优化代码结构

lib/infer_pack/attentions.py

@@ -0,0 +1,417 @@
+import copy
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from lib.infer_pack import commons
+from lib.infer_pack import modules
+from lib.infer_pack.modules import LayerNorm
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=10,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.attn_layers[i](x, x, attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        proximal_bias=False,
+        proximal_init=True,
+        **kwargs
+    ):
+        super().__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    proximal_bias=proximal_bias,
+                    proximal_init=proximal_init,
+                )
+            )
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+        x: decoder input
+        h: encoder output
+        """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype
+        )
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super().__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(self, x, c, attn_mask=None):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, self.attn = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(self, query, key, value, mask=None):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s, t_t = (*key.size(), query.size(2))
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length):
+        max_relative_position = 2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]]))
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [0, length - 1]])
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # padd along column
+        x = F.pad(
+            x, commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, length - 1]])
+        )
+        x_flat = x.view([batch, heads, length**2 + length * (length - 1)])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(x_flat, commons.convert_pad_shape([[0, 0], [0, 0], [length, 0]]))
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation=None,
+        causal=False,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+
+        if causal:
+            self.padding = self._causal_padding
+        else:
+            self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def forward(self, x, x_mask):
+        x = self.conv_1(self.padding(x * x_mask))
+        if self.activation == "gelu":
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+        x = self.conv_2(self.padding(x * x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = self.kernel_size - 1
+        pad_r = 0
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l = (self.kernel_size - 1) // 2
+        pad_r = self.kernel_size // 2
+        padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(x, commons.convert_pad_shape(padding))
+        return x

lib/infer_pack/commons.py

@@ -0,0 +1,166 @@
+import math
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+def convert_pad_shape(pad_shape):
+    l = pad_shape[::-1]
+    pad_shape = [item for sublist in l for item in sublist]
+    return pad_shape
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += (
+        0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
+    )
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, :, idx_str:idx_end]
+    return ret
+
+
+def slice_segments2(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, idx_str:idx_end]
+    return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = x_lengths - segment_size + 1
+    ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
+        num_timescales - 1
+    )
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
+    )
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+def convert_pad_shape(pad_shape):
+    l = pad_shape[::-1]
+    pad_shape = [item for sublist in l for item in sublist]
+    return pad_shape
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length, max_length=None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+    duration: [b, 1, t_x]
+    mask: [b, 1, t_y, t_x]
+    """
+    device = duration.device
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item() ** norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm ** (1.0 / norm_type)
+    return total_norm

lib/infer_pack/models_onnx.py

@@ -0,0 +1,819 @@
+import math, pdb, os
+from time import time as ttime
+import torch
+from torch import nn
+from torch.nn import functional as F
+from lib.infer_pack import modules
+from lib.infer_pack import attentions
+from lib.infer_pack import commons
+from lib.infer_pack.commons import init_weights, get_padding
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm, spectral_norm
+from lib.infer_pack.commons import init_weights
+import numpy as np
+from lib.infer_pack import commons
+
+
+class TextEncoder256(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class TextEncoder768(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(768, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+    def remove_weight_norm(self):
+        for i in range(self.n_flows):
+            self.flows[i * 2].remove_weight_norm()
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class SineGen(torch.nn.Module):
+    """Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(np.pi) or cos(0)
+    """
+
+    def __init__(
+        self,
+        samp_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        noise_std=0.003,
+        voiced_threshold=0,
+        flag_for_pulse=False,
+    ):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = torch.ones_like(f0)
+        uv = uv * (f0 > self.voiced_threshold)
+        return uv
+
+    def forward(self, f0, upp):
+        """sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            f0 = f0[:, None].transpose(1, 2)
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
+            # fundamental component
+            f0_buf[:, :, 0] = f0[:, :, 0]
+            for idx in np.arange(self.harmonic_num):
+                f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
+                    idx + 2
+                )  # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
+            rad_values = (f0_buf / self.sampling_rate) % 1  ###%1意味着n_har的乘积无法后处理优化
+            rand_ini = torch.rand(
+                f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
+            )
+            rand_ini[:, 0] = 0
+            rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+            tmp_over_one = torch.cumsum(rad_values, 1)  # % 1  #####%1意味着后面的cumsum无法再优化
+            tmp_over_one *= upp
+            tmp_over_one = F.interpolate(
+                tmp_over_one.transpose(2, 1),
+                scale_factor=upp,
+                mode="linear",
+                align_corners=True,
+            ).transpose(2, 1)
+            rad_values = F.interpolate(
+                rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(
+                2, 1
+            )  #######
+            tmp_over_one %= 1
+            tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+            sine_waves = torch.sin(
+                torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
+            )
+            sine_waves = sine_waves * self.sine_amp
+            uv = self._f02uv(f0)
+            uv = F.interpolate(
+                uv.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(2, 1)
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(
+        self,
+        sampling_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        add_noise_std=0.003,
+        voiced_threshod=0,
+        is_half=True,
+    ):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+        self.is_half = is_half
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(
+            sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
+        )
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+
+    def forward(self, x, upp=None):
+        sine_wavs, uv, _ = self.l_sin_gen(x, upp)
+        if self.is_half:
+            sine_wavs = sine_wavs.half()
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+        return sine_merge, None, None  # noise, uv
+
+
+class GeneratorNSF(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels,
+        sr,
+        is_half=False,
+    ):
+        super(GeneratorNSF, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=sr, harmonic_num=0, is_half=is_half
+        )
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            c_cur = upsample_initial_channel // (2 ** (i + 1))
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+            if i + 1 < len(upsample_rates):
+                stride_f0 = np.prod(upsample_rates[i + 1 :])
+                self.noise_convs.append(
+                    Conv1d(
+                        1,
+                        c_cur,
+                        kernel_size=stride_f0 * 2,
+                        stride=stride_f0,
+                        padding=stride_f0 // 2,
+                    )
+                )
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+        self.upp = np.prod(upsample_rates)
+
+    def forward(self, x, f0, g=None):
+        har_source, noi_source, uv = self.m_source(f0, self.upp)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            x_source = self.noise_convs[i](har_source)
+            x = x + x_source
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+sr2sr = {
+    "32k": 32000,
+    "40k": 40000,
+    "48k": 48000,
+}
+
+
+class SynthesizerTrnMsNSFsidM(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        version,
+        **kwargs
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        if version == "v1":
+            self.enc_p = TextEncoder256(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        else:
+            self.enc_p = TextEncoder768(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        self.speaker_map = None
+        print("gin_channels:", gin_channels, "self.spk_embed_dim:", self.spk_embed_dim)
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def construct_spkmixmap(self, n_speaker):
+        self.speaker_map = torch.zeros((n_speaker, 1, 1, self.gin_channels))
+        for i in range(n_speaker):
+            self.speaker_map[i] = self.emb_g(torch.LongTensor([[i]]))
+        self.speaker_map = self.speaker_map.unsqueeze(0)
+
+    def forward(self, phone, phone_lengths, pitch, nsff0, g, rnd, max_len=None):
+        if self.speaker_map is not None:  # [N, S]  *  [S, B, 1, H]
+            g = g.reshape((g.shape[0], g.shape[1], 1, 1, 1))  # [N, S, B, 1, 1]
+            g = g * self.speaker_map  # [N, S, B, 1, H]
+            g = torch.sum(g, dim=1)  # [N, 1, B, 1, H]
+            g = g.transpose(0, -1).transpose(0, -2).squeeze(0)  # [B, H, N]
+        else:
+            g = g.unsqueeze(0)
+            g = self.emb_g(g).transpose(1, 2)
+
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * rnd) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
+        return o
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11, 17]
+        # periods = [3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap

lib/infer_pack/modules.py

@@ -0,0 +1,522 @@
+import copy
+import math
+import numpy as np
+import scipy
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from torch.nn import Conv1d, ConvTranspose1d, AvgPool1d, Conv2d
+from torch.nn.utils import weight_norm, remove_weight_norm
+
+from lib.infer_pack import commons
+from lib.infer_pack.commons import init_weights, get_padding
+from lib.infer_pack.transforms import piecewise_rational_quadratic_transform
+
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super().__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+class ConvReluNorm(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        hidden_channels,
+        out_channels,
+        kernel_size,
+        n_layers,
+        p_dropout,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.hidden_channels = hidden_channels
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+        assert n_layers > 1, "Number of layers should be larger than 0."
+
+        self.conv_layers = nn.ModuleList()
+        self.norm_layers = nn.ModuleList()
+        self.conv_layers.append(
+            nn.Conv1d(
+                in_channels, hidden_channels, kernel_size, padding=kernel_size // 2
+            )
+        )
+        self.norm_layers.append(LayerNorm(hidden_channels))
+        self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(p_dropout))
+        for _ in range(n_layers - 1):
+            self.conv_layers.append(
+                nn.Conv1d(
+                    hidden_channels,
+                    hidden_channels,
+                    kernel_size,
+                    padding=kernel_size // 2,
+                )
+            )
+            self.norm_layers.append(LayerNorm(hidden_channels))
+        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask):
+        x_org = x
+        for i in range(self.n_layers):
+            x = self.conv_layers[i](x * x_mask)
+            x = self.norm_layers[i](x)
+            x = self.relu_drop(x)
+        x = x_org + self.proj(x)
+        return x * x_mask
+
+
+class DDSConv(nn.Module):
+    """
+    Dialted and Depth-Separable Convolution
+    """
+
+    def __init__(self, channels, kernel_size, n_layers, p_dropout=0.0):
+        super().__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = p_dropout
+
+        self.drop = nn.Dropout(p_dropout)
+        self.convs_sep = nn.ModuleList()
+        self.convs_1x1 = nn.ModuleList()
+        self.norms_1 = nn.ModuleList()
+        self.norms_2 = nn.ModuleList()
+        for i in range(n_layers):
+            dilation = kernel_size**i
+            padding = (kernel_size * dilation - dilation) // 2
+            self.convs_sep.append(
+                nn.Conv1d(
+                    channels,
+                    channels,
+                    kernel_size,
+                    groups=channels,
+                    dilation=dilation,
+                    padding=padding,
+                )
+            )
+            self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+            self.norms_1.append(LayerNorm(channels))
+            self.norms_2.append(LayerNorm(channels))
+
+    def forward(self, x, x_mask, g=None):
+        if g is not None:
+            x = x + g
+        for i in range(self.n_layers):
+            y = self.convs_sep[i](x * x_mask)
+            y = self.norms_1[i](y)
+            y = F.gelu(y)
+            y = self.convs_1x1[i](y)
+            y = self.norms_2[i](y)
+            y = F.gelu(y)
+            y = self.drop(y)
+            x = x + y
+        return x * x_mask
+
+
+class WN(torch.nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+        p_dropout=0,
+    ):
+        super(WN, self).__init__()
+        assert kernel_size % 2 == 1
+        self.hidden_channels = hidden_channels
+        self.kernel_size = (kernel_size,)
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+        self.p_dropout = p_dropout
+
+        self.in_layers = torch.nn.ModuleList()
+        self.res_skip_layers = torch.nn.ModuleList()
+        self.drop = nn.Dropout(p_dropout)
+
+        if gin_channels != 0:
+            cond_layer = torch.nn.Conv1d(
+                gin_channels, 2 * hidden_channels * n_layers, 1
+            )
+            self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name="weight")
+
+        for i in range(n_layers):
+            dilation = dilation_rate**i
+            padding = int((kernel_size * dilation - dilation) / 2)
+            in_layer = torch.nn.Conv1d(
+                hidden_channels,
+                2 * hidden_channels,
+                kernel_size,
+                dilation=dilation,
+                padding=padding,
+            )
+            in_layer = torch.nn.utils.weight_norm(in_layer, name="weight")
+            self.in_layers.append(in_layer)
+
+            # last one is not necessary
+            if i < n_layers - 1:
+                res_skip_channels = 2 * hidden_channels
+            else:
+                res_skip_channels = hidden_channels
+
+            res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+            res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name="weight")
+            self.res_skip_layers.append(res_skip_layer)
+
+    def forward(self, x, x_mask, g=None, **kwargs):
+        output = torch.zeros_like(x)
+        n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+        if g is not None:
+            g = self.cond_layer(g)
+
+        for i in range(self.n_layers):
+            x_in = self.in_layers[i](x)
+            if g is not None:
+                cond_offset = i * 2 * self.hidden_channels
+                g_l = g[:, cond_offset : cond_offset + 2 * self.hidden_channels, :]
+            else:
+                g_l = torch.zeros_like(x_in)
+
+            acts = commons.fused_add_tanh_sigmoid_multiply(x_in, g_l, n_channels_tensor)
+            acts = self.drop(acts)
+
+            res_skip_acts = self.res_skip_layers[i](acts)
+            if i < self.n_layers - 1:
+                res_acts = res_skip_acts[:, : self.hidden_channels, :]
+                x = (x + res_acts) * x_mask
+                output = output + res_skip_acts[:, self.hidden_channels :, :]
+            else:
+                output = output + res_skip_acts
+        return output * x_mask
+
+    def remove_weight_norm(self):
+        if self.gin_channels != 0:
+            torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            torch.nn.utils.remove_weight_norm(l)
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[2],
+                        padding=get_padding(kernel_size, dilation[2]),
+                    )
+                ),
+            ]
+        )
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+            ]
+        )
+        self.convs2.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+            ]
+        )
+        self.convs.apply(init_weights)
+
+    def forward(self, x, x_mask=None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, LRELU_SLOPE)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+
+class Log(nn.Module):
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+            logdet = torch.sum(-y, [1, 2])
+            return y, logdet
+        else:
+            x = torch.exp(x) * x_mask
+            return x
+
+
+class Flip(nn.Module):
+    def forward(self, x, *args, reverse=False, **kwargs):
+        x = torch.flip(x, [1])
+        if not reverse:
+            logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+            return x, logdet
+        else:
+            return x
+
+
+class ElementwiseAffine(nn.Module):
+    def __init__(self, channels):
+        super().__init__()
+        self.channels = channels
+        self.m = nn.Parameter(torch.zeros(channels, 1))
+        self.logs = nn.Parameter(torch.zeros(channels, 1))
+
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = self.m + torch.exp(self.logs) * x
+            y = y * x_mask
+            logdet = torch.sum(self.logs * x_mask, [1, 2])
+            return y, logdet
+        else:
+            x = (x - self.m) * torch.exp(-self.logs) * x_mask
+            return x
+
+
+class ResidualCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        p_dropout=0,
+        gin_channels=0,
+        mean_only=False,
+    ):
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            p_dropout=p_dropout,
+            gin_channels=gin_channels,
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class ConvFlow(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        filter_channels,
+        kernel_size,
+        n_layers,
+        num_bins=10,
+        tail_bound=5.0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.num_bins = num_bins
+        self.tail_bound = tail_bound
+        self.half_channels = in_channels // 2
+
+        self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+        self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.0)
+        self.proj = nn.Conv1d(
+            filter_channels, self.half_channels * (num_bins * 3 - 1), 1
+        )
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0)
+        h = self.convs(h, x_mask, g=g)
+        h = self.proj(h) * x_mask
+
+        b, c, t = x0.shape
+        h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2)  # [b, cx?, t] -> [b, c, t, ?]
+
+        unnormalized_widths = h[..., : self.num_bins] / math.sqrt(self.filter_channels)
+        unnormalized_heights = h[..., self.num_bins : 2 * self.num_bins] / math.sqrt(
+            self.filter_channels
+        )
+        unnormalized_derivatives = h[..., 2 * self.num_bins :]
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x

lib/infer_pack/modules/F0Predictor/DioF0Predictor.py

@@ -0,0 +1,90 @@
+from lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+import pyworld
+import numpy as np
+
+
+class DioF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def resize_f0(self, x, target_len):
+        source = np.array(x)
+        source[source < 0.001] = np.nan
+        target = np.interp(
+            np.arange(0, len(source) * target_len, len(source)) / target_len,
+            np.arange(0, len(source)),
+            source,
+        )
+        res = np.nan_to_num(target)
+        return res
+
+    def compute_f0(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.dio(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        for index, pitch in enumerate(f0):
+            f0[index] = round(pitch, 1)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))[0]
+
+    def compute_f0_uv(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.dio(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        for index, pitch in enumerate(f0):
+            f0[index] = round(pitch, 1)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))

lib/infer_pack/modules/F0Predictor/F0Predictor.py

@@ -0,0 +1,16 @@
+class F0Predictor(object):
+    def compute_f0(self, wav, p_len):
+        """
+        input: wav:[signal_length]
+               p_len:int
+        output: f0:[signal_length//hop_length]
+        """
+        pass
+
+    def compute_f0_uv(self, wav, p_len):
+        """
+        input: wav:[signal_length]
+               p_len:int
+        output: f0:[signal_length//hop_length],uv:[signal_length//hop_length]
+        """
+        pass

lib/infer_pack/modules/F0Predictor/HarvestF0Predictor.py

@@ -0,0 +1,86 @@
+from lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+import pyworld
+import numpy as np
+
+
+class HarvestF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def resize_f0(self, x, target_len):
+        source = np.array(x)
+        source[source < 0.001] = np.nan
+        target = np.interp(
+            np.arange(0, len(source) * target_len, len(source)) / target_len,
+            np.arange(0, len(source)),
+            source,
+        )
+        res = np.nan_to_num(target)
+        return res
+
+    def compute_f0(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.harvest(
+            wav.astype(np.double),
+            fs=self.hop_length,
+            f0_ceil=self.f0_max,
+            f0_floor=self.f0_min,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.fs)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))[0]
+
+    def compute_f0_uv(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.harvest(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))

lib/infer_pack/modules/F0Predictor/PMF0Predictor.py

@@ -0,0 +1,97 @@
+from lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+import parselmouth
+import numpy as np
+
+
+class PMF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def compute_f0(self, wav, p_len=None):
+        x = wav
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_length
+        else:
+            assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error"
+        time_step = self.hop_length / self.sampling_rate * 1000
+        f0 = (
+            parselmouth.Sound(x, self.sampling_rate)
+            .to_pitch_ac(
+                time_step=time_step / 1000,
+                voicing_threshold=0.6,
+                pitch_floor=self.f0_min,
+                pitch_ceiling=self.f0_max,
+            )
+            .selected_array["frequency"]
+        )
+
+        pad_size = (p_len - len(f0) + 1) // 2
+        if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+            f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant")
+        f0, uv = self.interpolate_f0(f0)
+        return f0
+
+    def compute_f0_uv(self, wav, p_len=None):
+        x = wav
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_length
+        else:
+            assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error"
+        time_step = self.hop_length / self.sampling_rate * 1000
+        f0 = (
+            parselmouth.Sound(x, self.sampling_rate)
+            .to_pitch_ac(
+                time_step=time_step / 1000,
+                voicing_threshold=0.6,
+                pitch_floor=self.f0_min,
+                pitch_ceiling=self.f0_max,
+            )
+            .selected_array["frequency"]
+        )
+
+        pad_size = (p_len - len(f0) + 1) // 2
+        if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+            f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant")
+        f0, uv = self.interpolate_f0(f0)
+        return f0, uv

lib/infer_pack/onnx_inference.py

@@ -0,0 +1,143 @@
+import onnxruntime
+import librosa
+import numpy as np
+import soundfile
+
+
+class ContentVec:
+    def __init__(self, vec_path="pretrained/vec-768-layer-12.onnx", device=None):
+        print("load model(s) from {}".format(vec_path))
+        if device == "cpu" or device is None:
+            providers = ["CPUExecutionProvider"]
+        elif device == "cuda":
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+        elif device == "dml":
+            providers = ["DmlExecutionProvider"]
+        else:
+            raise RuntimeError("Unsportted Device")
+        self.model = onnxruntime.InferenceSession(vec_path, providers=providers)
+
+    def __call__(self, wav):
+        return self.forward(wav)
+
+    def forward(self, wav):
+        feats = wav
+        if feats.ndim == 2:  # double channels
+            feats = feats.mean(-1)
+        assert feats.ndim == 1, feats.ndim
+        feats = np.expand_dims(np.expand_dims(feats, 0), 0)
+        onnx_input = {self.model.get_inputs()[0].name: feats}
+        logits = self.model.run(None, onnx_input)[0]
+        return logits.transpose(0, 2, 1)
+
+
+def get_f0_predictor(f0_predictor, hop_length, sampling_rate, **kargs):
+    if f0_predictor == "pm":
+        from lib.infer_pack.modules.F0Predictor.PMF0Predictor import PMF0Predictor
+
+        f0_predictor_object = PMF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    elif f0_predictor == "harvest":
+        from lib.infer_pack.modules.F0Predictor.HarvestF0Predictor import HarvestF0Predictor
+
+        f0_predictor_object = HarvestF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    elif f0_predictor == "dio":
+        from lib.infer_pack.modules.F0Predictor.DioF0Predictor import DioF0Predictor
+
+        f0_predictor_object = DioF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    else:
+        raise Exception("Unknown f0 predictor")
+    return f0_predictor_object
+
+
+class OnnxRVC:
+    def __init__(
+        self,
+        model_path,
+        sr=40000,
+        hop_size=512,
+        vec_path="vec-768-layer-12",
+        device="cpu",
+    ):
+        vec_path = f"pretrained/{vec_path}.onnx"
+        self.vec_model = ContentVec(vec_path, device)
+        if device == "cpu" or device is None:
+            providers = ["CPUExecutionProvider"]
+        elif device == "cuda":
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+        elif device == "dml":
+            providers = ["DmlExecutionProvider"]
+        else:
+            raise RuntimeError("Unsportted Device")
+        self.model = onnxruntime.InferenceSession(model_path, providers=providers)
+        self.sampling_rate = sr
+        self.hop_size = hop_size
+
+    def forward(self, hubert, hubert_length, pitch, pitchf, ds, rnd):
+        onnx_input = {
+            self.model.get_inputs()[0].name: hubert,
+            self.model.get_inputs()[1].name: hubert_length,
+            self.model.get_inputs()[2].name: pitch,
+            self.model.get_inputs()[3].name: pitchf,
+            self.model.get_inputs()[4].name: ds,
+            self.model.get_inputs()[5].name: rnd,
+        }
+        return (self.model.run(None, onnx_input)[0] * 32767).astype(np.int16)
+
+    def inference(
+        self,
+        raw_path,
+        sid,
+        f0_method="dio",
+        f0_up_key=0,
+        pad_time=0.5,
+        cr_threshold=0.02,
+    ):
+        f0_min = 50
+        f0_max = 1100
+        f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+        f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+        f0_predictor = get_f0_predictor(
+            f0_method,
+            hop_length=self.hop_size,
+            sampling_rate=self.sampling_rate,
+            threshold=cr_threshold,
+        )
+        wav, sr = librosa.load(raw_path, sr=self.sampling_rate)
+        org_length = len(wav)
+        if org_length / sr > 50.0:
+            raise RuntimeError("Reached Max Length")
+
+        wav16k = librosa.resample(wav, orig_sr=self.sampling_rate, target_sr=16000)
+        wav16k = wav16k
+
+        hubert = self.vec_model(wav16k)
+        hubert = np.repeat(hubert, 2, axis=2).transpose(0, 2, 1).astype(np.float32)
+        hubert_length = hubert.shape[1]
+
+        pitchf = f0_predictor.compute_f0(wav, hubert_length)
+        pitchf = pitchf * 2 ** (f0_up_key / 12)
+        pitch = pitchf.copy()
+        f0_mel = 1127 * np.log(1 + pitch / 700)
+        f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (
+            f0_mel_max - f0_mel_min
+        ) + 1
+        f0_mel[f0_mel <= 1] = 1
+        f0_mel[f0_mel > 255] = 255
+        pitch = np.rint(f0_mel).astype(np.int64)
+
+        pitchf = pitchf.reshape(1, len(pitchf)).astype(np.float32)
+        pitch = pitch.reshape(1, len(pitch))
+        ds = np.array([sid]).astype(np.int64)
+
+        rnd = np.random.randn(1, 192, hubert_length).astype(np.float32)
+        hubert_length = np.array([hubert_length]).astype(np.int64)
+
+        out_wav = self.forward(hubert, hubert_length, pitch, pitchf, ds, rnd).squeeze()
+        out_wav = np.pad(out_wav, (0, 2 * self.hop_size), "constant")
+        return out_wav[0:org_length]

lib/infer_pack/transforms.py

@@ -0,0 +1,209 @@
+import torch
+from torch.nn import functional as F
+
+import numpy as np
+
+
+DEFAULT_MIN_BIN_WIDTH = 1e-3
+DEFAULT_MIN_BIN_HEIGHT = 1e-3
+DEFAULT_MIN_DERIVATIVE = 1e-3
+
+
+def piecewise_rational_quadratic_transform(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails=None,
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if tails is None:
+        spline_fn = rational_quadratic_spline
+        spline_kwargs = {}
+    else:
+        spline_fn = unconstrained_rational_quadratic_spline
+        spline_kwargs = {"tails": tails, "tail_bound": tail_bound}
+
+    outputs, logabsdet = spline_fn(
+        inputs=inputs,
+        unnormalized_widths=unnormalized_widths,
+        unnormalized_heights=unnormalized_heights,
+        unnormalized_derivatives=unnormalized_derivatives,
+        inverse=inverse,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+        **spline_kwargs
+    )
+    return outputs, logabsdet
+
+
+def searchsorted(bin_locations, inputs, eps=1e-6):
+    bin_locations[..., -1] += eps
+    return torch.sum(inputs[..., None] >= bin_locations, dim=-1) - 1
+
+
+def unconstrained_rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails="linear",
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
+    outside_interval_mask = ~inside_interval_mask
+
+    outputs = torch.zeros_like(inputs)
+    logabsdet = torch.zeros_like(inputs)
+
+    if tails == "linear":
+        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
+        constant = np.log(np.exp(1 - min_derivative) - 1)
+        unnormalized_derivatives[..., 0] = constant
+        unnormalized_derivatives[..., -1] = constant
+
+        outputs[outside_interval_mask] = inputs[outside_interval_mask]
+        logabsdet[outside_interval_mask] = 0
+    else:
+        raise RuntimeError("{} tails are not implemented.".format(tails))
+
+    (
+        outputs[inside_interval_mask],
+        logabsdet[inside_interval_mask],
+    ) = rational_quadratic_spline(
+        inputs=inputs[inside_interval_mask],
+        unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
+        unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
+        unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
+        inverse=inverse,
+        left=-tail_bound,
+        right=tail_bound,
+        bottom=-tail_bound,
+        top=tail_bound,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+    )
+
+    return outputs, logabsdet
+
+
+def rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    left=0.0,
+    right=1.0,
+    bottom=0.0,
+    top=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if torch.min(inputs) < left or torch.max(inputs) > right:
+        raise ValueError("Input to a transform is not within its domain")
+
+    num_bins = unnormalized_widths.shape[-1]
+
+    if min_bin_width * num_bins > 1.0:
+        raise ValueError("Minimal bin width too large for the number of bins")
+    if min_bin_height * num_bins > 1.0:
+        raise ValueError("Minimal bin height too large for the number of bins")
+
+    widths = F.softmax(unnormalized_widths, dim=-1)
+    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
+    cumwidths = torch.cumsum(widths, dim=-1)
+    cumwidths = F.pad(cumwidths, pad=(1, 0), mode="constant", value=0.0)
+    cumwidths = (right - left) * cumwidths + left
+    cumwidths[..., 0] = left
+    cumwidths[..., -1] = right
+    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
+
+    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
+
+    heights = F.softmax(unnormalized_heights, dim=-1)
+    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
+    cumheights = torch.cumsum(heights, dim=-1)
+    cumheights = F.pad(cumheights, pad=(1, 0), mode="constant", value=0.0)
+    cumheights = (top - bottom) * cumheights + bottom
+    cumheights[..., 0] = bottom
+    cumheights[..., -1] = top
+    heights = cumheights[..., 1:] - cumheights[..., :-1]
+
+    if inverse:
+        bin_idx = searchsorted(cumheights, inputs)[..., None]
+    else:
+        bin_idx = searchsorted(cumwidths, inputs)[..., None]
+
+    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
+    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
+
+    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
+    delta = heights / widths
+    input_delta = delta.gather(-1, bin_idx)[..., 0]
+
+    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
+    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
+
+    input_heights = heights.gather(-1, bin_idx)[..., 0]
+
+    if inverse:
+        a = (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        ) + input_heights * (input_delta - input_derivatives)
+        b = input_heights * input_derivatives - (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        )
+        c = -input_delta * (inputs - input_cumheights)
+
+        discriminant = b.pow(2) - 4 * a * c
+        assert (discriminant >= 0).all()
+
+        root = (2 * c) / (-b - torch.sqrt(discriminant))
+        outputs = root * input_bin_widths + input_cumwidths
+
+        theta_one_minus_theta = root * (1 - root)
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * root.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - root).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, -logabsdet
+    else:
+        theta = (inputs - input_cumwidths) / input_bin_widths
+        theta_one_minus_theta = theta * (1 - theta)
+
+        numerator = input_heights * (
+            input_delta * theta.pow(2) + input_derivatives * theta_one_minus_theta
+        )
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        outputs = input_cumheights + numerator / denominator
+
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * theta.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - theta).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, logabsdet

lib/uvr5_pack/lib_v5/dataset.py

@@ -0,0 +1,183 @@
+import os
+import random
+
+import numpy as np
+import torch
+import torch.utils.data
+from tqdm import tqdm
+
+from . import spec_utils
+
+
+class VocalRemoverValidationSet(torch.utils.data.Dataset):
+    def __init__(self, patch_list):
+        self.patch_list = patch_list
+
+    def __len__(self):
+        return len(self.patch_list)
+
+    def __getitem__(self, idx):
+        path = self.patch_list[idx]
+        data = np.load(path)
+
+        X, y = data["X"], data["y"]
+
+        X_mag = np.abs(X)
+        y_mag = np.abs(y)
+
+        return X_mag, y_mag
+
+
+def make_pair(mix_dir, inst_dir):
+    input_exts = [".wav", ".m4a", ".mp3", ".mp4", ".flac"]
+
+    X_list = sorted(
+        [
+            os.path.join(mix_dir, fname)
+            for fname in os.listdir(mix_dir)
+            if os.path.splitext(fname)[1] in input_exts
+        ]
+    )
+    y_list = sorted(
+        [
+            os.path.join(inst_dir, fname)
+            for fname in os.listdir(inst_dir)
+            if os.path.splitext(fname)[1] in input_exts
+        ]
+    )
+
+    filelist = list(zip(X_list, y_list))
+
+    return filelist
+
+
+def train_val_split(dataset_dir, split_mode, val_rate, val_filelist):
+    if split_mode == "random":
+        filelist = make_pair(
+            os.path.join(dataset_dir, "mixtures"),
+            os.path.join(dataset_dir, "instruments"),
+        )
+
+        random.shuffle(filelist)
+
+        if len(val_filelist) == 0:
+            val_size = int(len(filelist) * val_rate)
+            train_filelist = filelist[:-val_size]
+            val_filelist = filelist[-val_size:]
+        else:
+            train_filelist = [
+                pair for pair in filelist if list(pair) not in val_filelist
+            ]
+    elif split_mode == "subdirs":
+        if len(val_filelist) != 0:
+            raise ValueError(
+                "The `val_filelist` option is not available in `subdirs` mode"
+            )
+
+        train_filelist = make_pair(
+            os.path.join(dataset_dir, "training/mixtures"),
+            os.path.join(dataset_dir, "training/instruments"),
+        )
+
+        val_filelist = make_pair(
+            os.path.join(dataset_dir, "validation/mixtures"),
+            os.path.join(dataset_dir, "validation/instruments"),
+        )
+
+    return train_filelist, val_filelist
+
+
+def augment(X, y, reduction_rate, reduction_mask, mixup_rate, mixup_alpha):
+    perm = np.random.permutation(len(X))
+    for i, idx in enumerate(tqdm(perm)):
+        if np.random.uniform() < reduction_rate:
+            y[idx] = spec_utils.reduce_vocal_aggressively(
+                X[idx], y[idx], reduction_mask
+            )
+
+        if np.random.uniform() < 0.5:
+            # swap channel
+            X[idx] = X[idx, ::-1]
+            y[idx] = y[idx, ::-1]
+        if np.random.uniform() < 0.02:
+            # mono
+            X[idx] = X[idx].mean(axis=0, keepdims=True)
+            y[idx] = y[idx].mean(axis=0, keepdims=True)
+        if np.random.uniform() < 0.02:
+            # inst
+            X[idx] = y[idx]
+
+        if np.random.uniform() < mixup_rate and i < len(perm) - 1:
+            lam = np.random.beta(mixup_alpha, mixup_alpha)
+            X[idx] = lam * X[idx] + (1 - lam) * X[perm[i + 1]]
+            y[idx] = lam * y[idx] + (1 - lam) * y[perm[i + 1]]
+
+    return X, y
+
+
+def make_padding(width, cropsize, offset):
+    left = offset
+    roi_size = cropsize - left * 2
+    if roi_size == 0:
+        roi_size = cropsize
+    right = roi_size - (width % roi_size) + left
+
+    return left, right, roi_size
+
+
+def make_training_set(filelist, cropsize, patches, sr, hop_length, n_fft, offset):
+    len_dataset = patches * len(filelist)
+
+    X_dataset = np.zeros((len_dataset, 2, n_fft // 2 + 1, cropsize), dtype=np.complex64)
+    y_dataset = np.zeros((len_dataset, 2, n_fft // 2 + 1, cropsize), dtype=np.complex64)
+
+    for i, (X_path, y_path) in enumerate(tqdm(filelist)):
+        X, y = spec_utils.cache_or_load(X_path, y_path, sr, hop_length, n_fft)
+        coef = np.max([np.abs(X).max(), np.abs(y).max()])
+        X, y = X / coef, y / coef
+
+        l, r, roi_size = make_padding(X.shape[2], cropsize, offset)
+        X_pad = np.pad(X, ((0, 0), (0, 0), (l, r)), mode="constant")
+        y_pad = np.pad(y, ((0, 0), (0, 0), (l, r)), mode="constant")
+
+        starts = np.random.randint(0, X_pad.shape[2] - cropsize, patches)
+        ends = starts + cropsize
+        for j in range(patches):
+            idx = i * patches + j
+            X_dataset[idx] = X_pad[:, :, starts[j] : ends[j]]
+            y_dataset[idx] = y_pad[:, :, starts[j] : ends[j]]
+
+    return X_dataset, y_dataset
+
+
+def make_validation_set(filelist, cropsize, sr, hop_length, n_fft, offset):
+    patch_list = []
+    patch_dir = "cs{}_sr{}_hl{}_nf{}_of{}".format(
+        cropsize, sr, hop_length, n_fft, offset
+    )
+    os.makedirs(patch_dir, exist_ok=True)
+
+    for i, (X_path, y_path) in enumerate(tqdm(filelist)):
+        basename = os.path.splitext(os.path.basename(X_path))[0]
+
+        X, y = spec_utils.cache_or_load(X_path, y_path, sr, hop_length, n_fft)
+        coef = np.max([np.abs(X).max(), np.abs(y).max()])
+        X, y = X / coef, y / coef
+
+        l, r, roi_size = make_padding(X.shape[2], cropsize, offset)
+        X_pad = np.pad(X, ((0, 0), (0, 0), (l, r)), mode="constant")
+        y_pad = np.pad(y, ((0, 0), (0, 0), (l, r)), mode="constant")
+
+        len_dataset = int(np.ceil(X.shape[2] / roi_size))
+        for j in range(len_dataset):
+            outpath = os.path.join(patch_dir, "{}_p{}.npz".format(basename, j))
+            start = j * roi_size
+            if not os.path.exists(outpath):
+                np.savez(
+                    outpath,
+                    X=X_pad[:, :, start : start + cropsize],
+                    y=y_pad[:, :, start : start + cropsize],
+                )
+            patch_list.append(outpath)
+
+    return VocalRemoverValidationSet(patch_list)

lib/uvr5_pack/lib_v5/layers.py

@@ -0,0 +1,118 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class SeperableConv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(SeperableConv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nin,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                groups=nin,
+                bias=False,
+            ),
+            nn.Conv2d(nin, nout, kernel_size=1, bias=False),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+
+    def __call__(self, x):
+        skip = self.conv1(x)
+        h = self.conv2(skip)
+
+        return h, skip
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+        h = self.conv(x)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv3 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = nn.Sequential(
+            Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
+        )
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
+        bottle = self.bottleneck(out)
+        return bottle

lib/uvr5_pack/lib_v5/layers_123812KB .py

@@ -0,0 +1,118 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class SeperableConv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(SeperableConv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nin,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                groups=nin,
+                bias=False,
+            ),
+            nn.Conv2d(nin, nout, kernel_size=1, bias=False),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+
+    def __call__(self, x):
+        skip = self.conv1(x)
+        h = self.conv2(skip)
+
+        return h, skip
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+        h = self.conv(x)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv3 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = nn.Sequential(
+            Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
+        )
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
+        bottle = self.bottleneck(out)
+        return bottle

lib/uvr5_pack/lib_v5/layers_123821KB.py

@@ -0,0 +1,118 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class SeperableConv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(SeperableConv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nin,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                groups=nin,
+                bias=False,
+            ),
+            nn.Conv2d(nin, nout, kernel_size=1, bias=False),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+
+    def __call__(self, x):
+        skip = self.conv1(x)
+        h = self.conv2(skip)
+
+        return h, skip
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+        h = self.conv(x)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv3 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = nn.Sequential(
+            Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
+        )
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
+        bottle = self.bottleneck(out)
+        return bottle

lib/uvr5_pack/lib_v5/layers_33966KB.py

@@ -0,0 +1,126 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class SeperableConv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(SeperableConv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nin,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                groups=nin,
+                bias=False,
+            ),
+            nn.Conv2d(nin, nout, kernel_size=1, bias=False),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+
+    def __call__(self, x):
+        skip = self.conv1(x)
+        h = self.conv2(skip)
+
+        return h, skip
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+        h = self.conv(x)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv3 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.conv6 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.conv7 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = nn.Sequential(
+            Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
+        )
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        feat6 = self.conv6(x)
+        feat7 = self.conv7(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
+        bottle = self.bottleneck(out)
+        return bottle

lib/uvr5_pack/lib_v5/layers_537227KB.py

@@ -0,0 +1,126 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class SeperableConv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(SeperableConv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nin,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                groups=nin,
+                bias=False,
+            ),
+            nn.Conv2d(nin, nout, kernel_size=1, bias=False),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+
+    def __call__(self, x):
+        skip = self.conv1(x)
+        h = self.conv2(skip)
+
+        return h, skip
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+        h = self.conv(x)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv3 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.conv6 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.conv7 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = nn.Sequential(
+            Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
+        )
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        feat6 = self.conv6(x)
+        feat7 = self.conv7(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
+        bottle = self.bottleneck(out)
+        return bottle

lib/uvr5_pack/lib_v5/layers_537238KB.py

@@ -0,0 +1,126 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class SeperableConv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(SeperableConv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nin,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                groups=nin,
+                bias=False,
+            ),
+            nn.Conv2d(nin, nout, kernel_size=1, bias=False),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
+
+    def __call__(self, x):
+        skip = self.conv1(x)
+        h = self.conv2(skip)
+
+        return h, skip
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+        h = self.conv(x)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
+        self.conv3 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.conv6 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.conv7 = SeperableConv2DBNActiv(
+            nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = nn.Sequential(
+            Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
+        )
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        feat6 = self.conv6(x)
+        feat7 = self.conv7(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
+        bottle = self.bottleneck(out)
+        return bottle

lib/uvr5_pack/lib_v5/layers_new.py

@@ -0,0 +1,125 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import spec_utils
+
+
+class Conv2DBNActiv(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
+        super(Conv2DBNActiv, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                nin,
+                nout,
+                kernel_size=ksize,
+                stride=stride,
+                padding=pad,
+                dilation=dilation,
+                bias=False,
+            ),
+            nn.BatchNorm2d(nout),
+            activ(),
+        )
+
+    def __call__(self, x):
+        return self.conv(x)
+
+
+class Encoder(nn.Module):
+    def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
+        super(Encoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, stride, pad, activ=activ)
+        self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ)
+
+    def __call__(self, x):
+        h = self.conv1(x)
+        h = self.conv2(h)
+
+        return h
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
+    ):
+        super(Decoder, self).__init__()
+        self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
+        # self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def __call__(self, x, skip=None):
+        x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
+
+        if skip is not None:
+            skip = spec_utils.crop_center(skip, x)
+            x = torch.cat([x, skip], dim=1)
+
+        h = self.conv1(x)
+        # h = self.conv2(h)
+
+        if self.dropout is not None:
+            h = self.dropout(h)
+
+        return h
+
+
+class ASPPModule(nn.Module):
+    def __init__(self, nin, nout, dilations=(4, 8, 12), activ=nn.ReLU, dropout=False):
+        super(ASPPModule, self).__init__()
+        self.conv1 = nn.Sequential(
+            nn.AdaptiveAvgPool2d((1, None)),
+            Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ),
+        )
+        self.conv2 = Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ)
+        self.conv3 = Conv2DBNActiv(
+            nin, nout, 3, 1, dilations[0], dilations[0], activ=activ
+        )
+        self.conv4 = Conv2DBNActiv(
+            nin, nout, 3, 1, dilations[1], dilations[1], activ=activ
+        )
+        self.conv5 = Conv2DBNActiv(
+            nin, nout, 3, 1, dilations[2], dilations[2], activ=activ
+        )
+        self.bottleneck = Conv2DBNActiv(nout * 5, nout, 1, 1, 0, activ=activ)
+        self.dropout = nn.Dropout2d(0.1) if dropout else None
+
+    def forward(self, x):
+        _, _, h, w = x.size()
+        feat1 = F.interpolate(
+            self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
+        )
+        feat2 = self.conv2(x)
+        feat3 = self.conv3(x)
+        feat4 = self.conv4(x)
+        feat5 = self.conv5(x)
+        out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
+        out = self.bottleneck(out)
+
+        if self.dropout is not None:
+            out = self.dropout(out)
+
+        return out
+
+
+class LSTMModule(nn.Module):
+    def __init__(self, nin_conv, nin_lstm, nout_lstm):
+        super(LSTMModule, self).__init__()
+        self.conv = Conv2DBNActiv(nin_conv, 1, 1, 1, 0)
+        self.lstm = nn.LSTM(
+            input_size=nin_lstm, hidden_size=nout_lstm // 2, bidirectional=True
+        )
+        self.dense = nn.Sequential(
+            nn.Linear(nout_lstm, nin_lstm), nn.BatchNorm1d(nin_lstm), nn.ReLU()
+        )
+
+    def forward(self, x):
+        N, _, nbins, nframes = x.size()
+        h = self.conv(x)[:, 0]  # N, nbins, nframes
+        h = h.permute(2, 0, 1)  # nframes, N, nbins
+        h, _ = self.lstm(h)
+        h = self.dense(h.reshape(-1, h.size()[-1]))  # nframes * N, nbins
+        h = h.reshape(nframes, N, 1, nbins)
+        h = h.permute(1, 2, 3, 0)
+
+        return h

lib/uvr5_pack/lib_v5/model_param_init.py

@@ -0,0 +1,69 @@
+import json
+import os
+import pathlib
+
+default_param = {}
+default_param["bins"] = 768
+default_param["unstable_bins"] = 9  # training only
+default_param["reduction_bins"] = 762  # training only
+default_param["sr"] = 44100
+default_param["pre_filter_start"] = 757
+default_param["pre_filter_stop"] = 768
+default_param["band"] = {}
+
+
+default_param["band"][1] = {
+    "sr": 11025,
+    "hl": 128,
+    "n_fft": 960,
+    "crop_start": 0,
+    "crop_stop": 245,
+    "lpf_start": 61,  # inference only
+    "res_type": "polyphase",
+}
+
+default_param["band"][2] = {
+    "sr": 44100,
+    "hl": 512,
+    "n_fft": 1536,
+    "crop_start": 24,
+    "crop_stop": 547,
+    "hpf_start": 81,  # inference only
+    "res_type": "sinc_best",
+}
+
+
+def int_keys(d):
+    r = {}
+    for k, v in d:
+        if k.isdigit():
+            k = int(k)
+        r[k] = v
+    return r
+
+
+class ModelParameters(object):
+    def __init__(self, config_path=""):
+        if ".pth" == pathlib.Path(config_path).suffix:
+            import zipfile
+
+            with zipfile.ZipFile(config_path, "r") as zip:
+                self.param = json.loads(
+                    zip.read("param.json"), object_pairs_hook=int_keys
+                )
+        elif ".json" == pathlib.Path(config_path).suffix:
+            with open(config_path, "r") as f:
+                self.param = json.loads(f.read(), object_pairs_hook=int_keys)
+        else:
+            self.param = default_param
+
+        for k in [
+            "mid_side",
+            "mid_side_b",
+            "mid_side_b2",
+            "stereo_w",
+            "stereo_n",
+            "reverse",
+        ]:
+            if not k in self.param:
+                self.param[k] = False

lib/uvr5_pack/lib_v5/modelparams/1band_sr16000_hl512.json

@@ -0,0 +1,19 @@
+{
+	"bins": 1024,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 16000,
+			"hl": 512,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 1024,
+			"hpf_start": -1,
+			"res_type": "sinc_best"
+		}
+	},
+	"sr": 16000,
+	"pre_filter_start": 1023,
+	"pre_filter_stop": 1024
+}

lib/uvr5_pack/lib_v5/modelparams/1band_sr32000_hl512.json

@@ -0,0 +1,19 @@
+{
+	"bins": 1024,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 32000,
+			"hl": 512,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 1024,
+			"hpf_start": -1,
+			"res_type": "kaiser_fast"
+		}
+	},
+	"sr": 32000,
+	"pre_filter_start": 1000,
+	"pre_filter_stop": 1021
+}

lib/uvr5_pack/lib_v5/modelparams/1band_sr33075_hl384.json

@@ -0,0 +1,19 @@
+{
+	"bins": 1024,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 33075,
+			"hl": 384,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 1024,
+			"hpf_start": -1,
+			"res_type": "sinc_best"
+		}
+	},
+	"sr": 33075,
+	"pre_filter_start": 1000,
+	"pre_filter_stop": 1021
+}

lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl1024.json

@@ -0,0 +1,19 @@
+{
+	"bins": 1024,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 44100,
+			"hl": 1024,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 1024,
+			"hpf_start": -1,
+			"res_type": "sinc_best"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 1023,
+	"pre_filter_stop": 1024
+}

lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl256.json

@@ -0,0 +1,19 @@
+{
+	"bins": 256,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 44100,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 0,
+			"crop_stop": 256,
+			"hpf_start": -1,
+			"res_type": "sinc_best"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 256,
+	"pre_filter_stop": 256
+}

lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl512.json

@@ -0,0 +1,19 @@
+{
+	"bins": 1024,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 1024,
+			"hpf_start": -1,
+			"res_type": "sinc_best"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 1023,
+	"pre_filter_stop": 1024
+}

lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl512_cut.json

@@ -0,0 +1,19 @@
+{
+	"bins": 1024,
+	"unstable_bins": 0,
+	"reduction_bins": 0,
+	"band": {
+		"1": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 700,
+			"hpf_start": -1,
+			"res_type": "sinc_best"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 1023,
+	"pre_filter_stop": 700
+}

lib/uvr5_pack/lib_v5/modelparams/2band_32000.json

@@ -0,0 +1,30 @@
+{
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 705,
+	"band": {
+		"1": {
+			"sr": 6000,
+			"hl": 66,
+			"n_fft": 512,
+			"crop_start": 0,
+			"crop_stop": 240,
+			"lpf_start": 60,
+			"lpf_stop": 118,
+			"res_type": "sinc_fastest"
+		},
+		"2": {
+			"sr": 32000,
+			"hl": 352,
+			"n_fft": 1024,
+			"crop_start": 22,
+			"crop_stop": 505,
+			"hpf_start": 44,
+			"hpf_stop": 23,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 32000,
+	"pre_filter_start": 710,
+	"pre_filter_stop": 731
+}

lib/uvr5_pack/lib_v5/modelparams/2band_44100_lofi.json

@@ -0,0 +1,30 @@
+{
+	"bins": 512,
+	"unstable_bins": 7,
+	"reduction_bins": 510,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 160,
+			"n_fft": 768,
+			"crop_start": 0,
+			"crop_stop": 192,
+			"lpf_start": 41,
+			"lpf_stop": 139,
+			"res_type": "sinc_fastest"
+		},
+		"2": {
+			"sr": 44100,
+			"hl": 640,
+			"n_fft": 1024,
+			"crop_start": 10,
+			"crop_stop": 320,
+			"hpf_start": 47,
+			"hpf_stop": 15,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 510,
+	"pre_filter_stop": 512
+}

lib/uvr5_pack/lib_v5/modelparams/2band_48000.json

@@ -0,0 +1,30 @@
+{
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 705,
+	"band": {
+		"1": {
+			"sr": 6000,
+			"hl": 66,
+			"n_fft": 512,
+			"crop_start": 0,
+			"crop_stop": 240,
+			"lpf_start": 60,
+			"lpf_stop": 240,
+			"res_type": "sinc_fastest"
+		},
+		"2": {
+			"sr": 48000,
+			"hl": 528,
+			"n_fft": 1536,
+			"crop_start": 22,
+			"crop_stop": 505,
+			"hpf_start": 82,
+			"hpf_stop": 22,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 48000,
+	"pre_filter_start": 710,
+	"pre_filter_stop": 731
+}

lib/uvr5_pack/lib_v5/modelparams/3band_44100.json

@@ -0,0 +1,42 @@
+{
+	"bins": 768,
+	"unstable_bins": 5,
+	"reduction_bins": 733,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 768,
+			"crop_start": 0,
+			"crop_stop": 278,
+			"lpf_start": 28,
+			"lpf_stop": 140,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 768,
+			"crop_start": 14,
+			"crop_stop": 322,
+			"hpf_start": 70,
+			"hpf_stop": 14,
+			"lpf_start": 283,
+			"lpf_stop": 314,
+			"res_type": "polyphase"
+		},	
+		"3": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 131,
+			"crop_stop": 313,
+			"hpf_start": 154,
+			"hpf_stop": 141,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 757,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/3band_44100_mid.json

@@ -0,0 +1,43 @@
+{
+	"mid_side": true,
+	"bins": 768,
+	"unstable_bins": 5,
+	"reduction_bins": 733,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 768,
+			"crop_start": 0,
+			"crop_stop": 278,
+			"lpf_start": 28,
+			"lpf_stop": 140,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 768,
+			"crop_start": 14,
+			"crop_stop": 322,
+			"hpf_start": 70,
+			"hpf_stop": 14,
+			"lpf_start": 283,
+			"lpf_stop": 314,
+			"res_type": "polyphase"
+		},	
+		"3": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 131,
+			"crop_stop": 313,
+			"hpf_start": 154,
+			"hpf_stop": 141,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 757,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/3band_44100_msb2.json

@@ -0,0 +1,43 @@
+{
+	"mid_side_b2": true,
+	"bins": 640,
+	"unstable_bins": 7,
+	"reduction_bins": 565,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 108,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 187,
+			"lpf_start": 92,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 22050,
+			"hl": 216,
+			"n_fft": 768,
+			"crop_start": 0,
+			"crop_stop": 212,
+			"hpf_start": 68,
+			"hpf_stop": 34,
+			"lpf_start": 174,
+			"lpf_stop": 209,
+			"res_type": "polyphase"
+		},	
+		"3": {
+			"sr": 44100,
+			"hl": 432,
+			"n_fft": 640,
+			"crop_start": 66,
+			"crop_stop": 307,
+			"hpf_start": 86,
+			"hpf_stop": 72,
+			"res_type": "kaiser_fast"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 639,
+	"pre_filter_stop": 640
+}

lib/uvr5_pack/lib_v5/modelparams/4band_44100.json

@@ -0,0 +1,54 @@
+{
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 668,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 186,
+			"lpf_start": 37,
+			"lpf_stop": 73,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 512,
+			"crop_start": 4,
+			"crop_stop": 185,			
+			"hpf_start": 36,
+			"hpf_stop": 18,
+			"lpf_start": 93,
+			"lpf_stop": 185,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 46,
+			"crop_stop": 186,
+			"hpf_start": 93,
+			"hpf_stop": 46,
+			"lpf_start": 164,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 121,
+			"crop_stop": 382,
+			"hpf_start": 138,
+			"hpf_stop": 123,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 740,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/4band_44100_mid.json

@@ -0,0 +1,55 @@
+{
+	"bins": 768,
+	"unstable_bins": 7,
+	"mid_side": true,
+	"reduction_bins": 668,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 186,
+			"lpf_start": 37,
+			"lpf_stop": 73,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 512,
+			"crop_start": 4,
+			"crop_stop": 185,			
+			"hpf_start": 36,
+			"hpf_stop": 18,
+			"lpf_start": 93,
+			"lpf_stop": 185,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 46,
+			"crop_stop": 186,
+			"hpf_start": 93,
+			"hpf_stop": 46,
+			"lpf_start": 164,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 121,
+			"crop_stop": 382,
+			"hpf_start": 138,
+			"hpf_stop": 123,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 740,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/4band_44100_msb.json

@@ -0,0 +1,55 @@
+{
+	"mid_side_b": true,
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 668,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 186,
+			"lpf_start": 37,
+			"lpf_stop": 73,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 512,
+			"crop_start": 4,
+			"crop_stop": 185,			
+			"hpf_start": 36,
+			"hpf_stop": 18,
+			"lpf_start": 93,
+			"lpf_stop": 185,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 46,
+			"crop_stop": 186,
+			"hpf_start": 93,
+			"hpf_stop": 46,
+			"lpf_start": 164,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 121,
+			"crop_stop": 382,
+			"hpf_start": 138,
+			"hpf_stop": 123,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 740,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/4band_44100_msb2.json

@@ -0,0 +1,55 @@
+{
+	"mid_side_b": true,
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 668,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 186,
+			"lpf_start": 37,
+			"lpf_stop": 73,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 512,
+			"crop_start": 4,
+			"crop_stop": 185,			
+			"hpf_start": 36,
+			"hpf_stop": 18,
+			"lpf_start": 93,
+			"lpf_stop": 185,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 46,
+			"crop_stop": 186,
+			"hpf_start": 93,
+			"hpf_stop": 46,
+			"lpf_start": 164,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 121,
+			"crop_stop": 382,
+			"hpf_start": 138,
+			"hpf_stop": 123,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 740,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/4band_44100_reverse.json

@@ -0,0 +1,55 @@
+{
+	"reverse": true,
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 668,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 186,
+			"lpf_start": 37,
+			"lpf_stop": 73,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 512,
+			"crop_start": 4,
+			"crop_stop": 185,			
+			"hpf_start": 36,
+			"hpf_stop": 18,
+			"lpf_start": 93,
+			"lpf_stop": 185,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 46,
+			"crop_stop": 186,
+			"hpf_start": 93,
+			"hpf_stop": 46,
+			"lpf_start": 164,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 121,
+			"crop_stop": 382,
+			"hpf_start": 138,
+			"hpf_stop": 123,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 740,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/4band_44100_sw.json

@@ -0,0 +1,55 @@
+{
+	"stereo_w": true,
+	"bins": 768,
+	"unstable_bins": 7,
+	"reduction_bins": 668,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 1024,
+			"crop_start": 0,
+			"crop_stop": 186,
+			"lpf_start": 37,
+			"lpf_stop": 73,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 11025,
+			"hl": 128,
+			"n_fft": 512,
+			"crop_start": 4,
+			"crop_stop": 185,			
+			"hpf_start": 36,
+			"hpf_stop": 18,
+			"lpf_start": 93,
+			"lpf_stop": 185,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 22050,
+			"hl": 256,
+			"n_fft": 512,
+			"crop_start": 46,
+			"crop_stop": 186,
+			"hpf_start": 93,
+			"hpf_stop": 46,
+			"lpf_start": 164,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 512,
+			"n_fft": 768,
+			"crop_start": 121,
+			"crop_stop": 382,
+			"hpf_start": 138,
+			"hpf_stop": 123,
+			"res_type": "sinc_medium"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 740,
+	"pre_filter_stop": 768
+}

lib/uvr5_pack/lib_v5/modelparams/4band_v2.json

@@ -0,0 +1,54 @@
+{
+	"bins": 672,
+	"unstable_bins": 8,
+	"reduction_bins": 637,
+	"band": {
+		"1": {
+			"sr": 7350,
+			"hl": 80,
+			"n_fft": 640,
+			"crop_start": 0,
+			"crop_stop": 85,
+			"lpf_start": 25,
+			"lpf_stop": 53,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 7350,
+			"hl": 80,
+			"n_fft": 320,
+			"crop_start": 4,
+			"crop_stop": 87,
+			"hpf_start": 25,
+			"hpf_stop": 12,
+			"lpf_start": 31,
+			"lpf_stop": 62,
+			"res_type": "polyphase"
+		},		
+		"3": {
+			"sr": 14700,
+			"hl": 160,
+			"n_fft": 512,
+			"crop_start": 17,
+			"crop_stop": 216,
+			"hpf_start": 48,
+			"hpf_stop": 24,
+			"lpf_start": 139,
+			"lpf_stop": 210,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 480,
+			"n_fft": 960,
+			"crop_start": 78,
+			"crop_stop": 383,
+			"hpf_start": 130,
+			"hpf_stop": 86,
+			"res_type": "kaiser_fast"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 668,
+	"pre_filter_stop": 672
+}

lib/uvr5_pack/lib_v5/modelparams/4band_v2_sn.json

@@ -0,0 +1,55 @@
+{
+	"bins": 672,
+	"unstable_bins": 8,
+	"reduction_bins": 637,
+	"band": {
+		"1": {
+			"sr": 7350,
+			"hl": 80,
+			"n_fft": 640,
+			"crop_start": 0,
+			"crop_stop": 85,
+			"lpf_start": 25,
+			"lpf_stop": 53,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 7350,
+			"hl": 80,
+			"n_fft": 320,
+			"crop_start": 4,
+			"crop_stop": 87,
+			"hpf_start": 25,
+			"hpf_stop": 12,
+			"lpf_start": 31,
+			"lpf_stop": 62,
+			"res_type": "polyphase"
+		},		
+		"3": {
+			"sr": 14700,
+			"hl": 160,
+			"n_fft": 512,
+			"crop_start": 17,
+			"crop_stop": 216,
+			"hpf_start": 48,
+			"hpf_stop": 24,
+			"lpf_start": 139,
+			"lpf_stop": 210,
+			"res_type": "polyphase"
+		},	
+		"4": {
+			"sr": 44100,
+			"hl": 480,
+			"n_fft": 960,
+			"crop_start": 78,
+			"crop_stop": 383,
+			"hpf_start": 130,
+			"hpf_stop": 86,
+			"convert_channels": "stereo_n",
+			"res_type": "kaiser_fast"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 668,
+	"pre_filter_stop": 672
+}

lib/uvr5_pack/lib_v5/modelparams/4band_v3.json

@@ -0,0 +1,54 @@
+{
+	"bins": 672,
+	"unstable_bins": 8,
+	"reduction_bins": 530,
+	"band": {
+		"1": {
+			"sr": 7350,
+			"hl": 80,
+			"n_fft": 640,
+			"crop_start": 0,
+			"crop_stop": 85,
+			"lpf_start": 25,
+			"lpf_stop": 53,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 7350,
+			"hl": 80,
+			"n_fft": 320,
+			"crop_start": 4,
+			"crop_stop": 87,
+			"hpf_start": 25,
+			"hpf_stop": 12,
+			"lpf_start": 31,
+			"lpf_stop": 62,
+			"res_type": "polyphase"
+		},
+		"3": {
+			"sr": 14700,
+			"hl": 160,
+			"n_fft": 512,
+			"crop_start": 17,
+			"crop_stop": 216,
+			"hpf_start": 48,
+			"hpf_stop": 24,
+			"lpf_start": 139,
+			"lpf_stop": 210,
+			"res_type": "polyphase"
+		},
+		"4": {
+			"sr": 44100,
+			"hl": 480,
+			"n_fft": 960,
+			"crop_start": 78,
+			"crop_stop": 383,
+			"hpf_start": 130,
+			"hpf_stop": 86,
+			"res_type": "kaiser_fast"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 668,
+	"pre_filter_stop": 672
+}

lib/uvr5_pack/lib_v5/modelparams/ensemble.json

@@ -0,0 +1,43 @@
+{
+	"mid_side_b2": true,
+	"bins": 1280,
+	"unstable_bins": 7,
+	"reduction_bins": 565,
+	"band": {
+		"1": {
+			"sr": 11025,
+			"hl": 108,
+			"n_fft": 2048,
+			"crop_start": 0,
+			"crop_stop": 374,
+			"lpf_start": 92,
+			"lpf_stop": 186,
+			"res_type": "polyphase"
+		},
+		"2": {
+			"sr": 22050,
+			"hl": 216,
+			"n_fft": 1536,
+			"crop_start": 0,
+			"crop_stop": 424,
+			"hpf_start": 68,
+			"hpf_stop": 34,
+			"lpf_start": 348,
+			"lpf_stop": 418,
+			"res_type": "polyphase"
+		},	
+		"3": {
+			"sr": 44100,
+			"hl": 432,
+			"n_fft": 1280,
+			"crop_start": 132,
+			"crop_stop": 614,
+			"hpf_start": 172,
+			"hpf_stop": 144,
+			"res_type": "polyphase"
+		}
+	},
+	"sr": 44100,
+	"pre_filter_start": 1280,
+	"pre_filter_stop": 1280
+}

lib/uvr5_pack/lib_v5/nets.py

@@ -0,0 +1,123 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+import layers
+from . import spec_utils
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 16)
+        self.stg1_high_band_net = BaseASPPNet(2, 16)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(18, 8, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(8, 16)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(16, 32)
+
+        self.out = nn.Conv2d(32, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(16, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(16, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_123812KB.py

@@ -0,0 +1,122 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import layers_123821KB as layers
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 32)
+        self.stg1_high_band_net = BaseASPPNet(2, 32)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(16, 32)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(32, 64)
+
+        self.out = nn.Conv2d(64, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(32, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_123821KB.py

@@ -0,0 +1,122 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import layers_123821KB as layers
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 32)
+        self.stg1_high_band_net = BaseASPPNet(2, 32)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(16, 32)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(32, 64)
+
+        self.out = nn.Conv2d(64, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(32, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_33966KB.py

@@ -0,0 +1,122 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import layers_33966KB as layers
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16, 32)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 16)
+        self.stg1_high_band_net = BaseASPPNet(2, 16)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(18, 8, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(8, 16)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(16, 32)
+
+        self.out = nn.Conv2d(32, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(16, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(16, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_537227KB.py

@@ -0,0 +1,123 @@
+import torch
+import numpy as np
+from torch import nn
+import torch.nn.functional as F
+
+from . import layers_537238KB as layers
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 64)
+        self.stg1_high_band_net = BaseASPPNet(2, 64)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(32, 64)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(130, 64, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(64, 128)
+
+        self.out = nn.Conv2d(128, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(64, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(64, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_537238KB.py

@@ -0,0 +1,123 @@
+import torch
+import numpy as np
+from torch import nn
+import torch.nn.functional as F
+
+from . import layers_537238KB as layers
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 64)
+        self.stg1_high_band_net = BaseASPPNet(2, 64)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(32, 64)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(130, 64, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(64, 128)
+
+        self.out = nn.Conv2d(128, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(64, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(64, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_61968KB.py

@@ -0,0 +1,122 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+
+from . import layers_123821KB as layers
+
+
+class BaseASPPNet(nn.Module):
+    def __init__(self, nin, ch, dilations=(4, 8, 16)):
+        super(BaseASPPNet, self).__init__()
+        self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
+        self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
+        self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
+        self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
+
+        self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
+
+        self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
+        self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
+        self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
+        self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
+
+    def __call__(self, x):
+        h, e1 = self.enc1(x)
+        h, e2 = self.enc2(h)
+        h, e3 = self.enc3(h)
+        h, e4 = self.enc4(h)
+
+        h = self.aspp(h)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedASPPNet(nn.Module):
+    def __init__(self, n_fft):
+        super(CascadedASPPNet, self).__init__()
+        self.stg1_low_band_net = BaseASPPNet(2, 32)
+        self.stg1_high_band_net = BaseASPPNet(2, 32)
+
+        self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
+        self.stg2_full_band_net = BaseASPPNet(16, 32)
+
+        self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
+        self.stg3_full_band_net = BaseASPPNet(32, 64)
+
+        self.out = nn.Conv2d(64, 2, 1, bias=False)
+        self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
+        self.aux2_out = nn.Conv2d(32, 2, 1, bias=False)
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+
+        self.offset = 128
+
+    def forward(self, x, aggressiveness=None):
+        mix = x.detach()
+        x = x.clone()
+
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        aux1 = torch.cat(
+            [
+                self.stg1_low_band_net(x[:, :, :bandw]),
+                self.stg1_high_band_net(x[:, :, bandw:]),
+            ],
+            dim=2,
+        )
+
+        h = torch.cat([x, aux1], dim=1)
+        aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
+
+        h = torch.cat([x, aux1, aux2], dim=1)
+        h = self.stg3_full_band_net(self.stg3_bridge(h))
+
+        mask = torch.sigmoid(self.out(h))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux1 = torch.sigmoid(self.aux1_out(aux1))
+            aux1 = F.pad(
+                input=aux1,
+                pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
+                mode="replicate",
+            )
+            aux2 = torch.sigmoid(self.aux2_out(aux2))
+            aux2 = F.pad(
+                input=aux2,
+                pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
+                mode="replicate",
+            )
+            return mask * mix, aux1 * mix, aux2 * mix
+        else:
+            if aggressiveness:
+                mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
+                    mask[:, :, : aggressiveness["split_bin"]],
+                    1 + aggressiveness["value"] / 3,
+                )
+                mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
+                    mask[:, :, aggressiveness["split_bin"] :],
+                    1 + aggressiveness["value"],
+                )
+
+            return mask * mix
+
+    def predict(self, x_mag, aggressiveness=None):
+        h = self.forward(x_mag, aggressiveness)
+
+        if self.offset > 0:
+            h = h[:, :, :, self.offset : -self.offset]
+            assert h.size()[3] > 0
+
+        return h

lib/uvr5_pack/lib_v5/nets_new.py

@@ -0,0 +1,132 @@
+import torch
+from torch import nn
+import torch.nn.functional as F
+from . import layers_new
+
+
+class BaseNet(nn.Module):
+    def __init__(
+        self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))
+    ):
+        super(BaseNet, self).__init__()
+        self.enc1 = layers_new.Conv2DBNActiv(nin, nout, 3, 1, 1)
+        self.enc2 = layers_new.Encoder(nout, nout * 2, 3, 2, 1)
+        self.enc3 = layers_new.Encoder(nout * 2, nout * 4, 3, 2, 1)
+        self.enc4 = layers_new.Encoder(nout * 4, nout * 6, 3, 2, 1)
+        self.enc5 = layers_new.Encoder(nout * 6, nout * 8, 3, 2, 1)
+
+        self.aspp = layers_new.ASPPModule(nout * 8, nout * 8, dilations, dropout=True)
+
+        self.dec4 = layers_new.Decoder(nout * (6 + 8), nout * 6, 3, 1, 1)
+        self.dec3 = layers_new.Decoder(nout * (4 + 6), nout * 4, 3, 1, 1)
+        self.dec2 = layers_new.Decoder(nout * (2 + 4), nout * 2, 3, 1, 1)
+        self.lstm_dec2 = layers_new.LSTMModule(nout * 2, nin_lstm, nout_lstm)
+        self.dec1 = layers_new.Decoder(nout * (1 + 2) + 1, nout * 1, 3, 1, 1)
+
+    def __call__(self, x):
+        e1 = self.enc1(x)
+        e2 = self.enc2(e1)
+        e3 = self.enc3(e2)
+        e4 = self.enc4(e3)
+        e5 = self.enc5(e4)
+
+        h = self.aspp(e5)
+
+        h = self.dec4(h, e4)
+        h = self.dec3(h, e3)
+        h = self.dec2(h, e2)
+        h = torch.cat([h, self.lstm_dec2(h)], dim=1)
+        h = self.dec1(h, e1)
+
+        return h
+
+
+class CascadedNet(nn.Module):
+    def __init__(self, n_fft, nout=32, nout_lstm=128):
+        super(CascadedNet, self).__init__()
+
+        self.max_bin = n_fft // 2
+        self.output_bin = n_fft // 2 + 1
+        self.nin_lstm = self.max_bin // 2
+        self.offset = 64
+
+        self.stg1_low_band_net = nn.Sequential(
+            BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm),
+            layers_new.Conv2DBNActiv(nout // 2, nout // 4, 1, 1, 0),
+        )
+
+        self.stg1_high_band_net = BaseNet(
+            2, nout // 4, self.nin_lstm // 2, nout_lstm // 2
+        )
+
+        self.stg2_low_band_net = nn.Sequential(
+            BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm),
+            layers_new.Conv2DBNActiv(nout, nout // 2, 1, 1, 0),
+        )
+        self.stg2_high_band_net = BaseNet(
+            nout // 4 + 2, nout // 2, self.nin_lstm // 2, nout_lstm // 2
+        )
+
+        self.stg3_full_band_net = BaseNet(
+            3 * nout // 4 + 2, nout, self.nin_lstm, nout_lstm
+        )
+
+        self.out = nn.Conv2d(nout, 2, 1, bias=False)
+        self.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False)
+
+    def forward(self, x):
+        x = x[:, :, : self.max_bin]
+
+        bandw = x.size()[2] // 2
+        l1_in = x[:, :, :bandw]
+        h1_in = x[:, :, bandw:]
+        l1 = self.stg1_low_band_net(l1_in)
+        h1 = self.stg1_high_band_net(h1_in)
+        aux1 = torch.cat([l1, h1], dim=2)
+
+        l2_in = torch.cat([l1_in, l1], dim=1)
+        h2_in = torch.cat([h1_in, h1], dim=1)
+        l2 = self.stg2_low_band_net(l2_in)
+        h2 = self.stg2_high_band_net(h2_in)
+        aux2 = torch.cat([l2, h2], dim=2)
+
+        f3_in = torch.cat([x, aux1, aux2], dim=1)
+        f3 = self.stg3_full_band_net(f3_in)
+
+        mask = torch.sigmoid(self.out(f3))
+        mask = F.pad(
+            input=mask,
+            pad=(0, 0, 0, self.output_bin - mask.size()[2]),
+            mode="replicate",
+        )
+
+        if self.training:
+            aux = torch.cat([aux1, aux2], dim=1)
+            aux = torch.sigmoid(self.aux_out(aux))
+            aux = F.pad(
+                input=aux,
+                pad=(0, 0, 0, self.output_bin - aux.size()[2]),
+                mode="replicate",
+            )
+            return mask, aux
+        else:
+            return mask
+
+    def predict_mask(self, x):
+        mask = self.forward(x)
+
+        if self.offset > 0:
+            mask = mask[:, :, :, self.offset : -self.offset]
+            assert mask.size()[3] > 0
+
+        return mask
+
+    def predict(self, x, aggressiveness=None):
+        mask = self.forward(x)
+        pred_mag = x * mask
+
+        if self.offset > 0:
+            pred_mag = pred_mag[:, :, :, self.offset : -self.offset]
+            assert pred_mag.size()[3] > 0
+
+        return pred_mag

lib/uvr5_pack/lib_v5/spec_utils.py

@@ -0,0 +1,667 @@
+import os, librosa
+import numpy as np
+import soundfile as sf
+from tqdm import tqdm
+import json, math, hashlib
+
+
+def crop_center(h1, h2):
+    h1_shape = h1.size()
+    h2_shape = h2.size()
+
+    if h1_shape[3] == h2_shape[3]:
+        return h1
+    elif h1_shape[3] < h2_shape[3]:
+        raise ValueError("h1_shape[3] must be greater than h2_shape[3]")
+
+    # s_freq = (h2_shape[2] - h1_shape[2]) // 2
+    # e_freq = s_freq + h1_shape[2]
+    s_time = (h1_shape[3] - h2_shape[3]) // 2
+    e_time = s_time + h2_shape[3]
+    h1 = h1[:, :, :, s_time:e_time]
+
+    return h1
+
+
+def wave_to_spectrogram(
+    wave, hop_length, n_fft, mid_side=False, mid_side_b2=False, reverse=False
+):
+    if reverse:
+        wave_left = np.flip(np.asfortranarray(wave[0]))
+        wave_right = np.flip(np.asfortranarray(wave[1]))
+    elif mid_side:
+        wave_left = np.asfortranarray(np.add(wave[0], wave[1]) / 2)
+        wave_right = np.asfortranarray(np.subtract(wave[0], wave[1]))
+    elif mid_side_b2:
+        wave_left = np.asfortranarray(np.add(wave[1], wave[0] * 0.5))
+        wave_right = np.asfortranarray(np.subtract(wave[0], wave[1] * 0.5))
+    else:
+        wave_left = np.asfortranarray(wave[0])
+        wave_right = np.asfortranarray(wave[1])
+
+    spec_left = librosa.stft(wave_left, n_fft, hop_length=hop_length)
+    spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length)
+
+    spec = np.asfortranarray([spec_left, spec_right])
+
+    return spec
+
+
+def wave_to_spectrogram_mt(
+    wave, hop_length, n_fft, mid_side=False, mid_side_b2=False, reverse=False
+):
+    import threading
+
+    if reverse:
+        wave_left = np.flip(np.asfortranarray(wave[0]))
+        wave_right = np.flip(np.asfortranarray(wave[1]))
+    elif mid_side:
+        wave_left = np.asfortranarray(np.add(wave[0], wave[1]) / 2)
+        wave_right = np.asfortranarray(np.subtract(wave[0], wave[1]))
+    elif mid_side_b2:
+        wave_left = np.asfortranarray(np.add(wave[1], wave[0] * 0.5))
+        wave_right = np.asfortranarray(np.subtract(wave[0], wave[1] * 0.5))
+    else:
+        wave_left = np.asfortranarray(wave[0])
+        wave_right = np.asfortranarray(wave[1])
+
+    def run_thread(**kwargs):
+        global spec_left
+        spec_left = librosa.stft(**kwargs)
+
+    thread = threading.Thread(
+        target=run_thread,
+        kwargs={"y": wave_left, "n_fft": n_fft, "hop_length": hop_length},
+    )
+    thread.start()
+    spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length)
+    thread.join()
+
+    spec = np.asfortranarray([spec_left, spec_right])
+
+    return spec
+
+
+def combine_spectrograms(specs, mp):
+    l = min([specs[i].shape[2] for i in specs])
+    spec_c = np.zeros(shape=(2, mp.param["bins"] + 1, l), dtype=np.complex64)
+    offset = 0
+    bands_n = len(mp.param["band"])
+
+    for d in range(1, bands_n + 1):
+        h = mp.param["band"][d]["crop_stop"] - mp.param["band"][d]["crop_start"]
+        spec_c[:, offset : offset + h, :l] = specs[d][
+            :, mp.param["band"][d]["crop_start"] : mp.param["band"][d]["crop_stop"], :l
+        ]
+        offset += h
+
+    if offset > mp.param["bins"]:
+        raise ValueError("Too much bins")
+
+    # lowpass fiter
+    if (
+        mp.param["pre_filter_start"] > 0
+    ):  # and mp.param['band'][bands_n]['res_type'] in ['scipy', 'polyphase']:
+        if bands_n == 1:
+            spec_c = fft_lp_filter(
+                spec_c, mp.param["pre_filter_start"], mp.param["pre_filter_stop"]
+            )
+        else:
+            gp = 1
+            for b in range(
+                mp.param["pre_filter_start"] + 1, mp.param["pre_filter_stop"]
+            ):
+                g = math.pow(
+                    10, -(b - mp.param["pre_filter_start"]) * (3.5 - gp) / 20.0
+                )
+                gp = g
+                spec_c[:, b, :] *= g
+
+    return np.asfortranarray(spec_c)
+
+
+def spectrogram_to_image(spec, mode="magnitude"):
+    if mode == "magnitude":
+        if np.iscomplexobj(spec):
+            y = np.abs(spec)
+        else:
+            y = spec
+        y = np.log10(y**2 + 1e-8)
+    elif mode == "phase":
+        if np.iscomplexobj(spec):
+            y = np.angle(spec)
+        else:
+            y = spec
+
+    y -= y.min()
+    y *= 255 / y.max()
+    img = np.uint8(y)
+
+    if y.ndim == 3:
+        img = img.transpose(1, 2, 0)
+        img = np.concatenate([np.max(img, axis=2, keepdims=True), img], axis=2)
+
+    return img
+
+
+def reduce_vocal_aggressively(X, y, softmask):
+    v = X - y
+    y_mag_tmp = np.abs(y)
+    v_mag_tmp = np.abs(v)
+
+    v_mask = v_mag_tmp > y_mag_tmp
+    y_mag = np.clip(y_mag_tmp - v_mag_tmp * v_mask * softmask, 0, np.inf)
+
+    return y_mag * np.exp(1.0j * np.angle(y))
+
+
+def mask_silence(mag, ref, thres=0.2, min_range=64, fade_size=32):
+    if min_range < fade_size * 2:
+        raise ValueError("min_range must be >= fade_area * 2")
+
+    mag = mag.copy()
+
+    idx = np.where(ref.mean(axis=(0, 1)) < thres)[0]
+    starts = np.insert(idx[np.where(np.diff(idx) != 1)[0] + 1], 0, idx[0])
+    ends = np.append(idx[np.where(np.diff(idx) != 1)[0]], idx[-1])
+    uninformative = np.where(ends - starts > min_range)[0]
+    if len(uninformative) > 0:
+        starts = starts[uninformative]
+        ends = ends[uninformative]
+        old_e = None
+        for s, e in zip(starts, ends):
+            if old_e is not None and s - old_e < fade_size:
+                s = old_e - fade_size * 2
+
+            if s != 0:
+                weight = np.linspace(0, 1, fade_size)
+                mag[:, :, s : s + fade_size] += weight * ref[:, :, s : s + fade_size]
+            else:
+                s -= fade_size
+
+            if e != mag.shape[2]:
+                weight = np.linspace(1, 0, fade_size)
+                mag[:, :, e - fade_size : e] += weight * ref[:, :, e - fade_size : e]
+            else:
+                e += fade_size
+
+            mag[:, :, s + fade_size : e - fade_size] += ref[
+                :, :, s + fade_size : e - fade_size
+            ]
+            old_e = e
+
+    return mag
+
+
+def align_wave_head_and_tail(a, b):
+    l = min([a[0].size, b[0].size])
+
+    return a[:l, :l], b[:l, :l]
+
+
+def cache_or_load(mix_path, inst_path, mp):
+    mix_basename = os.path.splitext(os.path.basename(mix_path))[0]
+    inst_basename = os.path.splitext(os.path.basename(inst_path))[0]
+
+    cache_dir = "mph{}".format(
+        hashlib.sha1(json.dumps(mp.param, sort_keys=True).encode("utf-8")).hexdigest()
+    )
+    mix_cache_dir = os.path.join("cache", cache_dir)
+    inst_cache_dir = os.path.join("cache", cache_dir)
+
+    os.makedirs(mix_cache_dir, exist_ok=True)
+    os.makedirs(inst_cache_dir, exist_ok=True)
+
+    mix_cache_path = os.path.join(mix_cache_dir, mix_basename + ".npy")
+    inst_cache_path = os.path.join(inst_cache_dir, inst_basename + ".npy")
+
+    if os.path.exists(mix_cache_path) and os.path.exists(inst_cache_path):
+        X_spec_m = np.load(mix_cache_path)
+        y_spec_m = np.load(inst_cache_path)
+    else:
+        X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {}
+
+        for d in range(len(mp.param["band"]), 0, -1):
+            bp = mp.param["band"][d]
+
+            if d == len(mp.param["band"]):  # high-end band
+                X_wave[d], _ = librosa.load(
+                    mix_path, bp["sr"], False, dtype=np.float32, res_type=bp["res_type"]
+                )
+                y_wave[d], _ = librosa.load(
+                    inst_path,
+                    bp["sr"],
+                    False,
+                    dtype=np.float32,
+                    res_type=bp["res_type"],
+                )
+            else:  # lower bands
+                X_wave[d] = librosa.resample(
+                    X_wave[d + 1],
+                    mp.param["band"][d + 1]["sr"],
+                    bp["sr"],
+                    res_type=bp["res_type"],
+                )
+                y_wave[d] = librosa.resample(
+                    y_wave[d + 1],
+                    mp.param["band"][d + 1]["sr"],
+                    bp["sr"],
+                    res_type=bp["res_type"],
+                )
+
+            X_wave[d], y_wave[d] = align_wave_head_and_tail(X_wave[d], y_wave[d])
+
+            X_spec_s[d] = wave_to_spectrogram(
+                X_wave[d],
+                bp["hl"],
+                bp["n_fft"],
+                mp.param["mid_side"],
+                mp.param["mid_side_b2"],
+                mp.param["reverse"],
+            )
+            y_spec_s[d] = wave_to_spectrogram(
+                y_wave[d],
+                bp["hl"],
+                bp["n_fft"],
+                mp.param["mid_side"],
+                mp.param["mid_side_b2"],
+                mp.param["reverse"],
+            )
+
+        del X_wave, y_wave
+
+        X_spec_m = combine_spectrograms(X_spec_s, mp)
+        y_spec_m = combine_spectrograms(y_spec_s, mp)
+
+        if X_spec_m.shape != y_spec_m.shape:
+            raise ValueError("The combined spectrograms are different: " + mix_path)
+
+        _, ext = os.path.splitext(mix_path)
+
+        np.save(mix_cache_path, X_spec_m)
+        np.save(inst_cache_path, y_spec_m)
+
+    return X_spec_m, y_spec_m
+
+
+def spectrogram_to_wave(spec, hop_length, mid_side, mid_side_b2, reverse):
+    spec_left = np.asfortranarray(spec[0])
+    spec_right = np.asfortranarray(spec[1])
+
+    wave_left = librosa.istft(spec_left, hop_length=hop_length)
+    wave_right = librosa.istft(spec_right, hop_length=hop_length)
+
+    if reverse:
+        return np.asfortranarray([np.flip(wave_left), np.flip(wave_right)])
+    elif mid_side:
+        return np.asfortranarray(
+            [np.add(wave_left, wave_right / 2), np.subtract(wave_left, wave_right / 2)]
+        )
+    elif mid_side_b2:
+        return np.asfortranarray(
+            [
+                np.add(wave_right / 1.25, 0.4 * wave_left),
+                np.subtract(wave_left / 1.25, 0.4 * wave_right),
+            ]
+        )
+    else:
+        return np.asfortranarray([wave_left, wave_right])
+
+
+def spectrogram_to_wave_mt(spec, hop_length, mid_side, reverse, mid_side_b2):
+    import threading
+
+    spec_left = np.asfortranarray(spec[0])
+    spec_right = np.asfortranarray(spec[1])
+
+    def run_thread(**kwargs):
+        global wave_left
+        wave_left = librosa.istft(**kwargs)
+
+    thread = threading.Thread(
+        target=run_thread, kwargs={"stft_matrix": spec_left, "hop_length": hop_length}
+    )
+    thread.start()
+    wave_right = librosa.istft(spec_right, hop_length=hop_length)
+    thread.join()
+
+    if reverse:
+        return np.asfortranarray([np.flip(wave_left), np.flip(wave_right)])
+    elif mid_side:
+        return np.asfortranarray(
+            [np.add(wave_left, wave_right / 2), np.subtract(wave_left, wave_right / 2)]
+        )
+    elif mid_side_b2:
+        return np.asfortranarray(
+            [
+                np.add(wave_right / 1.25, 0.4 * wave_left),
+                np.subtract(wave_left / 1.25, 0.4 * wave_right),
+            ]
+        )
+    else:
+        return np.asfortranarray([wave_left, wave_right])
+
+
+def cmb_spectrogram_to_wave(spec_m, mp, extra_bins_h=None, extra_bins=None):
+    wave_band = {}
+    bands_n = len(mp.param["band"])
+    offset = 0
+
+    for d in range(1, bands_n + 1):
+        bp = mp.param["band"][d]
+        spec_s = np.ndarray(
+            shape=(2, bp["n_fft"] // 2 + 1, spec_m.shape[2]), dtype=complex
+        )
+        h = bp["crop_stop"] - bp["crop_start"]
+        spec_s[:, bp["crop_start"] : bp["crop_stop"], :] = spec_m[
+            :, offset : offset + h, :
+        ]
+
+        offset += h
+        if d == bands_n:  # higher
+            if extra_bins_h:  # if --high_end_process bypass
+                max_bin = bp["n_fft"] // 2
+                spec_s[:, max_bin - extra_bins_h : max_bin, :] = extra_bins[
+                    :, :extra_bins_h, :
+                ]
+            if bp["hpf_start"] > 0:
+                spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1)
+            if bands_n == 1:
+                wave = spectrogram_to_wave(
+                    spec_s,
+                    bp["hl"],
+                    mp.param["mid_side"],
+                    mp.param["mid_side_b2"],
+                    mp.param["reverse"],
+                )
+            else:
+                wave = np.add(
+                    wave,
+                    spectrogram_to_wave(
+                        spec_s,
+                        bp["hl"],
+                        mp.param["mid_side"],
+                        mp.param["mid_side_b2"],
+                        mp.param["reverse"],
+                    ),
+                )
+        else:
+            sr = mp.param["band"][d + 1]["sr"]
+            if d == 1:  # lower
+                spec_s = fft_lp_filter(spec_s, bp["lpf_start"], bp["lpf_stop"])
+                wave = librosa.resample(
+                    spectrogram_to_wave(
+                        spec_s,
+                        bp["hl"],
+                        mp.param["mid_side"],
+                        mp.param["mid_side_b2"],
+                        mp.param["reverse"],
+                    ),
+                    bp["sr"],
+                    sr,
+                    res_type="sinc_fastest",
+                )
+            else:  # mid
+                spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1)
+                spec_s = fft_lp_filter(spec_s, bp["lpf_start"], bp["lpf_stop"])
+                wave2 = np.add(
+                    wave,
+                    spectrogram_to_wave(
+                        spec_s,
+                        bp["hl"],
+                        mp.param["mid_side"],
+                        mp.param["mid_side_b2"],
+                        mp.param["reverse"],
+                    ),
+                )
+                # wave = librosa.core.resample(wave2, bp['sr'], sr, res_type="sinc_fastest")
+                wave = librosa.core.resample(wave2, bp["sr"], sr, res_type="scipy")
+
+    return wave.T
+
+
+def fft_lp_filter(spec, bin_start, bin_stop):
+    g = 1.0
+    for b in range(bin_start, bin_stop):
+        g -= 1 / (bin_stop - bin_start)
+        spec[:, b, :] = g * spec[:, b, :]
+
+    spec[:, bin_stop:, :] *= 0
+
+    return spec
+
+
+def fft_hp_filter(spec, bin_start, bin_stop):
+    g = 1.0
+    for b in range(bin_start, bin_stop, -1):
+        g -= 1 / (bin_start - bin_stop)
+        spec[:, b, :] = g * spec[:, b, :]
+
+    spec[:, 0 : bin_stop + 1, :] *= 0
+
+    return spec
+
+
+def mirroring(a, spec_m, input_high_end, mp):
+    if "mirroring" == a:
+        mirror = np.flip(
+            np.abs(
+                spec_m[
+                    :,
+                    mp.param["pre_filter_start"]
+                    - 10
+                    - input_high_end.shape[1] : mp.param["pre_filter_start"]
+                    - 10,
+                    :,
+                ]
+            ),
+            1,
+        )
+        mirror = mirror * np.exp(1.0j * np.angle(input_high_end))
+
+        return np.where(
+            np.abs(input_high_end) <= np.abs(mirror), input_high_end, mirror
+        )
+
+    if "mirroring2" == a:
+        mirror = np.flip(
+            np.abs(
+                spec_m[
+                    :,
+                    mp.param["pre_filter_start"]
+                    - 10
+                    - input_high_end.shape[1] : mp.param["pre_filter_start"]
+                    - 10,
+                    :,
+                ]
+            ),
+            1,
+        )
+        mi = np.multiply(mirror, input_high_end * 1.7)
+
+        return np.where(np.abs(input_high_end) <= np.abs(mi), input_high_end, mi)
+
+
+def ensembling(a, specs):
+    for i in range(1, len(specs)):
+        if i == 1:
+            spec = specs[0]
+
+        ln = min([spec.shape[2], specs[i].shape[2]])
+        spec = spec[:, :, :ln]
+        specs[i] = specs[i][:, :, :ln]
+
+        if "min_mag" == a:
+            spec = np.where(np.abs(specs[i]) <= np.abs(spec), specs[i], spec)
+        if "max_mag" == a:
+            spec = np.where(np.abs(specs[i]) >= np.abs(spec), specs[i], spec)
+
+    return spec
+
+
+def stft(wave, nfft, hl):
+    wave_left = np.asfortranarray(wave[0])
+    wave_right = np.asfortranarray(wave[1])
+    spec_left = librosa.stft(wave_left, nfft, hop_length=hl)
+    spec_right = librosa.stft(wave_right, nfft, hop_length=hl)
+    spec = np.asfortranarray([spec_left, spec_right])
+
+    return spec
+
+
+def istft(spec, hl):
+    spec_left = np.asfortranarray(spec[0])
+    spec_right = np.asfortranarray(spec[1])
+
+    wave_left = librosa.istft(spec_left, hop_length=hl)
+    wave_right = librosa.istft(spec_right, hop_length=hl)
+    wave = np.asfortranarray([wave_left, wave_right])
+
+
+if __name__ == "__main__":
+    import cv2
+    import sys
+    import time
+    import argparse
+    from model_param_init import ModelParameters
+
+    p = argparse.ArgumentParser()
+    p.add_argument(
+        "--algorithm",
+        "-a",
+        type=str,
+        choices=["invert", "invert_p", "min_mag", "max_mag", "deep", "align"],
+        default="min_mag",
+    )
+    p.add_argument(
+        "--model_params",
+        "-m",
+        type=str,
+        default=os.path.join("modelparams", "1band_sr44100_hl512.json"),
+    )
+    p.add_argument("--output_name", "-o", type=str, default="output")
+    p.add_argument("--vocals_only", "-v", action="store_true")
+    p.add_argument("input", nargs="+")
+    args = p.parse_args()
+
+    start_time = time.time()
+
+    if args.algorithm.startswith("invert") and len(args.input) != 2:
+        raise ValueError("There should be two input files.")
+
+    if not args.algorithm.startswith("invert") and len(args.input) < 2:
+        raise ValueError("There must be at least two input files.")
+
+    wave, specs = {}, {}
+    mp = ModelParameters(args.model_params)
+
+    for i in range(len(args.input)):
+        spec = {}
+
+        for d in range(len(mp.param["band"]), 0, -1):
+            bp = mp.param["band"][d]
+
+            if d == len(mp.param["band"]):  # high-end band
+                wave[d], _ = librosa.load(
+                    args.input[i],
+                    bp["sr"],
+                    False,
+                    dtype=np.float32,
+                    res_type=bp["res_type"],
+                )
+
+                if len(wave[d].shape) == 1:  # mono to stereo
+                    wave[d] = np.array([wave[d], wave[d]])
+            else:  # lower bands
+                wave[d] = librosa.resample(
+                    wave[d + 1],
+                    mp.param["band"][d + 1]["sr"],
+                    bp["sr"],
+                    res_type=bp["res_type"],
+                )
+
+            spec[d] = wave_to_spectrogram(
+                wave[d],
+                bp["hl"],
+                bp["n_fft"],
+                mp.param["mid_side"],
+                mp.param["mid_side_b2"],
+                mp.param["reverse"],
+            )
+
+        specs[i] = combine_spectrograms(spec, mp)
+
+    del wave
+
+    if args.algorithm == "deep":
+        d_spec = np.where(np.abs(specs[0]) <= np.abs(spec[1]), specs[0], spec[1])
+        v_spec = d_spec - specs[1]
+        sf.write(
+            os.path.join("{}.wav".format(args.output_name)),
+            cmb_spectrogram_to_wave(v_spec, mp),
+            mp.param["sr"],
+        )
+
+    if args.algorithm.startswith("invert"):
+        ln = min([specs[0].shape[2], specs[1].shape[2]])
+        specs[0] = specs[0][:, :, :ln]
+        specs[1] = specs[1][:, :, :ln]
+
+        if "invert_p" == args.algorithm:
+            X_mag = np.abs(specs[0])
+            y_mag = np.abs(specs[1])
+            max_mag = np.where(X_mag >= y_mag, X_mag, y_mag)
+            v_spec = specs[1] - max_mag * np.exp(1.0j * np.angle(specs[0]))
+        else:
+            specs[1] = reduce_vocal_aggressively(specs[0], specs[1], 0.2)
+            v_spec = specs[0] - specs[1]
+
+            if not args.vocals_only:
+                X_mag = np.abs(specs[0])
+                y_mag = np.abs(specs[1])
+                v_mag = np.abs(v_spec)
+
+                X_image = spectrogram_to_image(X_mag)
+                y_image = spectrogram_to_image(y_mag)
+                v_image = spectrogram_to_image(v_mag)
+
+                cv2.imwrite("{}_X.png".format(args.output_name), X_image)
+                cv2.imwrite("{}_y.png".format(args.output_name), y_image)
+                cv2.imwrite("{}_v.png".format(args.output_name), v_image)
+
+                sf.write(
+                    "{}_X.wav".format(args.output_name),
+                    cmb_spectrogram_to_wave(specs[0], mp),
+                    mp.param["sr"],
+                )
+                sf.write(
+                    "{}_y.wav".format(args.output_name),
+                    cmb_spectrogram_to_wave(specs[1], mp),
+                    mp.param["sr"],
+                )
+
+        sf.write(
+            "{}_v.wav".format(args.output_name),
+            cmb_spectrogram_to_wave(v_spec, mp),
+            mp.param["sr"],
+        )
+    else:
+        if not args.algorithm == "deep":
+            sf.write(
+                os.path.join("ensembled", "{}.wav".format(args.output_name)),
+                cmb_spectrogram_to_wave(ensembling(args.algorithm, specs), mp),
+                mp.param["sr"],
+            )
+
+    if args.algorithm == "align":
+        trackalignment = [
+            {
+                "file1": '"{}"'.format(args.input[0]),
+                "file2": '"{}"'.format(args.input[1]),
+            }
+        ]
+
+        for i, e in tqdm(enumerate(trackalignment), desc="Performing Alignment..."):
+            os.system(f"python lib/align_tracks.py {e['file1']} {e['file2']}")
+
+    # print('Total time: {0:.{1}f}s'.format(time.time() - start_time, 1))

lib/uvr5_pack/name_params.json

@@ -0,0 +1,263 @@
+{
+    "equivalent" : [
+        {
+            "model_hash_name" : [
+                {
+                    "hash_name": "47939caf0cfe52a0e81442b85b971dfd",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100"
+                },
+                {
+                    "hash_name": "4e4ecb9764c50a8c414fee6e10395bbe",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_v2.json",
+                    "param_name": "4band_v2"
+                },
+                {
+                    "hash_name": "ca106edd563e034bde0bdec4bb7a4b36",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_v2.json",
+                    "param_name": "4band_v2"
+                },
+                {
+                    "hash_name": "e60a1e84803ce4efc0a6551206cc4b71",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100"
+                },
+                {
+                    "hash_name": "a82f14e75892e55e994376edbf0c8435",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100"
+                },
+                {
+                    "hash_name": "6dd9eaa6f0420af9f1d403aaafa4cc06",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_v2_sn.json",
+                    "param_name": "4band_v2_sn"
+                },
+                {
+                    "hash_name": "08611fb99bd59eaa79ad27c58d137727",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_v2_sn.json",
+                    "param_name": "4band_v2_sn"
+                },
+                {
+                    "hash_name": "5c7bbca45a187e81abbbd351606164e5",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100_msb2.json",
+                    "param_name": "3band_44100_msb2"
+                },
+                {
+                    "hash_name": "d6b2cb685a058a091e5e7098192d3233",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100_msb2.json",
+                    "param_name": "3band_44100_msb2"
+                },
+                {
+                    "hash_name": "c1b9f38170a7c90e96f027992eb7c62b",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100"
+                },
+                {
+                    "hash_name": "c3448ec923fa0edf3d03a19e633faa53",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100"
+                },
+                {
+                    "hash_name": "68aa2c8093d0080704b200d140f59e54",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100.json",
+                    "param_name": "3band_44100"
+                },
+                {
+                    "hash_name": "fdc83be5b798e4bd29fe00fe6600e147",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100_mid.json",
+                    "param_name": "3band_44100_mid.json"
+                },
+                {
+                    "hash_name": "2ce34bc92fd57f55db16b7a4def3d745",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100_mid.json",
+                    "param_name": "3band_44100_mid.json"
+                },
+                {
+                    "hash_name": "52fdca89576f06cf4340b74a4730ee5f",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100.json"
+                },
+                {
+                    "hash_name": "41191165b05d38fc77f072fa9e8e8a30",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100.json"
+                },
+                {
+                    "hash_name": "89e83b511ad474592689e562d5b1f80e",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/2band_32000.json",
+                    "param_name": "2band_32000.json"
+                },
+                {
+                    "hash_name": "0b954da81d453b716b114d6d7c95177f",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/2band_32000.json",
+                    "param_name": "2band_32000.json"
+                }
+
+            ],
+            "v4 Models": [
+                {
+                    "hash_name": "6a00461c51c2920fd68937d4609ed6c8",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr16000_hl512.json",
+                    "param_name": "1band_sr16000_hl512"
+                },
+                {
+                    "hash_name": "0ab504864d20f1bd378fe9c81ef37140",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr32000_hl512.json",
+                    "param_name": "1band_sr32000_hl512"
+                },
+                {
+                    "hash_name": "7dd21065bf91c10f7fccb57d7d83b07f",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr32000_hl512.json",
+                    "param_name": "1band_sr32000_hl512"
+                },
+                {
+                    "hash_name": "80ab74d65e515caa3622728d2de07d23",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr32000_hl512.json",
+                    "param_name": "1band_sr32000_hl512"
+                },
+                {
+                    "hash_name": "edc115e7fc523245062200c00caa847f",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr33075_hl384.json",
+                    "param_name": "1band_sr33075_hl384"
+                },
+                {
+                    "hash_name": "28063e9f6ab5b341c5f6d3c67f2045b7",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr33075_hl384.json",
+                    "param_name": "1band_sr33075_hl384"
+                },
+                {
+                    "hash_name": "b58090534c52cbc3e9b5104bad666ef2",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl512.json",
+                    "param_name": "1band_sr44100_hl512"
+                },
+                {
+                    "hash_name": "0cdab9947f1b0928705f518f3c78ea8f",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl512.json",
+                    "param_name": "1band_sr44100_hl512"
+                },
+                {
+                    "hash_name": "ae702fed0238afb5346db8356fe25f13",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl1024.json",
+                    "param_name": "1band_sr44100_hl1024"
+                }
+            ]
+        }
+    ],
+    "User Models" : [
+        {
+            "1 Band": [
+                {
+                    "hash_name": "1band_sr16000_hl512",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr16000_hl512.json",
+                    "param_name": "1band_sr16000_hl512"
+                },
+                {
+                    "hash_name": "1band_sr32000_hl512",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr32000_hl512.json",
+                    "param_name": "1band_sr16000_hl512"
+                },
+                {
+                    "hash_name": "1band_sr33075_hl384",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr33075_hl384.json",
+                    "param_name": "1band_sr33075_hl384"
+                },
+                {
+                    "hash_name": "1band_sr44100_hl256",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl256.json",
+                    "param_name": "1band_sr44100_hl256"
+                },
+                {
+                    "hash_name": "1band_sr44100_hl512",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl512.json",
+                    "param_name": "1band_sr44100_hl512"
+                },
+                {
+                    "hash_name": "1band_sr44100_hl1024",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/1band_sr44100_hl1024.json",
+                    "param_name": "1band_sr44100_hl1024"
+                }
+            ],
+            "2 Band": [
+                {
+                    "hash_name": "2band_44100_lofi",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/2band_44100_lofi.json",
+                    "param_name": "2band_44100_lofi"
+                },
+                {
+                    "hash_name": "2band_32000",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/2band_32000.json",
+                    "param_name": "2band_32000"
+                },
+                {
+                    "hash_name": "2band_48000",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/2band_48000.json",
+                    "param_name": "2band_48000"
+                }
+            ],
+            "3 Band": [
+                {
+                    "hash_name": "3band_44100",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100.json",
+                    "param_name": "3band_44100"
+                },
+                {
+                    "hash_name": "3band_44100_mid",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100_mid.json",
+                    "param_name": "3band_44100_mid"
+                },
+                {
+                    "hash_name": "3band_44100_msb2",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/3band_44100_msb2.json",
+                    "param_name": "3band_44100_msb2"
+                }
+            ],
+            "4 Band": [
+                {
+                    "hash_name": "4band_44100",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100.json",
+                    "param_name": "4band_44100"
+                },
+                {
+                    "hash_name": "4band_44100_mid",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100_mid.json",
+                    "param_name": "4band_44100_mid"
+                },
+                {
+                    "hash_name": "4band_44100_msb",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100_msb.json",
+                    "param_name": "4band_44100_msb"
+                },
+                {
+                    "hash_name": "4band_44100_msb2",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100_msb2.json",
+                    "param_name": "4band_44100_msb2"
+                },
+                {
+                    "hash_name": "4band_44100_reverse",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100_reverse.json",
+                    "param_name": "4band_44100_reverse"
+                },
+                {
+                    "hash_name": "4band_44100_sw",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_44100_sw.json",
+                    "param_name": "4band_44100_sw"
+                },
+                {
+                    "hash_name": "4band_v2",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_v2.json",
+                    "param_name": "4band_v2"
+                },
+                {
+                    "hash_name": "4band_v2_sn",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/4band_v2_sn.json",
+                    "param_name": "4band_v2_sn"
+                },
+                {
+                    "hash_name": "tmodelparam",
+                    "model_params": "lib/uvr5_pack/lib_v5/modelparams/tmodelparam.json",
+                    "param_name": "User Model Param Set"
+                }
+            ]
+        }
+    ]
+}

lib/uvr5_pack/utils.py

@@ -0,0 +1,120 @@
+import torch
+import numpy as np
+from tqdm import tqdm
+import json
+
+
+def load_data(file_name: str = "./lib/uvr5_pack/name_params.json") -> dict:
+    with open(file_name, "r") as f:
+        data = json.load(f)
+
+    return data
+
+
+def make_padding(width, cropsize, offset):
+    left = offset
+    roi_size = cropsize - left * 2
+    if roi_size == 0:
+        roi_size = cropsize
+    right = roi_size - (width % roi_size) + left
+
+    return left, right, roi_size
+
+
+def inference(X_spec, device, model, aggressiveness, data):
+    """
+    data ： dic configs
+    """
+
+    def _execute(
+        X_mag_pad, roi_size, n_window, device, model, aggressiveness, is_half=True
+    ):
+        model.eval()
+        with torch.no_grad():
+            preds = []
+
+            iterations = [n_window]
+
+            total_iterations = sum(iterations)
+            for i in tqdm(range(n_window)):
+                start = i * roi_size
+                X_mag_window = X_mag_pad[
+                    None, :, :, start : start + data["window_size"]
+                ]
+                X_mag_window = torch.from_numpy(X_mag_window)
+                if is_half:
+                    X_mag_window = X_mag_window.half()
+                X_mag_window = X_mag_window.to(device)
+
+                pred = model.predict(X_mag_window, aggressiveness)
+
+                pred = pred.detach().cpu().numpy()
+                preds.append(pred[0])
+
+            pred = np.concatenate(preds, axis=2)
+        return pred
+
+    def preprocess(X_spec):
+        X_mag = np.abs(X_spec)
+        X_phase = np.angle(X_spec)
+
+        return X_mag, X_phase
+
+    X_mag, X_phase = preprocess(X_spec)
+
+    coef = X_mag.max()
+    X_mag_pre = X_mag / coef
+
+    n_frame = X_mag_pre.shape[2]
+    pad_l, pad_r, roi_size = make_padding(n_frame, data["window_size"], model.offset)
+    n_window = int(np.ceil(n_frame / roi_size))
+
+    X_mag_pad = np.pad(X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode="constant")
+
+    if list(model.state_dict().values())[0].dtype == torch.float16:
+        is_half = True
+    else:
+        is_half = False
+    pred = _execute(
+        X_mag_pad, roi_size, n_window, device, model, aggressiveness, is_half
+    )
+    pred = pred[:, :, :n_frame]
+
+    if data["tta"]:
+        pad_l += roi_size // 2
+        pad_r += roi_size // 2
+        n_window += 1
+
+        X_mag_pad = np.pad(X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode="constant")
+
+        pred_tta = _execute(
+            X_mag_pad, roi_size, n_window, device, model, aggressiveness, is_half
+        )
+        pred_tta = pred_tta[:, :, roi_size // 2 :]
+        pred_tta = pred_tta[:, :, :n_frame]
+
+        return (pred + pred_tta) * 0.5 * coef, X_mag, np.exp(1.0j * X_phase)
+    else:
+        return pred * coef, X_mag, np.exp(1.0j * X_phase)
+
+
+def _get_name_params(model_path, model_hash):
+    data = load_data()
+    flag = False
+    ModelName = model_path
+    for type in list(data):
+        for model in list(data[type][0]):
+            for i in range(len(data[type][0][model])):
+                if str(data[type][0][model][i]["hash_name"]) == model_hash:
+                    flag = True
+                elif str(data[type][0][model][i]["hash_name"]) in ModelName:
+                    flag = True
+
+                if flag:
+                    model_params_auto = data[type][0][model][i]["model_params"]
+                    param_name_auto = data[type][0][model][i]["param_name"]
+                    if type == "equivalent":
+                        return param_name_auto, model_params_auto
+                    else:
+                        flag = False
+    return param_name_auto, model_params_auto