replace files

tools/app.py

@@ -0,0 +1,146 @@
+import logging
+import os
+
+# os.system("wget -P cvec/ https://huggingface.co/lj1995/VoiceConversionWebUI/resolve/main/hubert_base.pt")
+import gradio as gr
+from dotenv import load_dotenv
+
+from configs.config import Config
+from i18n.i18n import I18nAuto
+from infer.modules.vc.modules import VC
+
+logging.getLogger("numba").setLevel(logging.WARNING)
+logging.getLogger("markdown_it").setLevel(logging.WARNING)
+logging.getLogger("urllib3").setLevel(logging.WARNING)
+logging.getLogger("matplotlib").setLevel(logging.WARNING)
+
+i18n = I18nAuto()
+i18n.print()
+
+load_dotenv()
+config = Config()
+vc = VC(config)
+
+weight_root = os.getenv("weight_root")
+weight_uvr5_root = os.getenv("weight_uvr5_root")
+index_root = "logs"
+names = []
+hubert_model = None
+for name in os.listdir(weight_root):
+    if name.endswith(".pth"):
+        names.append(name)
+index_paths = []
+for root, dirs, files in os.walk(index_root, topdown=False):
+    for name in files:
+        if name.endswith(".index") and "trained" not in name:
+            index_paths.append("%s/%s" % (root, name))
+
+
+app = gr.Blocks()
+with app:
+    with gr.Tabs():
+        with gr.TabItem("在线demo"):
+            gr.Markdown(
+                value="""
+                RVC 在线demo
+                """
+            )
+            sid = gr.Dropdown(label=i18n("推理音色"), choices=sorted(names))
+            with gr.Column():
+                spk_item = gr.Slider(
+                    minimum=0,
+                    maximum=2333,
+                    step=1,
+                    label=i18n("请选择说话人id"),
+                    value=0,
+                    visible=False,
+                    interactive=True,
+                )
+            sid.change(fn=vc.get_vc, inputs=[sid], outputs=[spk_item])
+            gr.Markdown(
+                value=i18n("男转女推荐+12key, 女转男推荐-12key, 如果音域爆炸导致音色失真也可以自己调整到合适音域. ")
+            )
+            vc_input3 = gr.Audio(label="上传音频（长度小于90秒）")
+            vc_transform0 = gr.Number(label=i18n("变调(整数, 半音数量, 升八度12降八度-12)"), value=0)
+            f0method0 = gr.Radio(
+                label=i18n("选择音高提取算法,输入歌声可用pm提速,harvest低音好但巨慢无比,crepe效果好但吃GPU"),
+                choices=["pm", "harvest", "crepe", "rmvpe"],
+                value="pm",
+                interactive=True,
+            )
+            filter_radius0 = gr.Slider(
+                minimum=0,
+                maximum=7,
+                label=i18n(">=3则使用对harvest音高识别的结果使用中值滤波，数值为滤波半径，使用可以削弱哑音"),
+                value=3,
+                step=1,
+                interactive=True,
+            )
+            with gr.Column():
+                file_index1 = gr.Textbox(
+                    label=i18n("特征检索库文件路径,为空则使用下拉的选择结果"),
+                    value="",
+                    interactive=False,
+                    visible=False,
+                )
+            file_index2 = gr.Dropdown(
+                label=i18n("自动检测index路径,下拉式选择(dropdown)"),
+                choices=sorted(index_paths),
+                interactive=True,
+            )
+            index_rate1 = gr.Slider(
+                minimum=0,
+                maximum=1,
+                label=i18n("检索特征占比"),
+                value=0.88,
+                interactive=True,
+            )
+            resample_sr0 = gr.Slider(
+                minimum=0,
+                maximum=48000,
+                label=i18n("后处理重采样至最终采样率，0为不进行重采样"),
+                value=0,
+                step=1,
+                interactive=True,
+            )
+            rms_mix_rate0 = gr.Slider(
+                minimum=0,
+                maximum=1,
+                label=i18n("输入源音量包络替换输出音量包络融合比例，越靠近1越使用输出包络"),
+                value=1,
+                interactive=True,
+            )
+            protect0 = gr.Slider(
+                minimum=0,
+                maximum=0.5,
+                label=i18n("保护清辅音和呼吸声，防止电音撕裂等artifact，拉满0.5不开启，调低加大保护力度但可能降低索引效果"),
+                value=0.33,
+                step=0.01,
+                interactive=True,
+            )
+            f0_file = gr.File(label=i18n("F0曲线文件, 可选, 一行一个音高, 代替默认F0及升降调"))
+            but0 = gr.Button(i18n("转换"), variant="primary")
+            vc_output1 = gr.Textbox(label=i18n("输出信息"))
+            vc_output2 = gr.Audio(label=i18n("输出音频(右下角三个点,点了可以下载)"))
+            but0.click(
+                vc.vc_single,
+                [
+                    spk_item,
+                    vc_input3,
+                    vc_transform0,
+                    f0_file,
+                    f0method0,
+                    file_index1,
+                    file_index2,
+                    # file_big_npy1,
+                    index_rate1,
+                    filter_radius0,
+                    resample_sr0,
+                    rms_mix_rate0,
+                    protect0,
+                ],
+                [vc_output1, vc_output2],
+            )
+
+
+app.launch()

tools/infer_cli.py

@@ -4,11 +4,11 @@ import sys
 
 now_dir = os.getcwd()
 sys.path.append(now_dir)
+from dotenv import load_dotenv
 from scipy.io import wavfile
 
 from configs.config import Config
 from infer.modules.vc.modules import VC
-from dotenv import load_dotenv
 
 ####
 # USAGE

tools/rvc_for_realtime.py

@@ -0,0 +1,331 @@
+import os
+import sys
+import traceback
+from time import time as ttime
+
+import fairseq
+import faiss
+import numpy as np
+import parselmouth
+import pyworld
+import scipy.signal as signal
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torchcrepe
+
+from infer.lib.infer_pack.models import (
+    SynthesizerTrnMs256NSFsid,
+    SynthesizerTrnMs256NSFsid_nono,
+    SynthesizerTrnMs768NSFsid,
+    SynthesizerTrnMs768NSFsid_nono,
+)
+
+now_dir = os.getcwd()
+sys.path.append(now_dir)
+from multiprocessing import Manager as M
+
+from configs.config import Config
+
+Config()
+
+mm = M()
+if config.dml == True:
+
+    def forward_dml(ctx, x, scale):
+        ctx.scale = scale
+        res = x.clone().detach()
+        return res
+
+    fairseq.modules.grad_multiply.GradMultiply.forward = forward_dml
+
+
+# config.device=torch.device("cpu")########强制cpu测试
+# config.is_half=False########强制cpu测试
+class RVC:
+    def __init__(
+        self, key, pth_path, index_path, index_rate, n_cpu, inp_q, opt_q, device
+    ) -> None:
+        """
+        初始化
+        """
+        try:
+            global config
+            self.inp_q = inp_q
+            self.opt_q = opt_q
+            # device="cpu"########强制cpu测试
+            self.device = device
+            self.f0_up_key = key
+            self.time_step = 160 / 16000 * 1000
+            self.f0_min = 50
+            self.f0_max = 1100
+            self.f0_mel_min = 1127 * np.log(1 + self.f0_min / 700)
+            self.f0_mel_max = 1127 * np.log(1 + self.f0_max / 700)
+            self.sr = 16000
+            self.window = 160
+            self.n_cpu = n_cpu
+            if index_rate != 0:
+                self.index = faiss.read_index(index_path)
+                self.big_npy = self.index.reconstruct_n(0, self.index.ntotal)
+                print("index search enabled")
+            self.index_rate = index_rate
+            models, _, _ = fairseq.checkpoint_utils.load_model_ensemble_and_task(
+                ["hubert_base.pt"],
+                suffix="",
+            )
+            hubert_model = models[0]
+            hubert_model = hubert_model.to(config.device)
+            if config.is_half:
+                hubert_model = hubert_model.half()
+            else:
+                hubert_model = hubert_model.float()
+            hubert_model.eval()
+            self.model = hubert_model
+            cpt = torch.load(pth_path, map_location="cpu")
+            self.tgt_sr = cpt["config"][-1]
+            cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]
+            self.if_f0 = cpt.get("f0", 1)
+            self.version = cpt.get("version", "v1")
+            if self.version == "v1":
+                if self.if_f0 == 1:
+                    self.net_g = SynthesizerTrnMs256NSFsid(
+                        *cpt["config"], is_half=config.is_half
+                    )
+                else:
+                    self.net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
+            elif self.version == "v2":
+                if self.if_f0 == 1:
+                    self.net_g = SynthesizerTrnMs768NSFsid(
+                        *cpt["config"], is_half=config.is_half
+                    )
+                else:
+                    self.net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
+            del self.net_g.enc_q
+            print(self.net_g.load_state_dict(cpt["weight"], strict=False))
+            self.net_g.eval().to(device)
+            # print(2333333333,device,config.device,self.device)#net_g是device，hubert是config.device
+            if config.is_half:
+                self.net_g = self.net_g.half()
+            else:
+                self.net_g = self.net_g.float()
+            self.is_half = config.is_half
+        except:
+            print(traceback.format_exc())
+
+    def get_f0_post(self, f0):
+        f0_min = self.f0_min
+        f0_max = self.f0_max
+        f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+        f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+        f0bak = f0.copy()
+        f0_mel = 1127 * np.log(1 + f0 / 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
+        f0_coarse = np.rint(f0_mel).astype(np.int32)
+        return f0_coarse, f0bak
+
+    def get_f0(self, x, f0_up_key, n_cpu, method="harvest"):
+        n_cpu = int(n_cpu)
+        if method == "crepe":
+            return self.get_f0_crepe(x, f0_up_key)
+        if method == "rmvpe":
+            return self.get_f0_rmvpe(x, f0_up_key)
+        if method == "pm":
+            p_len = x.shape[0] // 160
+            f0 = (
+                parselmouth.Sound(x, 16000)
+                .to_pitch_ac(
+                    time_step=0.01,
+                    voicing_threshold=0.6,
+                    pitch_floor=50,
+                    pitch_ceiling=1100,
+                )
+                .selected_array["frequency"]
+            )
+
+            pad_size = (p_len - len(f0) + 1) // 2
+            if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+                # print(pad_size, p_len - len(f0) - pad_size)
+                f0 = np.pad(
+                    f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant"
+                )
+
+            f0 *= pow(2, f0_up_key / 12)
+            return self.get_f0_post(f0)
+        if n_cpu == 1:
+            f0, t = pyworld.harvest(
+                x.astype(np.double),
+                fs=16000,
+                f0_ceil=1100,
+                f0_floor=50,
+                frame_period=10,
+            )
+            f0 = signal.medfilt(f0, 3)
+            f0 *= pow(2, f0_up_key / 12)
+            return self.get_f0_post(f0)
+        f0bak = np.zeros(x.shape[0] // 160, dtype=np.float64)
+        length = len(x)
+        part_length = int(length / n_cpu / 160) * 160
+        ts = ttime()
+        res_f0 = mm.dict()
+        for idx in range(n_cpu):
+            tail = part_length * (idx + 1) + 320
+            if idx == 0:
+                self.inp_q.put((idx, x[:tail], res_f0, n_cpu, ts))
+            else:
+                self.inp_q.put(
+                    (idx, x[part_length * idx - 320 : tail], res_f0, n_cpu, ts)
+                )
+        while 1:
+            res_ts = self.opt_q.get()
+            if res_ts == ts:
+                break
+        f0s = [i[1] for i in sorted(res_f0.items(), key=lambda x: x[0])]
+        for idx, f0 in enumerate(f0s):
+            if idx == 0:
+                f0 = f0[:-3]
+            elif idx != n_cpu - 1:
+                f0 = f0[2:-3]
+            else:
+                f0 = f0[2:-1]
+            f0bak[
+                part_length * idx // 160 : part_length * idx // 160 + f0.shape[0]
+            ] = f0
+        f0bak = signal.medfilt(f0bak, 3)
+        f0bak *= pow(2, f0_up_key / 12)
+        return self.get_f0_post(f0bak)
+
+    def get_f0_crepe(self, x, f0_up_key):
+        if self.device.type == "privateuseone":  ###不支持dml，cpu又太慢用不成，拿pm顶替
+            return self.get_f0(x, f0_up_key, 1, "pm")
+        audio = torch.tensor(np.copy(x))[None].float()
+        # print("using crepe,device:%s"%self.device)
+        f0, pd = torchcrepe.predict(
+            audio,
+            self.sr,
+            160,
+            self.f0_min,
+            self.f0_max,
+            "full",
+            batch_size=512,
+            # device=self.device if self.device.type!="privateuseone" else "cpu",###crepe不用半精度全部是全精度所以不愁###cpu延迟高到没法用
+            device=self.device,
+            return_periodicity=True,
+        )
+        pd = torchcrepe.filter.median(pd, 3)
+        f0 = torchcrepe.filter.mean(f0, 3)
+        f0[pd < 0.1] = 0
+        f0 = f0[0].cpu().numpy()
+        f0 *= pow(2, f0_up_key / 12)
+        return self.get_f0_post(f0)
+
+    def get_f0_rmvpe(self, x, f0_up_key):
+        if hasattr(self, "model_rmvpe") == False:
+            from lib.rmvpe import RMVPE
+
+            print("loading rmvpe model")
+            self.model_rmvpe = RMVPE(
+                # "rmvpe.pt", is_half=self.is_half if self.device.type!="privateuseone" else False, device=self.device if self.device.type!="privateuseone"else "cpu"####dml时强制对rmvpe用cpu跑
+                #  "rmvpe.pt", is_half=False, device=self.device####dml配置
+                # "rmvpe.pt", is_half=False, device="cpu"####锁定cpu配置
+                "rmvpe.pt",
+                is_half=self.is_half,
+                device=self.device,  ####正常逻辑
+            )
+            # self.model_rmvpe = RMVPE("aug2_58000_half.pt", is_half=self.is_half, device=self.device)
+        f0 = self.model_rmvpe.infer_from_audio(x, thred=0.03)
+        f0 *= pow(2, f0_up_key / 12)
+        return self.get_f0_post(f0)
+
+    def infer(
+        self,
+        feats: torch.Tensor,
+        indata: np.ndarray,
+        rate1,
+        rate2,
+        cache_pitch,
+        cache_pitchf,
+        f0method,
+    ) -> np.ndarray:
+        feats = feats.view(1, -1)
+        if config.is_half:
+            feats = feats.half()
+        else:
+            feats = feats.float()
+        feats = feats.to(self.device)
+        t1 = ttime()
+        with torch.no_grad():
+            padding_mask = torch.BoolTensor(feats.shape).to(self.device).fill_(False)
+            inputs = {
+                "source": feats,
+                "padding_mask": padding_mask,
+                "output_layer": 9 if self.version == "v1" else 12,
+            }
+            logits = self.model.extract_features(**inputs)
+            feats = (
+                self.model.final_proj(logits[0]) if self.version == "v1" else logits[0]
+            )
+        t2 = ttime()
+        try:
+            if hasattr(self, "index") and self.index_rate != 0:
+                leng_replace_head = int(rate1 * feats[0].shape[0])
+                npy = feats[0][-leng_replace_head:].cpu().numpy().astype("float32")
+                score, ix = self.index.search(npy, k=8)
+                weight = np.square(1 / score)
+                weight /= weight.sum(axis=1, keepdims=True)
+                npy = np.sum(self.big_npy[ix] * np.expand_dims(weight, axis=2), axis=1)
+                if config.is_half:
+                    npy = npy.astype("float16")
+                feats[0][-leng_replace_head:] = (
+                    torch.from_numpy(npy).unsqueeze(0).to(self.device) * self.index_rate
+                    + (1 - self.index_rate) * feats[0][-leng_replace_head:]
+                )
+            else:
+                print("index search FAIL or disabled")
+        except:
+            traceback.print_exc()
+            print("index search FAIL")
+        feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1)
+        t3 = ttime()
+        if self.if_f0 == 1:
+            pitch, pitchf = self.get_f0(indata, self.f0_up_key, self.n_cpu, f0method)
+            cache_pitch[:] = np.append(cache_pitch[pitch[:-1].shape[0] :], pitch[:-1])
+            cache_pitchf[:] = np.append(
+                cache_pitchf[pitchf[:-1].shape[0] :], pitchf[:-1]
+            )
+            p_len = min(feats.shape[1], 13000, cache_pitch.shape[0])
+        else:
+            cache_pitch, cache_pitchf = None, None
+            p_len = min(feats.shape[1], 13000)
+        t4 = ttime()
+        feats = feats[:, :p_len, :]
+        if self.if_f0 == 1:
+            cache_pitch = cache_pitch[:p_len]
+            cache_pitchf = cache_pitchf[:p_len]
+            cache_pitch = torch.LongTensor(cache_pitch).unsqueeze(0).to(self.device)
+            cache_pitchf = torch.FloatTensor(cache_pitchf).unsqueeze(0).to(self.device)
+        p_len = torch.LongTensor([p_len]).to(self.device)
+        ii = 0  # sid
+        sid = torch.LongTensor([ii]).to(self.device)
+        with torch.no_grad():
+            if self.if_f0 == 1:
+                # print(12222222222,feats.device,p_len.device,cache_pitch.device,cache_pitchf.device,sid.device,rate2)
+                infered_audio = (
+                    self.net_g.infer(
+                        feats, p_len, cache_pitch, cache_pitchf, sid, rate2
+                    )[0][0, 0]
+                    .data.cpu()
+                    .float()
+                )
+            else:
+                infered_audio = (
+                    self.net_g.infer(feats, p_len, sid, rate2)[0][0, 0]
+                    .data.cpu()
+                    .float()
+                )
+        t5 = ttime()
+        print("time->fea-index-f0-model:", t2 - t1, t3 - t2, t4 - t3, t5 - t4)
+        return infered_audio