Add Voxtral MLX streaming backend

Integrates the voxmlx-based Voxtral Mini Realtime streaming pipeline:
- VoxtralStreamingASR and VoxtralStreamingOnlineProcessor
- Incremental audio encoding and token-by-token autoregressive decoding
- Selectable via --backend voxtral-mlx

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Dockerfile

@@ -37,9 +37,10 @@ RUN pip3 install --upgrade pip setuptools wheel && \
 COPY . .
 
 # Install WhisperLiveKit directly, allowing for optional dependencies
+# Example: --build-arg EXTRAS="translation"
 RUN if [ -n "$EXTRAS" ]; then \
       echo "Installing with extras: [$EXTRAS]"; \
-      pip install --no-cache-dir whisperlivekit[$EXTRAS]; \
+      pip install --no-cache-dir "whisperlivekit[$EXTRAS]"; \
     else \
       echo "Installing base package only"; \
       pip install --no-cache-dir whisperlivekit; \

README.md

@@ -147,8 +147,8 @@ async def websocket_endpoint(websocket: WebSocket):
 |-----------|-------------|---------|
 | `--model` | Whisper model size. List and recommandations [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/docs/default_and_custom_models.md) | `small` |
 | `--model-path` | Local .pt file/directory **or** Hugging Face repo ID containing the Whisper model. Overrides `--model`. Recommandations [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/docs/default_and_custom_models.md) | `None` |
-| `--language` | List [here](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/whisperlivekit/whisper/tokenizer.py). If you use `auto`, the model attempts to detect the language automatically, but it tends to bias towards English. | `auto` |
+| `--language` | List [here](docs/supported_languages.md). If you use `auto`, the model attempts to detect the language automatically, but it tends to bias towards English. | `auto` |
-| `--target-language` | If sets, translates using [NLLW](https://github.com/QuentinFuxa/NoLanguageLeftWaiting). [200 languages available](https://github.com/QuentinFuxa/WhisperLiveKit/blob/main/docs/supported_languages.md). If you want to translate to english, you can also use `--direct-english-translation`. The STT model will try to directly output the translation. | `None` |
+| `--target-language` | If sets, translates using [NLLW](https://github.com/QuentinFuxa/NoLanguageLeftWaiting). [200 languages available](docs/supported_languages.md). If you want to translate to english, you can also use `--direct-english-translation`. The STT model will try to directly output the translation. | `None` |
 | `--diarization` | Enable speaker identification | `False` |
 | `--backend-policy` | Streaming strategy: `1`/`simulstreaming` uses AlignAtt SimulStreaming, `2`/`localagreement` uses the LocalAgreement policy | `simulstreaming` |
 | `--backend` | Whisper implementation selector. `auto` picks MLX on macOS (if installed), otherwise Faster-Whisper, otherwise vanilla Whisper. You can also force `mlx-whisper`, `faster-whisper`, `whisper`, or `openai-api` (LocalAgreement only) | `auto` |
@@ -267,7 +267,7 @@ docker run --gpus all -p 8000:8000 --name wlk wlk --model large-v3 --language fr
 #### Customization
 
 - `--build-arg` Options:
-  - `EXTRAS="whisper-timestamped"` - Add extras to the image's installation (no spaces). Remember to set necessary container options!
+  - `EXTRAS="translation"` - Add extras to the image's installation (no spaces). Remember to set necessary container options!
   - `HF_PRECACHE_DIR="./.cache/"` - Pre-load a model cache for faster first-time start
   - `HF_TKN_FILE="./token"` - Add your Hugging Face Hub access token to download gated models
 

chrome-extension/README.md

@@ -6,7 +6,7 @@ Capture the audio of your current tab, transcribe diarize and translate it using
 <img src="https://raw.githubusercontent.com/QuentinFuxa/WhisperLiveKit/refs/heads/main/chrome-extension/demo-extension.png" alt="WhisperLiveKit Demo" width="730">
 
 ## Running this extension
-1. Run `python sync_extension.py` to copy frontend files to the `chrome-extension` directory.
+1. Run `python scripts/sync_extension.py` to copy frontend files to the `chrome-extension` directory.
 2. Load the `chrome-extension` directory in Chrome as an unpacked extension.
 
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "whisperlivekit"
-version = "0.2.16.dev0"
+version = "0.2.18"
 description = "Real-time speech-to-text with speaker diarization using Whisper"
 readme = "README.md"
 authors = [
@@ -35,6 +35,7 @@ dependencies = [
     "torchaudio>=2.0.0",
     "torch>=2.0.0",
     "huggingface-hub>=0.25.0",
+    "faster-whisper>=1.2.0",
     "tqdm",
     "tiktoken",
     'triton>=2.0.0; platform_machine == "x86_64" and (sys_platform == "linux" or sys_platform == "linux2")'
@@ -56,6 +57,7 @@ packages = [
     "whisperlivekit",
     "whisperlivekit.diarization",
     "whisperlivekit.simul_whisper",
+    "whisperlivekit.simul_whisper.mlx",
     "whisperlivekit.whisper",
     "whisperlivekit.whisper.assets",
     "whisperlivekit.whisper.normalizers",
@@ -67,4 +69,5 @@ packages = [
 [tool.setuptools.package-data]
 whisperlivekit = ["web/*.html", "web/*.css", "web/*.js", "web/src/*.svg"]
 "whisperlivekit.whisper.assets" = ["*.tiktoken", "*.npz"]
+"whisperlivekit.whisper.normalizers" = ["*.json"]
 "whisperlivekit.silero_vad_models" = ["*.jit", "*.onnx"]

whisperlivekit/audio_processor.py

@@ -10,7 +10,7 @@ from whisperlivekit.core import (TranscriptionEngine,
                                  online_diarization_factory, online_factory,
                                  online_translation_factory)
 from whisperlivekit.ffmpeg_manager import FFmpegManager, FFmpegState
-from whisperlivekit.silero_vad_iterator import FixedVADIterator
+from whisperlivekit.silero_vad_iterator import FixedVADIterator, OnnxWrapper, load_jit_vad
 from whisperlivekit.timed_objects import (ASRToken, ChangeSpeaker, FrontData,
                                           Segment, Silence, State, Transcript)
 from whisperlivekit.tokens_alignment import TokensAlignment
@@ -85,12 +85,14 @@ class AudioProcessor:
 
         # Models and processing
         self.asr: Any = models.asr
-        self.vac_model: Any = models.vac_model
+        self.vac: Optional[FixedVADIterator] = None
-        if self.args.vac:
-            self.vac: Optional[FixedVADIterator] = FixedVADIterator(models.vac_model)
-        else:
-            self.vac: Optional[FixedVADIterator] = None
 
+        if self.args.vac:
+            if models.vac_session is not None:
+                vac_model = OnnxWrapper(session=models.vac_session)
+                self.vac = FixedVADIterator(vac_model)
+            else:
+                self.vac = FixedVADIterator(load_jit_vad())
         self.ffmpeg_manager: Optional[FFmpegManager] = None
         self.ffmpeg_reader_task: Optional[asyncio.Task] = None
         self._ffmpeg_error: Optional[str] = None
@@ -351,11 +353,14 @@ class AudioProcessor:
                     if item.has_ended:
                         self.diarization.insert_silence(item.duration)
                     continue
-
                 self.diarization.insert_audio_chunk(item)
                 diarization_segments = await self.diarization.diarize()
-                self.state.new_diarization = diarization_segments
+                diar_end = 0.0
-                
+                if diarization_segments:
+                    diar_end = max(getattr(s, "end", 0.0) for s in diarization_segments)
+                async with self.lock:
+                    self.state.new_diarization = diarization_segments
+                    self.state.end_attributed_speaker = max(self.state.end_attributed_speaker, diar_end)
             except Exception as e:
                 logger.warning(f"Exception in diarization_processor: {e}")
                 logger.warning(f"Traceback: {traceback.format_exc()}")

whisperlivekit/backend_support.py

@@ -29,6 +29,13 @@ def mlx_backend_available(warn_on_missing = False):
     return available
 
 
+def voxmlx_backend_available():
+    """Return True if voxmlx (Voxtral MLX backend) is available."""
+    is_macos = platform.system() == "Darwin"
+    is_arm = platform.machine() == "arm64"
+    return is_macos and is_arm and module_available("voxmlx")
+
+
 def faster_backend_available(warn_on_missing = False):
     available = module_available("faster_whisper")
     if not available and warn_on_missing and platform.system() != "Darwin":

whisperlivekit/core.py

@@ -1,5 +1,6 @@
 import logging
 import sys
+import threading
 from argparse import Namespace
 
 from whisperlivekit.local_agreement.online_asr import OnlineASRProcessor
@@ -19,16 +20,26 @@ logger = logging.getLogger(__name__)
 class TranscriptionEngine:
     _instance = None
     _initialized = False
+    _lock = threading.Lock()  # Thread-safe singleton lock
     
     def __new__(cls, *args, **kwargs):
+        # Double-checked locking pattern for thread-safe singleton
         if cls._instance is None:
-            cls._instance = super().__new__(cls)
+            with cls._lock:
+                # Check again inside lock to prevent race condition
+                if cls._instance is None:
+                    cls._instance = super().__new__(cls)
         return cls._instance
     
     def __init__(self, **kwargs):
-        if TranscriptionEngine._initialized:
+        # Thread-safe initialization check
-            return
+        with TranscriptionEngine._lock:
+            if TranscriptionEngine._initialized:
+                return
+            # Set flag immediately to prevent re-initialization
+            TranscriptionEngine._initialized = True
 
+        # Perform initialization outside lock to avoid holding lock during slow operations
         global_params = {
             "host": "localhost",
             "port": 8000,
@@ -36,7 +47,6 @@ class TranscriptionEngine:
             "punctuation_split": False,
             "target_language": "",
             "vac": True,
-            "vac_onnx": False,
             "vac_chunk_size": 0.04,
             "log_level": "DEBUG",
             "ssl_certfile": None,
@@ -79,18 +89,27 @@ class TranscriptionEngine:
         self.asr = None
         self.tokenizer = None
         self.diarization = None
-        self.vac_model = None
+        self.vac_session = None
         
         if self.args.vac:
-            from whisperlivekit.silero_vad_iterator import load_silero_vad
+            from whisperlivekit.silero_vad_iterator import is_onnx_available
-
-            # Use ONNX if specified, otherwise use JIT (default)
-            use_onnx = kwargs.get('vac_onnx', False)
-            self.vac_model = load_silero_vad(onnx=use_onnx)
             
+            if is_onnx_available():
+                from whisperlivekit.silero_vad_iterator import load_onnx_session
+                self.vac_session = load_onnx_session()
+            else:
+                logger.warning(
+                    "onnxruntime not installed. VAC will use JIT model which is loaded per-session. "
+                    "For multi-user scenarios, install onnxruntime: pip install onnxruntime"
+                )
         backend_policy = self.args.backend_policy
         if self.args.transcription:
-            if backend_policy == "simulstreaming":                 
+            if self.args.backend == "voxtral-mlx":
+                from whisperlivekit.voxtral_streaming import VoxtralStreamingASR
+                self.tokenizer = None
+                self.asr = VoxtralStreamingASR(**transcription_common_params)
+                logger.info("Using Voxtral MLX streaming backend")
+            elif backend_policy == "simulstreaming":
                 simulstreaming_params = {
                     "disable_fast_encoder": False,
                     "custom_alignment_heads": None,
@@ -169,16 +188,16 @@ class TranscriptionEngine:
                 }
                 translation_params = update_with_kwargs(translation_params, kwargs)
                 self.translation_model = load_model([self.args.lan], **translation_params) #in the future we want to handle different languages for different speakers
-        TranscriptionEngine._initialized = True
 
 
 def online_factory(args, asr):
+    if getattr(args, 'backend', None) == "voxtral-mlx":
+        from whisperlivekit.voxtral_streaming import VoxtralStreamingOnlineProcessor
+        return VoxtralStreamingOnlineProcessor(asr)
     if args.backend_policy == "simulstreaming":
         from whisperlivekit.simul_whisper import SimulStreamingOnlineProcessor
-        online = SimulStreamingOnlineProcessor(asr)
+        return SimulStreamingOnlineProcessor(asr)
-    else:
+    return OnlineASRProcessor(asr)
-        online = OnlineASRProcessor(asr)
-    return online
   
   
 def online_diarization_factory(args, diarization_backend):

whisperlivekit/diarization/diart_backend.py

@@ -202,14 +202,14 @@ class DiartDiarization:
     def insert_silence(self, silence_duration):
         self.observer.global_time_offset += silence_duration
 
-    async def diarize(self, pcm_array: np.ndarray):
+    def insert_audio_chunk(self, pcm_array: np.ndarray):
-        """
+        """Buffer audio for the next diarization step."""
-        Process audio data for diarization.
-        Only used when working with WebSocketAudioSource.
-        """
         if self.custom_source:
             self.custom_source.push_audio(pcm_array)
-        # self.observer.clear_old_segments()        
+
+    async def diarize(self):
+        """Return the current speaker segments from the diarization pipeline."""
+        return self.observer.get_segments()        
 
     def close(self):
         """Close the audio source."""

whisperlivekit/local_agreement/backends.py

@@ -249,6 +249,7 @@ class OpenaiApiASR(ASRBase):
         self.load_model()
         self.use_vad_opt = False
         self.direct_english_translation = False
+        self.task = "transcribe"
 
     def load_model(self, *args, **kwargs):
         from openai import OpenAI
@@ -294,7 +295,8 @@ class OpenaiApiASR(ASRBase):
             params["language"] = self.original_language
         if prompt:
             params["prompt"] = prompt
-        proc = self.client.audio.translations if self.task == "translate" else self.client.audio.transcriptions
+        task = self.transcribe_kargs.get("task", self.task)
+        proc = self.client.audio.translations if task == "translate" else self.client.audio.transcriptions
         transcript = proc.create(**params)
         logger.debug(f"OpenAI API processed accumulated {self.transcribed_seconds} seconds")
         return transcript

whisperlivekit/local_agreement/whisper_online.py

@@ -146,6 +146,7 @@ def backend_factory(
 
     if direct_english_translation:
         tgt_language = "en"  # Whisper translates into English
+        asr.transcribe_kargs["task"] = "translate"
     else:
         tgt_language = lan  # Whisper transcribes in this language
 

whisperlivekit/parse_args.py

@@ -147,8 +147,8 @@ def parse_args():
         "--backend",
         type=str,
         default="auto",
-        choices=["auto", "mlx-whisper", "faster-whisper", "whisper", "openai-api"],
+        choices=["auto", "mlx-whisper", "faster-whisper", "whisper", "openai-api", "voxtral-mlx"],
-        help="Select the Whisper backend implementation (auto: prefer MLX on macOS, otherwise Faster-Whisper, else Whisper). Use 'openai-api' with --backend-policy localagreement to call OpenAI's API.",
+        help="Select the Whisper backend implementation (auto: prefer MLX on macOS, otherwise Faster-Whisper, else Whisper). Use 'openai-api' with --backend-policy localagreement to call OpenAI's API. Use 'voxtral-mlx' for Voxtral streaming on Apple Silicon.",
     )
     parser.add_argument(
         "--no-vac",

whisperlivekit/silero_vad_iterator.py

@@ -8,6 +8,15 @@ import torch
 Code is adapted from silero-vad v6: https://github.com/snakers4/silero-vad
 """
 
+def is_onnx_available() -> bool:
+    """Check if onnxruntime is installed."""
+    try:
+        import onnxruntime
+        return True
+    except ImportError:
+        return False
+
+
 def init_jit_model(model_path: str, device=torch.device('cpu')):
     """Load a JIT model from file."""
     model = torch.jit.load(model_path, map_location=device)
@@ -15,12 +24,12 @@ def init_jit_model(model_path: str, device=torch.device('cpu')):
     return model
 
 
-class OnnxWrapper():
+class OnnxSession():
-    """ONNX Runtime wrapper for Silero VAD model."""
+    """
+    Shared ONNX session for Silero VAD model (stateless).
+    """
 
     def __init__(self, path, force_onnx_cpu=False):
-        global np
-        import numpy as np
         import onnxruntime
 
         opts = onnxruntime.SessionOptions()
@@ -32,13 +41,28 @@ class OnnxWrapper():
         else:
             self.session = onnxruntime.InferenceSession(path, sess_options=opts)
 
-        self.reset_states()
+        self.path = path
         if '16k' in path:
             warnings.warn('This model support only 16000 sampling rate!')
             self.sample_rates = [16000]
         else:
             self.sample_rates = [8000, 16000]
 
+
+class OnnxWrapper():
+    """
+    ONNX Runtime wrapper for Silero VAD model with per-instance state.
+    """
+
+    def __init__(self, session: OnnxSession, force_onnx_cpu=False):
+        self._shared_session = session
+        self.sample_rates = session.sample_rates
+        self.reset_states()
+
+    @property
+    def session(self):
+        return self._shared_session.session
+
     def _validate_input(self, x, sr: int):
         if x.dim() == 1:
             x = x.unsqueeze(0)
@@ -101,38 +125,20 @@ class OnnxWrapper():
         return out
 
 
-def load_silero_vad(model_path: str = None, onnx: bool = False, opset_version: int = 16):
+def _get_onnx_model_path(model_path: str = None, opset_version: int = 16) -> Path:
-    """
+    """Get the path to the ONNX model file."""
-    Load Silero VAD model (JIT or ONNX).
-    
-    Parameters
-    ----------
-    model_path : str, optional
-        Path to model file. If None, uses default bundled model.
-    onnx : bool, default False
-        Whether to use ONNX runtime (requires onnxruntime package).
-    opset_version : int, default 16
-        ONNX opset version (15 or 16). Only used if onnx=True.
-    
-    Returns
-    -------
-    model
-        Loaded VAD model (JIT or ONNX wrapper)
-    """
     available_ops = [15, 16]
-    if onnx and opset_version not in available_ops:
+    if opset_version not in available_ops:
         raise Exception(f'Available ONNX opset_version: {available_ops}')
+    
     if model_path is None:
         current_dir = Path(__file__).parent
         data_dir = current_dir / 'silero_vad_models'
         
-        if onnx:
+        if opset_version == 16:
-            if opset_version == 16:
+            model_name = 'silero_vad.onnx'
-                model_name = 'silero_vad.onnx'
-            else:
-                model_name = f'silero_vad_16k_op{opset_version}.onnx'
         else:
-            model_name = 'silero_vad.jit'
+            model_name = f'silero_vad_16k_op{opset_version}.onnx'
         
         model_path = data_dir / model_name
         
@@ -143,16 +149,38 @@ def load_silero_vad(model_path: str = None, onnx: bool = False, opset_version: i
             )
     else:
         model_path = Path(model_path)
-    if onnx:
+    
-        try:
+    return model_path
-            model = OnnxWrapper(str(model_path), force_onnx_cpu=True)
+
-        except ImportError:
+
-            raise ImportError(
+def load_onnx_session(model_path: str = None, opset_version: int = 16, force_onnx_cpu: bool = True) -> OnnxSession:
-                "ONNX runtime not available. Install with: pip install onnxruntime\n"
+    """
-                "Or use JIT model by setting onnx=False"
+    Load a shared ONNX session for Silero VAD.
+    """
+    path = _get_onnx_model_path(model_path, opset_version)
+    return OnnxSession(str(path), force_onnx_cpu=force_onnx_cpu)
+
+
+def load_jit_vad(model_path: str = None):
+    """
+    Load Silero VAD model in JIT format.
+    """
+    if model_path is None:
+        current_dir = Path(__file__).parent
+        data_dir = current_dir / 'silero_vad_models'
+        model_name = 'silero_vad.jit'
+        
+        model_path = data_dir / model_name
+        
+        if not model_path.exists():
+            raise FileNotFoundError(
+                f"Model file not found: {model_path}\n"
+                f"Please ensure the whisperlivekit/silero_vad_models/ directory contains the model files."
             )
     else:
-        model = init_jit_model(str(model_path))
+        model_path = Path(model_path)
+
+    model = init_jit_model(str(model_path))
  
     return model
 
@@ -285,8 +313,8 @@ class FixedVADIterator(VADIterator):
 
 
 if __name__ == "__main__":
-    model = load_silero_vad(onnx=False)
+    # vad = FixedVADIterator(load_jit_vad())
-    vad = FixedVADIterator(model)
+    vad = FixedVADIterator(OnnxWrapper(session=load_onnx_session()))
 
     audio_buffer = np.array([0] * 512, dtype=np.float32)
     result = vad(audio_buffer)
@@ -295,3 +323,4 @@ if __name__ == "__main__":
     # test with 511 samples
     audio_buffer = np.array([0] * 511, dtype=np.float32)
     result = vad(audio_buffer)
+    print(f"   511 samples: {result}")

whisperlivekit/simul_whisper/backend.py

@@ -24,9 +24,11 @@ logger = logging.getLogger(__name__)
 
 HAS_MLX_WHISPER = mlx_backend_available(warn_on_missing=True)
 if HAS_MLX_WHISPER:
-    from .mlx_encoder import load_mlx_encoder, mlx_model_mapping
+    from .mlx_encoder import load_mlx_encoder, load_mlx_model, mlx_model_mapping
+    from .mlx import MLXAlignAtt
 else:
     mlx_model_mapping = {}
+    MLXAlignAtt = None
 HAS_FASTER_WHISPER = faster_backend_available(warn_on_missing=not HAS_MLX_WHISPER)
 if HAS_FASTER_WHISPER:
     from faster_whisper import WhisperModel
@@ -36,50 +38,47 @@ else:
 MIN_DURATION_REAL_SILENCE = 5
 
 class SimulStreamingOnlineProcessor:
+    """Online processor for SimulStreaming ASR."""
     SAMPLING_RATE = 16000
 
-    def __init__(
+    def __init__(self, asr, logfile=sys.stderr):
-        self,
-        asr,
-        logfile=sys.stderr,
-    ):        
         self.asr = asr
         self.logfile = logfile
         self.end = 0.0
         self.buffer = []
-        self.committed: List[ASRToken] = []
+        self.model = self._create_alignatt()
-        self.last_result_tokens: List[ASRToken] = []
-        self.load_new_alignatt_instance()
         
         if asr.tokenizer:
             self.model.tokenizer = asr.tokenizer
+            self.model.state.tokenizer = asr.tokenizer
 
-    def load_new_alignatt_instance(self):
+    def _create_alignatt(self):
-        """Initialize AlignAtt decoder using the shared model."""
+        """Create the AlignAtt decoder instance based on ASR mode."""
-        self.model = AlignAtt(
+        if self.asr.use_full_mlx and HAS_MLX_WHISPER:
-            cfg=self.asr.cfg,
+            return MLXAlignAtt(cfg=self.asr.cfg, mlx_model=self.asr.mlx_model)
-            loaded_model=self.asr.shared_model,
+        else:
-            mlx_encoder=self.asr.mlx_encoder,
+            return AlignAtt(
-            fw_encoder=self.asr.fw_encoder,
+                cfg=self.asr.cfg,
-        )
+                loaded_model=self.asr.shared_model,
+                mlx_encoder=self.asr.mlx_encoder,
+                fw_encoder=self.asr.fw_encoder,
+            )
 
     def start_silence(self):
         tokens, processed_upto = self.process_iter(is_last=True)
         return tokens, processed_upto
 
     def end_silence(self, silence_duration, offset):
-        """
+        """Handle silence period."""
-        Handle silence period.
-        
-        If silence > MIN_DURATION_REAL_SILENCE, do a complete context clear.
-        Otherwise, insert a small silence and shift the last_attend_frame.
-        """
         self.end += silence_duration
         long_silence = silence_duration >= MIN_DURATION_REAL_SILENCE
         if not long_silence:
             gap_len = int(16000 * silence_duration)
             if gap_len > 0:
-                gap_silence = torch.zeros(gap_len)
+                if self.asr.use_full_mlx:
+                    gap_silence = np.zeros(gap_len, dtype=np.float32)
+                else:
+                    gap_silence = torch.zeros(gap_len)
                 self.model.insert_audio(gap_silence)
         if long_silence:
             self.model.refresh_segment(complete=True)
@@ -87,11 +86,12 @@ class SimulStreamingOnlineProcessor:
 
     def insert_audio_chunk(self, audio: np.ndarray, audio_stream_end_time):
         """Append an audio chunk to be processed by SimulStreaming."""
-            
+        self.end = audio_stream_end_time
-        # Convert numpy array to torch tensor
+        if self.asr.use_full_mlx:
-        audio_tensor = torch.from_numpy(audio).float()
+            self.model.insert_audio(audio)
-        self.end = audio_stream_end_time  # Aligned with whisperstreaming backend behavior
+        else:
-        self.model.insert_audio(audio_tensor)
+            audio_tensor = torch.from_numpy(audio).float()
+            self.model.insert_audio(audio_tensor)
 
     def new_speaker(self, change_speaker: ChangeSpeaker):
         """Handle speaker change event."""
@@ -120,7 +120,6 @@ class SimulStreamingOnlineProcessor:
                 self.buffer.extend(timestamped_words)
                 return [], self.end
             
-            self.committed.extend(timestamped_words)
             self.buffer = []
             return timestamped_words, self.end
         except Exception as e:
@@ -130,6 +129,10 @@ class SimulStreamingOnlineProcessor:
     def warmup(self, audio, init_prompt=""):
         """Warmup the SimulStreaming model."""
         try:
+            if self.asr.use_full_mlx:
+                # MLX mode: ensure numpy array
+                if hasattr(audio, 'numpy'):
+                    audio = audio.numpy()
             self.model.insert_audio(audio)
             self.model.infer(True)
             self.model.refresh_segment(complete=True)
@@ -139,9 +142,14 @@ class SimulStreamingOnlineProcessor:
 
     def __del__(self):
         gc.collect()
-        torch.cuda.empty_cache()
+        if not getattr(self.asr, 'use_full_mlx', True) and torch is not None:
+            try:
+                torch.cuda.empty_cache()
+            except Exception:
+                pass
 
-class SimulStreamingASR():
+
+class SimulStreamingASR:
     """SimulStreaming backend with AlignAtt policy."""
     sep = ""
 
@@ -158,6 +166,7 @@ class SimulStreamingASR():
         self.fast_encoder = False
         self._resolved_model_path = None
         self.encoder_backend = "whisper"
+        self.use_full_mlx = getattr(self, "use_full_mlx", False)
         preferred_backend = getattr(self, "backend", "auto")
         compatible_whisper_mlx, compatible_faster_whisper = True, True
         
@@ -170,7 +179,7 @@ class SimulStreamingASR():
             compatible_whisper_mlx = model_info.compatible_whisper_mlx
             compatible_faster_whisper = model_info.compatible_faster_whisper
             
-            if not model_info.has_pytorch:
+            if not self.use_full_mlx and not model_info.has_pytorch:
                 raise FileNotFoundError(
                     f"No PyTorch checkpoint (.pt/.bin/.safetensors) found under {self.model_path}"
                 )            
@@ -191,6 +200,10 @@ class SimulStreamingASR():
         if self.encoder_backend == "whisper":
             self.disable_fast_encoder = True
         
+        if self.encoder_backend == "mlx-whisper" and platform.system() == "Darwin":
+            if not hasattr(self, '_full_mlx_disabled'):
+                self.use_full_mlx = True
+                    
         self.cfg = AlignAttConfig(
                 tokenizer_is_multilingual= is_multilingual,
                 segment_length=self.min_chunk_size,
@@ -201,7 +214,7 @@ class SimulStreamingASR():
                 cif_ckpt_path=self.cif_ckpt_path,
                 decoder_type="beam",
                 beam_size=self.beams,
-                task=self.direct_english_translation,
+                task="translate" if self.direct_english_translation else "transcribe",
                 never_fire=self.never_fire,
                 init_prompt=self.init_prompt,
                 max_context_tokens=self.max_context_tokens,
@@ -214,20 +227,36 @@ class SimulStreamingASR():
         else:
             self.tokenizer = None
 
-        self.mlx_encoder, self.fw_encoder = None, None
+        self.mlx_encoder, self.fw_encoder, self.mlx_model = None, None, None
-        if self.encoder_backend == "mlx-whisper":
+        self.shared_model = None
-            print('Simulstreaming will use MLX whisper to increase encoding speed.')
+        
+        if self.use_full_mlx and HAS_MLX_WHISPER:
+            logger.info('MLX Whisper backend used.')
             if self._resolved_model_path is not None:
-                mlx_model = str(self._resolved_model_path)
+                mlx_model_path = str(self._resolved_model_path)
             else:
-                mlx_model = mlx_model_mapping.get(self.model_name)
+                mlx_model_path = mlx_model_mapping.get(self.model_name)
-            if not mlx_model:
+            if not mlx_model_path:
                 raise FileNotFoundError(
                     f"MLX Whisper backend requested but no compatible weights found for model '{self.model_name}'."
                 )
-            self.mlx_encoder = load_mlx_encoder(path_or_hf_repo=mlx_model)
+            self.mlx_model = load_mlx_model(path_or_hf_repo=mlx_model_path)
+            self._warmup_mlx_model()
+        elif self.encoder_backend == "mlx-whisper":
+            # hybrid mode: mlx encoder + pytorch decoder
+            logger.info('SimulStreaming will use MLX Whisper encoder with PyTorch decoder.')
+            if self._resolved_model_path is not None:
+                mlx_model_path = str(self._resolved_model_path)
+            else:
+                mlx_model_path = mlx_model_mapping.get(self.model_name)
+            if not mlx_model_path:
+                raise FileNotFoundError(
+                    f"MLX Whisper backend requested but no compatible weights found for model '{self.model_name}'."
+                )
+            self.mlx_encoder = load_mlx_encoder(path_or_hf_repo=mlx_model_path)
+            self.shared_model = self.load_model()
         elif self.encoder_backend == "faster-whisper":
-            print('Simulstreaming will use Faster Whisper for the encoder.')
+            print('SimulStreaming will use Faster Whisper for the encoder.')
             if self._resolved_model_path is not None:
                 fw_model = str(self._resolved_model_path)
             else:
@@ -237,7 +266,20 @@ class SimulStreamingASR():
                 device='auto',
                 compute_type='auto',
             )
-        self.shared_model = self.load_model()
+            self.shared_model = self.load_model()
+        else:
+            self.shared_model = self.load_model()
+    
+    def _warmup_mlx_model(self):
+        """Warmup the full MLX model."""
+        warmup_audio = load_file(self.warmup_file)
+        if warmup_audio is not None:
+            temp_model = MLXAlignAtt(
+                cfg=self.cfg,
+                mlx_model=self.mlx_model,
+            )
+            temp_model.warmup(warmup_audio)
+            logger.info("Full MLX model warmed up successfully")
 
 
     def _resolve_encoder_backend(self, preferred_backend, compatible_whisper_mlx, compatible_faster_whisper):
@@ -285,7 +327,7 @@ class SimulStreamingASR():
         lora_path = getattr(self, 'lora_path', None)
         whisper_model = load_model(
             name=model_ref,
-            download_root=None,
+            download_root=getattr(self, 'model_cache_dir', None),
             decoder_only=self.fast_encoder,
             custom_alignment_heads=self.custom_alignment_heads,
             lora_path=lora_path,

whisperlivekit/simul_whisper/decoder_state.py

@@ -47,9 +47,24 @@ class DecoderState:
     
     def clean_cache(self):
         """Clean the kv_cache after each inference step."""
-        self.kv_cache = {}
+        # Explicitly delete tensor references to free GPU memory
+        if self.kv_cache:
+            for key in list(self.kv_cache.keys()):
+                tensor = self.kv_cache.pop(key, None)
+                if tensor is not None:
+                    del tensor
+
+        # Clear the dict
+        self.kv_cache.clear()
+
+        # Force GPU cache cleanup (only if CUDA is available)
+        import torch
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
         if self.decoder_type == "beam" and self.inference is not None:
-            self.inference.kv_cache = self.kv_cache
+            # Create NEW dict instead of sharing reference
+            self.inference.kv_cache = {}
             if self.token_decoder is not None:
                 self.token_decoder.reset()
     

whisperlivekit/simul_whisper/mlx/__init__.py

@@ -0,0 +1,11 @@
+from .decoder_state import MLXDecoderState
+from .decoders import MLXBeamSearchDecoder, MLXGreedyDecoder, MLXInference
+from .simul_whisper import MLXAlignAtt
+
+__all__ = [
+    "MLXAlignAtt",
+    "MLXBeamSearchDecoder",
+    "MLXDecoderState",
+    "MLXGreedyDecoder",
+    "MLXInference",
+]

whisperlivekit/simul_whisper/mlx/decoder_state.py

@@ -0,0 +1,76 @@
+from dataclasses import dataclass, field
+from typing import Any, Dict, List, Optional, Tuple
+
+import mlx.core as mx
+import numpy as np
+
+
+@dataclass
+class MLXDecoderState:
+    """
+    mlx kv cache format: List of ((k, v), (cross_k, cross_v)) tuples per layer,
+    where each element is a tuple of mx.arrays.
+    """
+
+    kv_cache: Optional[List[Tuple[Tuple[mx.array, mx.array], Tuple[mx.array, mx.array]]]] = None
+    
+    tokenizer: Any = None
+    detected_language: Optional[str] = None
+    reset_tokenizer_to_auto_next_call: bool = False
+    
+    tokens: List[mx.array] = field(default_factory=list)
+    initial_tokens: Optional[mx.array] = None
+    initial_token_length: int = 0
+    sot_index: int = 0    
+    align_source: Dict[int, List[Tuple[int, int]]] = field(default_factory=dict)
+    num_align_heads: int = 0    
+    segments: List[np.ndarray] = field(default_factory=list)
+    
+    context: Any = None
+    
+    pending_incomplete_tokens: List[int] = field(default_factory=list)
+    
+    global_time_offset: float = 0.0
+    cumulative_time_offset: float = 0.0
+    first_timestamp: Optional[float] = None
+    last_attend_frame: int = 0
+    
+    speaker: int = -1
+    log_segments: int = 0    
+    cif_weights: Optional[mx.array] = None
+    always_fire: bool = False
+    never_fire: bool = False
+    
+    suppress_tokens: Optional[Tuple[int, ...]] = None
+    
+    token_decoder: Any = None
+    decoder_type: str = "greedy"
+    
+    inference: Any = None
+    
+    def clean_cache(self):
+        self.kv_cache = None
+        if self.decoder_type == "beam" and self.inference is not None:
+            self.inference.kv_cache = None
+            if self.token_decoder is not None:
+                self.token_decoder.reset()
+    
+    def reset(self, rewind_threshold: int = 200):
+        self.last_attend_frame = -rewind_threshold
+        self.cumulative_time_offset = 0.0
+        self.pending_incomplete_tokens = []
+        self.log_segments += 1
+    
+    def full_reset(self, rewind_threshold: int = 200):
+        """
+        Full reset including audio segments and tokens.
+        
+        Args:
+            rewind_threshold: Value for resetting last_attend_frame
+        """
+        self.reset(rewind_threshold)
+        self.segments = []
+        self.tokens = []
+        self.kv_cache = None
+        self.first_timestamp = None
+

whisperlivekit/simul_whisper/mlx/decoders.py

@@ -0,0 +1,219 @@
+"""
+MLX-native token decoders for streaming ASR.
+"""
+from typing import Any, Dict, List, Optional, Tuple
+
+import mlx.core as mx
+import numpy as np
+
+
+class MLXGreedyDecoder:
+    """Greedy decoder using MLX operations."""
+    
+    def __init__(self, temperature: float, eot: int):
+        self.temperature = temperature
+        self.eot = eot
+
+    def update(
+        self, tokens: mx.array, logits: mx.array, sum_logprobs: mx.array
+    ) -> Tuple[mx.array, bool]:
+        """
+        Update tokens with next predicted token.
+        
+        Args:
+            tokens: Current token sequence, shape (batch, seq_len)
+            logits: Logits for next token, shape (batch, vocab_size)
+            sum_logprobs: Cumulative log probabilities, shape (batch,)
+            
+        Returns:
+            Updated tokens and completion flag
+        """
+        if self.temperature == 0:
+            next_tokens = mx.argmax(logits, axis=-1)
+        else:
+            probs = mx.softmax(logits / self.temperature, axis=-1)
+            next_tokens = mx.random.categorical(mx.log(probs + 1e-10))
+        
+        logprobs = mx.softmax(logits, axis=-1)
+        logprobs = mx.log(logprobs + 1e-10)        
+        batch_size = logprobs.shape[0]
+        current_logprobs = logprobs[mx.arange(batch_size), next_tokens]        
+        mask = (tokens[:, -1] != self.eot).astype(mx.float32)
+        sum_logprobs = sum_logprobs + current_logprobs * mask        
+        eot_mask = (tokens[:, -1] == self.eot)
+        next_tokens = mx.where(eot_mask, mx.array(self.eot), next_tokens)        
+        tokens = mx.concatenate([tokens, next_tokens[:, None]], axis=1)        
+        completed = bool(mx.all(tokens[:, -1] == self.eot))
+        
+        return tokens, completed
+
+    def finalize(self, tokens: mx.array, sum_logprobs: mx.array):
+        """Finalize decoding by ensuring EOT at end."""
+        eot_column = mx.full((tokens.shape[0], 1), self.eot, dtype=tokens.dtype)
+        tokens = mx.concatenate([tokens, eot_column], axis=1)
+        return tokens, sum_logprobs.tolist()
+
+
+class MLXBeamSearchDecoder:
+    """Beam search decoder using MLX operations."""
+    
+    def __init__(
+        self,
+        beam_size: int,
+        eot: int,
+        inference: Any,
+        patience: Optional[float] = None,
+    ):
+        self.beam_size = beam_size
+        self.eot = eot
+        self.inference = inference
+        self.patience = patience or 1.0
+        self.max_candidates: int = round(beam_size * self.patience)
+        self.finished_sequences: Optional[List[Dict]] = None
+
+        assert (
+            self.max_candidates > 0
+        ), f"Invalid beam size ({beam_size}) or patience ({patience})"
+
+    def reset(self):
+        """Reset finished sequences for new segment."""
+        self.finished_sequences = None
+
+    def update(
+        self, tokens: mx.array, logits: mx.array, sum_logprobs: mx.array
+    ) -> Tuple[mx.array, bool]:
+        """
+        Update tokens using beam search.
+        
+        Args:
+            tokens: Current token sequences, shape (batch * beam_size, seq_len)
+            logits: Logits for next token, shape (batch * beam_size, vocab_size)
+            sum_logprobs: Cumulative log probabilities, shape (batch * beam_size,)
+            
+        Returns:
+            Updated tokens and completion flag
+        """
+        if tokens.shape[0] % self.beam_size != 0:
+            raise ValueError(f"{tokens.shape}[0] % {self.beam_size} != 0")
+
+        n_audio = tokens.shape[0] // self.beam_size
+        if self.finished_sequences is None:
+            self.finished_sequences = [{} for _ in range(n_audio)]
+        logprobs = mx.softmax(logits, axis=-1)
+        logprobs = mx.log(logprobs + 1e-10)        
+        logprobs_np = np.array(logprobs)
+        tokens_np = np.array(tokens)
+        sum_logprobs_np = np.array(sum_logprobs)
+        
+        next_tokens, source_indices, finished_sequences = [], [], []
+        new_sum_logprobs = []
+        
+        for i in range(n_audio):
+            scores, sources, finished = {}, {}, {}
+            for j in range(self.beam_size):
+                idx = i * self.beam_size + j
+                prefix = tokens_np[idx].tolist()                
+                top_k_indices = np.argsort(logprobs_np[idx])[-self.beam_size - 1:][::-1]
+                
+                for token_idx in top_k_indices:
+                    logprob = logprobs_np[idx, token_idx]
+                    new_logprob = sum_logprobs_np[idx] + logprob
+                    sequence = tuple(prefix + [int(token_idx)])
+                    scores[sequence] = new_logprob
+                    sources[sequence] = idx
+            saved = 0
+            for sequence in sorted(scores, key=scores.get, reverse=True):
+                if sequence[-1] == self.eot:
+                    finished[sequence] = scores[sequence]
+                else:
+                    new_sum_logprobs.append(scores[sequence])
+                    next_tokens.append(sequence)
+                    source_indices.append(sources[sequence])
+
+                    saved += 1
+                    if saved == self.beam_size:
+                        break
+
+            finished_sequences.append(finished)
+        tokens = mx.array(np.array(next_tokens, dtype=np.int32))
+        sum_logprobs = mx.array(np.array(new_sum_logprobs, dtype=np.float32))        
+        self.inference.rearrange_kv_cache(source_indices)
+        assert len(self.finished_sequences) == len(finished_sequences)
+        for previously_finished, newly_finished in zip(
+            self.finished_sequences, finished_sequences
+        ):
+            for seq in sorted(newly_finished, key=newly_finished.get, reverse=True):
+                if len(previously_finished) >= self.max_candidates:
+                    break
+                previously_finished[seq] = newly_finished[seq]
+        completed = all(
+            len(sequences) >= self.max_candidates
+            for sequences in self.finished_sequences
+        )
+        
+        return tokens, completed
+
+    def finalize(self, preceding_tokens: mx.array, sum_logprobs: mx.array):
+        """Finalize beam search by selecting best sequences."""
+        preceding_tokens_np = np.array(preceding_tokens)
+        sum_logprobs_np = np.array(sum_logprobs)
+        
+        n_audio = preceding_tokens_np.shape[0] // self.beam_size
+        tokens_list: List[List[int]] = [[] for _ in range(n_audio)]
+        sum_logprobs_list: List[float] = [0.0] * n_audio
+
+        for i, sequences in enumerate(self.finished_sequences):
+            if sequences:
+                best_seq = max(sequences, key=sequences.get)
+                tokens_list[i] = list(best_seq)
+                sum_logprobs_list[i] = sequences[best_seq]
+            else:
+                idx = i * self.beam_size
+                tokens_list[i] = preceding_tokens_np[idx].tolist() + [self.eot]
+                sum_logprobs_list[i] = float(sum_logprobs_np[idx])
+        max_len = max(len(t) for t in tokens_list)
+        for i, t in enumerate(tokens_list):
+            tokens_list[i] = t + [self.eot] * (max_len - len(t))
+
+        tokens = mx.array(np.array(tokens_list, dtype=np.int32))
+        return tokens, sum_logprobs_list
+
+
+class MLXInference:
+    """MLX inference wrapper for beam search KV cache management."""
+    
+    def __init__(self, model, initial_token_length: int):
+        self.model = model
+        self.initial_token_length = initial_token_length
+        self.kv_cache = None
+    
+    def rearrange_kv_cache(self, source_indices: List[int]):
+        """Rearrange KV cache based on beam search source indices."""
+        if self.kv_cache is None:
+            return
+            
+        if source_indices == list(range(len(source_indices))):
+            return
+        
+        source_indices_mx = mx.array(source_indices, dtype=mx.int32)
+        
+        new_cache = []
+        for layer_cache in self.kv_cache:
+            (k, v), (cross_k, cross_v) = layer_cache            
+            new_k = k[source_indices_mx]
+            new_v = v[source_indices_mx]
+            new_cache.append(((new_k, new_v), (cross_k, cross_v)))
+        
+        self.kv_cache = new_cache
+    
+    def logits(
+        self, 
+        tokens: mx.array, 
+        audio_features: mx.array,
+    ) -> Tuple[mx.array, List]:
+        """Get logits from decoder with KV cache."""
+        logits, self.kv_cache, cross_qk = self.model.decoder(
+            tokens, audio_features, kv_cache=self.kv_cache
+        )
+        return logits, cross_qk
+

whisperlivekit/simul_whisper/mlx/simul_whisper.py

@@ -0,0 +1,756 @@
+"""
+MLX whisper AlignAtt streaming decoder
+"""
+import logging
+from time import time
+from typing import Any, List, Optional, Tuple
+
+import mlx.core as mx
+import numpy as np
+
+from mlx_whisper.audio import log_mel_spectrogram as mlx_log_mel_spectrogram
+from mlx_whisper.transcribe import pad_or_trim as mlx_pad_or_trim
+
+from whisperlivekit.timed_objects import ASRToken
+from whisperlivekit.whisper import DecodingOptions, tokenizer
+from whisperlivekit.whisper.audio import N_FRAMES, N_SAMPLES, TOKENS_PER_SECOND
+
+from ..config import AlignAttConfig
+from .decoder_state import MLXDecoderState
+from .decoders import MLXBeamSearchDecoder, MLXGreedyDecoder, MLXInference
+
+DEC_PAD = 50257
+logger = logging.getLogger(__name__)
+
+
+class MLXTokenBuffer: #should try to make it heritate from classic simul whisper class
+    """Token buffer for MLX-based decoding."""
+
+    def __init__(self, text="", tokenizer=None, prefix_token_ids=None):
+        self.text = text
+        self.prefix_token_ids = prefix_token_ids or []
+        self.tokenizer = tokenizer
+        self.pending_token_ids = []
+
+    def as_token_ids(self, tokenizer=None):
+        if tokenizer is None:
+            tokenizer = self.tokenizer
+        if tokenizer is None:
+            raise ValueError("Tokenizer is not set.")
+        return self.prefix_token_ids + tokenizer.encode(self.text)
+
+    def as_mlx_array(self) -> mx.array:
+        """Return tokens as MLX array."""
+        tok_ids = self.as_token_ids()
+        return mx.array([tok_ids], dtype=mx.int32)
+
+    def as_mlx_array_beam(self, beam: int) -> mx.array:
+        """Return tokens as MLX array repeated for beam search."""
+        t = self.as_mlx_array()
+        return mx.repeat(t, beam, axis=0)
+
+    def as_text(self):
+        return self.text
+
+    @staticmethod
+    def empty(*a, **kw):
+        return MLXTokenBuffer(*a, **kw)
+
+    @staticmethod
+    def from_text(text, *a, **kw):
+        return MLXTokenBuffer(*a, text=text, **kw)
+
+    def is_empty(self):
+        return self.text is None or self.text == ""
+
+    def trim_words(self, num=1, after=0):
+        """Trim words from the beginning of the context."""
+        tokenizer = self.tokenizer
+        assert tokenizer is not None, "Tokenizer is not set."
+
+        ids = tokenizer.encode(self.text[after:])
+        words, wids = self.tokenizer.split_to_word_tokens(ids)
+        if not words:
+            return 0
+        self.text = self.text[:after] + "".join(words[num:])
+        return sum(len(wi) for wi in wids[:num])
+
+    def append_token_ids(self, token_ids):
+        """Append token IDs to the buffer, handling incomplete UTF-8."""
+        tokenizer = self.tokenizer
+        assert tokenizer is not None, "Tokenizer is not set."
+
+        all_tokens = self.pending_token_ids + token_ids
+        decoded = tokenizer.decode(all_tokens)
+        replacement_char = "\ufffd"
+
+        if replacement_char in decoded:
+            if len(all_tokens) > 1:
+                decoded_partial = tokenizer.decode(all_tokens[:-1])
+                if replacement_char not in decoded_partial:
+                    self.text += decoded_partial
+                    self.pending_token_ids = [all_tokens[-1]]
+                else:
+                    self.pending_token_ids = all_tokens
+            else:
+                self.pending_token_ids = all_tokens
+        else:
+            self.text += decoded
+            self.pending_token_ids = []
+
+
+def mlx_median_filter(x: mx.array, filter_width: int) -> mx.array:
+    """
+    Apply median filter along the last axis.
+    
+    Args:
+        x: Input array of shape (..., T)
+        filter_width: Width of the median filter (should be odd)
+        
+    Returns:
+        Filtered array of same shape
+    """
+    if filter_width <= 1:
+        return x
+    
+    pad_width = filter_width // 2
+    shape = x.shape
+    
+    left_pad = mx.repeat(x[..., :1], pad_width, axis=-1)
+    right_pad = mx.repeat(x[..., -1:], pad_width, axis=-1)
+    x_padded = mx.concatenate([left_pad, x, right_pad], axis=-1)
+    
+    result_shape = list(shape)
+    result = []
+    
+    for i in range(shape[-1]):
+        window = x_padded[..., i:i + filter_width]
+        sorted_window = mx.sort(window, axis=-1)
+        median_val = sorted_window[..., filter_width // 2:filter_width // 2 + 1]
+        result.append(median_val)
+    
+    return mx.concatenate(result, axis=-1)
+
+
+class MLXAlignAtt:
+    """
+    MLX-native Alignment-based Attention decoder for SimulStreaming.
+    
+    This class runs entirely on MLX, with no PyTorch dependencies for inference.
+    """
+
+    @property
+    def speaker(self):
+        return self.state.speaker
+
+    @speaker.setter
+    def speaker(self, value):
+        self.state.speaker = value
+
+    @property
+    def global_time_offset(self):
+        return self.state.global_time_offset
+
+    @global_time_offset.setter
+    def global_time_offset(self, value):
+        self.state.global_time_offset = value
+
+    def __init__(
+        self,
+        cfg: AlignAttConfig,
+        mlx_model: Any,
+    ) -> None:
+        """
+        Initialize MLX AlignAtt decoder.
+        
+        Args:
+            cfg: AlignAtt configuration
+            mlx_model: MLX Whisper model (full model, not just encoder)
+        """
+        self.model = mlx_model
+        self.cfg = cfg
+        
+        logger.info(f"MLX Model dimensions: {self.model.dims}")
+        
+        self.decode_options = DecodingOptions(
+            language=cfg.language,
+            without_timestamps=True,
+            task=cfg.task
+        )
+        self.tokenizer_is_multilingual = cfg.tokenizer_is_multilingual
+        
+        self.max_text_len = self.model.dims.n_text_ctx
+        self.num_decoder_layers = len(self.model.decoder.blocks)
+
+        if self.cfg.max_context_tokens is None:
+            self.max_context_tokens = self.max_text_len
+        else:
+            self.max_context_tokens = self.cfg.max_context_tokens
+
+        # Initialize per-session state
+        self.state = MLXDecoderState()
+        self._init_state(cfg)
+
+    def _init_state(self, cfg: AlignAttConfig):
+        """Initialize the per-session decoder state."""
+        self.create_tokenizer(cfg.language if cfg.language != "auto" else None)
+        self.state.tokenizer = self.tokenizer
+        self.state.detected_language = cfg.language if cfg.language != "auto" else None
+        self.state.global_time_offset = 0.0
+        self.state.last_attend_frame = -cfg.rewind_threshold
+        self.state.speaker = -1
+
+        if cfg.cif_ckpt_path is None or not cfg.cif_ckpt_path:
+            if cfg.never_fire:
+                self.state.never_fire = True
+                self.state.always_fire = False
+            else:
+                self.state.always_fire = True
+                self.state.never_fire = False
+        else:
+            logger.warning("CIF checkpoint provided but MLX CIF not implemented. Using always_fire=True")
+            self.state.always_fire = True
+            self.state.never_fire = cfg.never_fire
+
+        self._build_alignment_source()
+
+        suppress_tokens = [
+            self.tokenizer.transcribe,
+            self.tokenizer.translate,
+            self.tokenizer.sot,
+            self.tokenizer.sot_prev,
+            self.tokenizer.sot_lm,
+            self.tokenizer.no_timestamps,
+        ] + list(self.tokenizer.all_language_tokens)
+        if self.tokenizer.no_speech is not None:
+            suppress_tokens.append(self.tokenizer.no_speech)
+        self.state.suppress_tokens = tuple(sorted(set(suppress_tokens)))
+        logger.debug(f"Suppress tokens: {self.state.suppress_tokens}")
+
+        self.init_tokens()
+        self.init_context()
+
+        self.state.decoder_type = cfg.decoder_type
+        if cfg.decoder_type == "greedy":
+            logger.info("Using MLX greedy decoder")
+            self.state.token_decoder = MLXGreedyDecoder(0.0, self.tokenizer.eot)
+        elif cfg.decoder_type == "beam":
+            logger.info("Using MLX beam decoder")
+            self.state.inference = MLXInference(self.model, self.state.initial_token_length)
+            self.state.token_decoder = MLXBeamSearchDecoder(
+                inference=self.state.inference,
+                eot=self.tokenizer.eot,
+                beam_size=cfg.beam_size
+            )
+
+    def _build_alignment_source(self):
+        """Build alignment source mapping from model's alignment_heads."""
+        self.state.align_source = {}
+        self.state.num_align_heads = 0
+        
+        alignment_heads = self.model.alignment_heads
+        
+        if alignment_heads is None:
+            logger.warning("No alignment heads found in model")
+            return
+            
+        if hasattr(alignment_heads, 'tolist'):
+            heads_list = alignment_heads.tolist()
+        else:
+            heads_list = np.array(alignment_heads).tolist()
+            
+        for layer_rank, head_id in heads_list:
+            layer_rank = int(layer_rank)
+            head_id = int(head_id)
+            heads = self.state.align_source.get(layer_rank, [])
+            heads.append((self.state.num_align_heads, head_id))
+            self.state.align_source[layer_rank] = heads
+            self.state.num_align_heads += 1
+
+    def warmup(self, audio: np.ndarray):
+        """Warmup the model with sample audio."""
+        try:
+            self.insert_audio(audio)
+            self.infer(is_last=True)
+            self.refresh_segment(complete=True)
+            logger.info("MLX model warmed up successfully")
+        except Exception as e:
+            logger.exception(f"MLX model warmup failed: {e}")
+
+    def create_tokenizer(self, language=None):
+        """Create tokenizer for the given language."""
+        self.tokenizer = tokenizer.get_tokenizer(
+            multilingual=self.tokenizer_is_multilingual,
+            language=language,
+            num_languages=self.model.num_languages,
+            task=self.decode_options.task
+        )
+        self.state.tokenizer = self.tokenizer
+
+    def init_context(self):
+        """Initialize context buffer."""
+        kw = {
+            'tokenizer': self.tokenizer,
+            'prefix_token_ids': [self.tokenizer.sot_prev]
+        }
+        self.state.context = MLXTokenBuffer.empty(**kw)
+        if self.cfg.static_init_prompt is not None:
+            self.state.context = MLXTokenBuffer.from_text(self.cfg.static_init_prompt, **kw)
+        if self.cfg.init_prompt is not None:
+            self.state.context.text += self.cfg.init_prompt
+
+    def init_tokens(self):
+        """Initialize token sequence."""
+        logger.debug(f"init tokens, {len(self.state.segments)}")
+        self.state.initial_tokens = mx.array(
+            [self.tokenizer.sot_sequence_including_notimestamps],
+            dtype=mx.int32
+        )
+        self.state.initial_token_length = self.state.initial_tokens.shape[1]
+        self.state.sot_index = self.tokenizer.sot_sequence.index(self.tokenizer.sot)
+        logger.debug(f"init tokens after, {len(self.state.segments)}")
+        self.state.tokens = [self.state.initial_tokens]
+
+    def trim_context(self):
+        """Trim context if too long."""
+        logger.info("Trimming context")
+        c = len(self.state.context.as_token_ids()) - len(self.state.context.prefix_token_ids)
+        logger.info(f"Context text: {self.state.context.as_text()}")
+        l = sum(t.shape[1] for t in self.state.tokens) + c
+        if self.cfg.static_init_prompt is None:
+            after = 0
+        else:
+            after = len(self.cfg.static_init_prompt)
+        while c > self.max_context_tokens or l > self.max_text_len - 20:
+            t = self.state.context.trim_words(after=after)
+            l -= t
+            c -= t
+            logger.debug(f"len {l}, c {c}, max_context_tokens {self.max_context_tokens}")
+            if t == 0:
+                break
+        logger.info(f"Context after trim: {self.state.context.text} (len: {l})")
+
+    def refresh_segment(self, complete=False):
+        """Refresh segment state."""
+        logger.debug("Refreshing segment:")
+        self.init_tokens()
+        self.state.last_attend_frame = -self.cfg.rewind_threshold
+        self.state.cumulative_time_offset = 0.0
+        self.init_context()
+        logger.debug(f"Context: {self.state.context}")
+        if not complete and len(self.state.segments) > 2:
+            self.state.segments = self.state.segments[-2:]
+        else:
+            logger.debug("removing all segments.")
+            self.state.segments = []
+        self.state.log_segments += 1
+        self.state.pending_incomplete_tokens = []
+
+    def fire_at_boundary(self, chunked_encoder_feature: mx.array) -> bool:
+        """Check if we should fire at word boundary (CIF-based)."""
+        if self.state.always_fire:
+            return True
+        if self.state.never_fire:
+            return False
+        return True
+
+    def _current_tokens(self) -> mx.array:
+        """Get current token sequence for decoding."""
+        toks = self.state.tokens
+        
+        if toks[0].shape[0] == 1:
+            toks[0] = mx.repeat(toks[0], self.cfg.beam_size, axis=0)
+
+        if not self.state.context.is_empty():
+            context_toks = self.state.context.as_mlx_array_beam(self.cfg.beam_size)
+            toks = [context_toks] + toks
+
+        # Concatenate all tokens
+        if len(toks) > 1:
+            current_tokens = mx.concatenate(toks, axis=1)
+        else:
+            current_tokens = toks[0]
+            
+        logger.debug("debug print current_tokens:")
+        self.debug_print_tokens(current_tokens)
+        return current_tokens
+
+    def debug_print_tokens(self, tokens: mx.array):
+        """Debug print token sequences."""
+        tokens_np = np.array(tokens)
+        for i in range(min(self.cfg.beam_size, tokens_np.shape[0])):
+            logger.debug(self.tokenizer.decode_with_timestamps(tokens_np[i].tolist()))
+
+    def segments_len(self) -> float:
+        """Get total length of audio segments in seconds."""
+        return sum(s.shape[0] for s in self.state.segments) / 16000
+
+    def _apply_minseglen(self) -> bool:
+        """Check if we have enough audio to process."""
+        segments_len = self.segments_len()
+        if segments_len < self.cfg.audio_min_len:
+            logger.debug("waiting for next segment")
+            return False
+        return True
+
+    def insert_audio(self, segment: np.ndarray = None):
+        """Insert audio segment into buffer."""
+        if segment is not None:
+            if hasattr(segment, 'numpy'):
+                segment = segment.numpy()
+            self.state.segments.append(segment)
+
+        removed_len = 0
+        segments_len = self.segments_len()
+        
+        while len(self.state.segments) > 1 and segments_len > self.cfg.audio_max_len:
+            removed_len = self.state.segments[0].shape[0] / 16000
+            segments_len -= removed_len
+            self.state.last_attend_frame -= int(TOKENS_PER_SECOND * removed_len)
+            self.state.cumulative_time_offset += removed_len
+            self.state.segments = self.state.segments[1:]
+            logger.debug(f"remove segments: {len(self.state.segments)} {len(self.state.tokens)}, cumulative offset: {self.state.cumulative_time_offset:.2f}s")
+            
+            if len(self.state.tokens) > 1:
+                # Convert MLX array to list for context
+                token_list = np.array(self.state.tokens[1][0, :]).tolist()
+                self.state.context.append_token_ids(token_list)
+                self.state.tokens = [self.state.initial_tokens] + self.state.tokens[2:]
+                
+        return removed_len
+
+    def _clean_cache(self):
+        """Clean the kv_cache after each inference step."""
+        self.state.clean_cache()
+
+    def _suppress_tokens(self, logits: mx.array) -> mx.array:
+        """Apply token suppression to logits."""
+        if self.state.suppress_tokens:
+            suppress_indices = mx.array(list(self.state.suppress_tokens), dtype=mx.int32)
+            logits = logits.at[:, suppress_indices].add(-float('inf'))
+        return logits
+
+    def lang_id(self, encoder_features: mx.array) -> Tuple[mx.array, List[dict]]:
+        """Language detection from encoder features."""
+        n_audio = encoder_features.shape[0]
+        x = mx.array([[self.tokenizer.sot]] * n_audio, dtype=mx.int32)
+        
+        logits, _, _ = self.model.decoder(x, encoder_features, kv_cache=None)
+        logits = logits[:, 0]
+        
+        mask = mx.ones(logits.shape[-1], dtype=mx.bool_)
+        language_token_indices = mx.array(list(self.tokenizer.all_language_tokens), dtype=mx.int32)
+        mask = mask.at[language_token_indices].add(False)
+        
+        logits = mx.where(mask, mx.array(-float('inf')), logits)
+        
+        language_tokens = mx.argmax(logits, axis=-1)
+        language_token_probs = mx.softmax(logits, axis=-1)
+        
+        probs_np = np.array(language_token_probs)
+        
+        language_probs = [
+            {
+                c: float(probs_np[i, j])
+                for j, c in zip(self.tokenizer.all_language_tokens, self.tokenizer.all_language_codes)
+            }
+            for i in range(n_audio)
+        ]
+
+        self._clean_cache()
+        return language_tokens, language_probs
+
+    def infer(self, is_last: bool = False) -> List[ASRToken]:
+        """
+        Main inference method.
+        
+        Args:
+            is_last: Whether this is the final chunk
+            
+        Returns:
+            List of timestamped ASR tokens
+        """
+        new_segment = True
+        
+        if len(self.state.segments) == 0:
+            logger.debug("No segments, nothing to do")
+            return []
+            
+        if not self._apply_minseglen():
+            logger.debug(f"applied minseglen {self.cfg.audio_min_len} > {self.segments_len()}.")
+            return []
+
+        if len(self.state.segments) > 1:
+            input_segments = np.concatenate(self.state.segments, axis=0)
+        else:
+            input_segments = self.state.segments[0]
+
+        beg_encode = time()
+        
+        mlx_mel_padded = mlx_log_mel_spectrogram(
+            audio=input_segments,
+            n_mels=self.model.dims.n_mels,
+            padding=N_SAMPLES
+        )
+        mlx_mel = mlx_pad_or_trim(mlx_mel_padded, N_FRAMES, axis=-2)
+        encoder_feature = self.model.encoder(mlx_mel[None])
+        content_mel_len = int((mlx_mel_padded.shape[0] - mlx_mel.shape[0]) / 2)
+        
+        mx.eval(encoder_feature)
+        
+        end_encode = time()
+        logger.debug(f'MLX Encoder duration: {end_encode - beg_encode:.3f}s')
+
+        if self.cfg.language == "auto" and self.state.detected_language is None and self.state.first_timestamp:
+            seconds_since_start = self.segments_len() - self.state.first_timestamp
+            if seconds_since_start >= 2.0:
+                language_tokens, language_probs = self.lang_id(encoder_feature)
+                top_lan, p = max(language_probs[0].items(), key=lambda x: x[1])
+                print(f"Detected language: {top_lan} with p={p:.4f}")
+                self.create_tokenizer(top_lan)
+                self.state.last_attend_frame = -self.cfg.rewind_threshold
+                self.state.cumulative_time_offset = 0.0
+                self.init_tokens()
+                self.init_context()
+                self.state.detected_language = top_lan
+                logger.info(f"Tokenizer language: {self.tokenizer.language}")
+
+        self.trim_context()
+        current_tokens = self._current_tokens()
+
+        fire_detected = self.fire_at_boundary(encoder_feature[:, :content_mel_len, :])
+
+        sum_logprobs = mx.zeros((self.cfg.beam_size,), dtype=mx.float32)
+        completed = False
+
+        attn_of_alignment_heads = None
+        most_attended_frame = None
+
+        token_len_before_decoding = current_tokens.shape[1]
+
+        l_absolute_timestamps = []
+        accumulated_cross_attns = []
+
+        audio_duration_s = self.segments_len()
+        # ~15 text tokens/s is a generous upper bound for speech; TOKENS_PER_SECOND (50)
+        # is the mel-frame rate and was causing 10-40x over-allocation on repetition loops.
+        max_tokens_per_chunk = max(50, int(audio_duration_s * 15 * 1.5))
+        tokens_produced_this_chunk = 0
+
+        while not completed and current_tokens.shape[1] < self.max_text_len:
+            tokens_produced_this_chunk += 1
+
+            if tokens_produced_this_chunk > max_tokens_per_chunk:
+                logger.warning(f"[Loop Detection] Too many tokens ({tokens_produced_this_chunk}) for {audio_duration_s:.2f}s audio. Breaking.")
+                current_tokens = current_tokens[:, :token_len_before_decoding]
+                break
+
+            if new_segment:
+                tokens_for_logits = current_tokens
+            else:
+                tokens_for_logits = current_tokens[:, -1:]
+
+            if self.state.decoder_type == "greedy":
+                logits, self.state.kv_cache, cross_qk = self.model.decoder(
+                    tokens_for_logits, encoder_feature, kv_cache=self.state.kv_cache
+                )
+            else:
+                logits, cross_qk = self.state.inference.logits(tokens_for_logits, encoder_feature)
+
+            mx.eval(logits)
+            
+            accumulated_cross_attns.append(cross_qk)
+            if len(accumulated_cross_attns) > 16:
+                accumulated_cross_attns = accumulated_cross_attns[-16:]
+
+            if new_segment and self.tokenizer.no_speech is not None:
+                probs_at_sot = mx.softmax(logits[:, self.state.sot_index, :], axis=-1)
+                no_speech_probs = np.array(probs_at_sot[:, self.tokenizer.no_speech]).tolist()
+                if no_speech_probs[0] > self.cfg.nonspeech_prob:
+                    logger.info("no speech, stop")
+                    break
+
+            logits = logits[:, -1, :]  # Last token logits
+
+            # Suppress tokens at segment start
+            if new_segment:
+                blank_tokens = self.tokenizer.encode(" ") + [self.tokenizer.eot]
+                logits = logits.at[:, blank_tokens].add(-float('inf'))
+            new_segment = False
+            
+            logits = self._suppress_tokens(logits)
+            
+            current_tokens, completed = self.state.token_decoder.update(
+                current_tokens, logits, sum_logprobs
+            )
+            mx.eval(current_tokens)
+
+            logger.debug(f"Decoding completed: {completed}")
+            self.debug_print_tokens(current_tokens)
+
+            attn_of_alignment_heads = self._process_cross_attention(
+                accumulated_cross_attns, content_mel_len
+            )
+
+            most_attended_frames = mx.argmax(attn_of_alignment_heads[:, -1, :], axis=-1)
+            most_attended_frames_np = np.array(most_attended_frames)
+
+            absolute_timestamps = [
+                (frame * 0.02 + self.state.cumulative_time_offset)
+                for frame in most_attended_frames_np.tolist()
+            ]
+
+            logger.debug(str(most_attended_frames_np.tolist()) + " most att frames")
+            logger.debug(f"Absolute timestamps: {absolute_timestamps}")
+
+            most_attended_frame = int(most_attended_frames_np[0])
+            l_absolute_timestamps.append(absolute_timestamps[0])
+
+            if completed:
+                current_tokens = current_tokens[:, :-1]
+                break
+            if not is_last and self.state.last_attend_frame - most_attended_frame > self.cfg.rewind_threshold:
+                current_tokens_np = np.array(current_tokens)
+                if current_tokens.shape[1] > 1 and current_tokens_np[0, -2] >= DEC_PAD:
+                    logger.debug("omit rewinding from special tokens")
+                    self.state.last_attend_frame = most_attended_frame
+                else:
+                    logger.debug(f"[rewind detected] current: {most_attended_frame}, last: {self.state.last_attend_frame}")
+                    self.state.last_attend_frame = -self.cfg.rewind_threshold
+                    current_tokens = mx.concatenate(self.state.tokens, axis=1) if len(self.state.tokens) > 0 else self.state.tokens[0]
+                    break
+            else:
+                self.state.last_attend_frame = most_attended_frame
+            if content_mel_len - most_attended_frame <= (4 if is_last else self.cfg.frame_threshold):
+                logger.debug(f"attention reaches the end: {most_attended_frame}/{content_mel_len}")
+                current_tokens = current_tokens[:, :-1]
+                break
+        tokens_to_split = np.array(current_tokens[0, token_len_before_decoding:]).tolist()
+        if self.state.pending_incomplete_tokens:
+            logger.debug(f"[UTF-8 Fix] Prepending pending tokens: {self.state.pending_incomplete_tokens}")
+            tokens_to_split = self.state.pending_incomplete_tokens + tokens_to_split
+
+        if fire_detected or is_last:
+            new_hypothesis = tokens_to_split
+            split_words, split_tokens = self.tokenizer.split_to_word_tokens(new_hypothesis)
+        else:
+            split_words, split_tokens = self.tokenizer.split_to_word_tokens(tokens_to_split)
+            if len(split_words) > 1:
+                new_hypothesis = [i for sublist in split_tokens[:-1] for i in sublist]
+            else:
+                new_hypothesis = []
+
+        logger.debug(f"new_hypothesis: {new_hypothesis}")
+        new_tokens = mx.array([new_hypothesis], dtype=mx.int32)
+        new_tokens = mx.repeat(new_tokens, self.cfg.beam_size, axis=0)
+        self.state.tokens.append(new_tokens)
+
+        logger.info(f"Output: {self.tokenizer.decode(new_hypothesis)}")
+
+        self._clean_cache()
+
+        if len(l_absolute_timestamps) >= 2 and self.state.first_timestamp is None:
+            self.state.first_timestamp = l_absolute_timestamps[0]
+        timestamped_words = []
+        timestamp_idx = 0
+        replacement_char = "\ufffd"
+        
+        for word, word_tokens in zip(split_words, split_tokens):
+            if replacement_char in word:
+                logger.warning(f"[UTF-8 Filter] Skipping: {repr(word)}")
+                timestamp_idx += len(word_tokens)
+                continue
+
+            try:
+                current_timestamp = l_absolute_timestamps[timestamp_idx]
+            except IndexError:
+                pass
+            timestamp_idx += len(word_tokens)
+
+            timestamp_entry = ASRToken(
+                start=round(current_timestamp, 2),
+                end=round(current_timestamp + 0.1, 2),
+                text=word,
+                speaker=self.state.speaker,
+                detected_language=self.state.detected_language
+            ).with_offset(self.state.global_time_offset)
+            timestamped_words.append(timestamp_entry)
+        self.state.pending_incomplete_tokens = []
+        MAX_PENDING_TOKENS = 10
+        if split_words and replacement_char in split_words[-1]:
+            if len(split_tokens[-1]) <= MAX_PENDING_TOKENS:
+                self.state.pending_incomplete_tokens = split_tokens[-1]
+                logger.debug(f"[UTF-8 Fix] Holding incomplete tokens")
+            else:
+                logger.warning(f"[UTF-8 Fix] Skipping too many tokens")
+
+        return timestamped_words
+
+    def _process_cross_attention(
+        self,
+        cross_attns: List[List[mx.array]],
+        content_mel_len: int
+    ) -> mx.array:
+        """
+        Process cross-attention weights for alignment.
+        
+        Args:
+            cross_attns: List of cross-attention from each forward pass
+                        Each element is a list of mx.arrays per layer
+            content_mel_len: Length of actual audio content
+            
+        Returns:
+            Processed attention tensor, shape (batch, seq_len, content_mel_len)
+        """
+        attn_of_alignment_heads = [[] for _ in range(self.state.num_align_heads)]
+        num_decoder_layers = self.num_decoder_layers
+
+        if cross_attns and isinstance(cross_attns[0], list):
+            flattened_attns = [attn for layer_list in cross_attns for attn in layer_list]
+        else:
+            flattened_attns = cross_attns
+
+        for idx, attn_mat in enumerate(flattened_attns):
+            if attn_mat is None:
+                continue
+                
+            layer_rank = idx % num_decoder_layers
+            align_heads_in_layer = self.state.align_source.get(layer_rank, [])
+            
+            if len(align_heads_in_layer) == 0:
+                continue
+            attn_mat = mx.softmax(attn_mat, axis=-1)
+
+            for align_head_rank, head_id in align_heads_in_layer:
+                if self.cfg.beam_size == 1:
+                    if attn_mat.ndim == 4:
+                        a = attn_mat[0, head_id, :, :]
+                    else:
+                        a = attn_mat[head_id, :, :]
+                    a = a[None, :, :]
+                else:
+                    a = attn_mat[:, head_id, :, :]
+                attn_of_alignment_heads[align_head_rank].append(a)
+        tmp = []
+        for mat in attn_of_alignment_heads:
+            if mat:
+                t = mx.concatenate(mat, axis=1)
+                tmp.append(t)
+
+        if not tmp:
+            return mx.zeros((self.cfg.beam_size, 1, content_mel_len))
+        attn_of_alignment_heads = mx.stack(tmp, axis=1)
+
+        std = mx.std(attn_of_alignment_heads, axis=-2, keepdims=True)
+        mean = mx.mean(attn_of_alignment_heads, axis=-2, keepdims=True)
+        attn_of_alignment_heads = (attn_of_alignment_heads - mean) / (std + 1e-8)
+
+        attn_of_alignment_heads = mlx_median_filter(attn_of_alignment_heads, 7)
+
+        attn_of_alignment_heads = mx.mean(attn_of_alignment_heads, axis=1)
+
+        attn_of_alignment_heads = attn_of_alignment_heads[:, :, :content_mel_len]
+
+        mx.eval(attn_of_alignment_heads)
+        return attn_of_alignment_heads
+

whisperlivekit/simul_whisper/mlx_encoder.py

@@ -69,3 +69,39 @@ def load_mlx_encoder(
     model.update(encoder_weights)
     mx.eval(model.parameters())
     return model
+
+
+def load_mlx_model(
+    path_or_hf_repo: str,
+    dtype: mx.Dtype = mx.float32,
+) -> whisper.Whisper:
+    model_path = Path(path_or_hf_repo)
+    if not model_path.exists():
+        model_path = Path(snapshot_download(repo_id=path_or_hf_repo))
+
+    with open(str(model_path / "config.json"), "r") as f:
+        config = json.loads(f.read())
+        config.pop("model_type", None)
+        quantization = config.pop("quantization", None)
+
+    model_args = whisper.ModelDimensions(**config)
+
+    wf = model_path / "weights.safetensors"
+    if not wf.exists():
+        wf = model_path / "weights.npz"
+    weights = mx.load(str(wf))
+
+    model = whisper.Whisper(model_args, dtype)
+
+    if quantization is not None:
+        class_predicate = (
+            lambda p, m: isinstance(m, (nn.Linear, nn.Embedding))
+            and f"{p}.scales" in weights
+        )
+        nn.quantize(model, **quantization, class_predicate=class_predicate)
+
+    weights = tree_unflatten(list(weights.items()))
+    
+    model.update(weights)
+    mx.eval(model.parameters())
+    return model

whisperlivekit/simul_whisper/simul_whisper.py

@@ -390,7 +390,6 @@ class AlignAtt:
             return []
         if not self._apply_minseglen():
             logger.debug(f"applied minseglen {self.cfg.audio_min_len} > {self.segments_len()}.")
-            input_segments = torch.cat(self.state.segments, dim=0)
             return []
 
         # input_segments is concatenation of audio, it's one array
@@ -485,7 +484,9 @@ class AlignAtt:
         accumulated_cross_attns = []
         
         audio_duration_s = self.segments_len()
-        max_tokens_per_chunk = max(50, int(audio_duration_s * TOKENS_PER_SECOND * 2.0))  # 2x margin, min 50
+        # ~15 text tokens/s is a generous upper bound for speech; TOKENS_PER_SECOND (50)
+        # is the mel-frame rate and was causing 10-40x over-allocation on repetition loops.
+        max_tokens_per_chunk = max(50, int(audio_duration_s * 15 * 1.5))
         tokens_produced_this_chunk = 0
         
         while not completed and current_tokens.shape[1] < self.max_text_len:  # bos is 3 tokens
@@ -506,8 +507,12 @@ class AlignAtt:
             result = self.logits(tokens_for_logits, encoder_feature, return_cross_attn=True)
             logits, cross_attns = result
             
-            # Accumulate cross-attention from this forward pass
+            # Accumulate cross-attention from this forward pass (rolling window to
+            # bound VRAM — only the last entry matters for alignment, and the
+            # median_filter kernel is 7, so 16 entries is more than enough).
             accumulated_cross_attns.append(cross_attns)
+            if len(accumulated_cross_attns) > 16:
+                accumulated_cross_attns = accumulated_cross_attns[-16:]
 
             if new_segment and self.tokenizer.no_speech is not None:
                 probs_at_sot = logits[:, self.state.sot_index, :].float().softmax(dim=-1)
@@ -626,8 +631,10 @@ class AlignAtt:
 
             try:
                 current_timestamp = l_absolute_timestamps[timestamp_idx]
-            except:
+            except IndexError:
-                pass
+                # Use last timestamp if index out of range
+                logger.warning(f"Timestamp index {timestamp_idx} out of range, using last timestamp")
+                current_timestamp = l_absolute_timestamps[-1] if l_absolute_timestamps else 0.0
             timestamp_idx += len(word_tokens)
 
             timestamp_entry = ASRToken(

whisperlivekit/thread_safety.py

@@ -0,0 +1,139 @@
+"""
+Thread Safety Configuration for WhisperLiveKit
+
+This module provides thread safety configuration and utilities.
+
+Environment Variables:
+    WHISPERLIVEKIT_MODEL_LOCK: Enable/disable model locking (default: 1)
+        Set to "0" to disable for single-connection deployments
+
+    WHISPERLIVEKIT_LOCK_TIMEOUT: Lock acquisition timeout in seconds (default: 30)
+
+Usage:
+    # Enable model locking (default)
+    export WHISPERLIVEKIT_MODEL_LOCK=1
+
+    # Disable for single-connection deployment
+    export WHISPERLIVEKIT_MODEL_LOCK=0
+
+    # Custom timeout
+    export WHISPERLIVEKIT_LOCK_TIMEOUT=60
+"""
+
+import os
+import logging
+import threading
+
+logger = logging.getLogger(__name__)
+
+# Configuration
+USE_MODEL_LOCK = os.environ.get("WHISPERLIVEKIT_MODEL_LOCK", "1") == "1"
+LOCK_TIMEOUT = float(os.environ.get("WHISPERLIVEKIT_LOCK_TIMEOUT", "30.0"))
+
+# Global model lock
+_model_lock = threading.Lock()
+
+# Log configuration on import
+if USE_MODEL_LOCK:
+    logger.info(f"Model locking ENABLED (timeout: {LOCK_TIMEOUT}s)")
+    logger.info("For single-connection deployments, set WHISPERLIVEKIT_MODEL_LOCK=0")
+else:
+    logger.warning("Model locking DISABLED - only safe for single-connection deployments")
+
+
+def get_model_lock():
+    """Get the global model lock instance"""
+    return _model_lock
+
+
+def acquire_model_lock(timeout=None):
+    """
+    Acquire model lock with timeout.
+
+    Args:
+        timeout: Lock acquisition timeout (default: use LOCK_TIMEOUT)
+
+    Returns:
+        bool: True if lock acquired, False on timeout
+    """
+    if not USE_MODEL_LOCK:
+        return True
+
+    timeout = timeout or LOCK_TIMEOUT
+    acquired = _model_lock.acquire(timeout=timeout)
+
+    if not acquired:
+        logger.error(f"Failed to acquire model lock within {timeout}s")
+
+    return acquired
+
+
+def release_model_lock():
+    """Release model lock"""
+    if not USE_MODEL_LOCK:
+        return
+
+    try:
+        _model_lock.release()
+    except RuntimeError:
+        # Lock not held - this is fine
+        pass
+
+
+class ModelLockContext:
+    """Context manager for model lock"""
+
+    def __init__(self, timeout=None):
+        self.timeout = timeout
+        self.acquired = False
+
+    def __enter__(self):
+        self.acquired = acquire_model_lock(self.timeout)
+        return self.acquired
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        if self.acquired:
+            release_model_lock()
+        return False
+
+
+# Concurrency recommendations
+RECOMMENDED_CONNECTIONS_PER_WORKER = 1 if USE_MODEL_LOCK else 1
+RECOMMENDED_WORKERS = 4
+
+def print_deployment_recommendations():
+    """Print recommended deployment configuration"""
+    print("\n" + "="*60)
+    print("WhisperLiveKit Deployment Recommendations")
+    print("="*60)
+
+    if USE_MODEL_LOCK:
+        print("⚠️  Model locking is ENABLED")
+        print("   This serializes inference across connections.")
+        print()
+        print("Recommended deployment:")
+        print(f"  gunicorn -w {RECOMMENDED_WORKERS} \\")
+        print("    -k uvicorn.workers.UvicornWorker \\")
+        print("    --worker-connections 1 \\")
+        print("    whisperlivekit.basic_server:app")
+        print()
+        print("Expected capacity:")
+        print(f"  - {RECOMMENDED_WORKERS} concurrent users (1 per worker)")
+        print(f"  - Memory: ~{RECOMMENDED_WORKERS}x model size")
+    else:
+        print("✅ Model locking is DISABLED")
+        print("   ⚠️  ONLY safe for single-connection deployments")
+        print()
+        print("Recommended deployment:")
+        print("  uvicorn whisperlivekit.basic_server:app \\")
+        print("    --host 0.0.0.0 --port 8000 \\")
+        print("    --workers 1")
+        print()
+        print("Expected capacity:")
+        print("  - 1 concurrent user only")
+
+    print("="*60 + "\n")
+
+
+if __name__ == "__main__":
+    print_deployment_recommendations()

whisperlivekit/timed_objects.py

@@ -39,10 +39,11 @@ class TimedText(Timed):
 
 @dataclass()
 class ASRToken(TimedText):
+    probability: Optional[float] = None
 
     def with_offset(self, offset: float) -> "ASRToken":
         """Return a new token with the time offset added."""
-        return ASRToken(self.start + offset, self.end + offset, self.text, self.speaker, detected_language=self.detected_language)
+        return ASRToken(self.start + offset, self.end + offset, self.text, self.speaker, detected_language=self.detected_language, probability=self.probability)
 
     def is_silence(self) -> bool:
         return False

whisperlivekit/tokens_alignment.py

@@ -49,9 +49,12 @@ class TokensAlignment:
 
     def add_translation(self, segment: Segment) -> None:
         """Append translated text segments that overlap with a segment."""
+        if segment.translation is None:
+            segment.translation = ''
         for ts in self.all_translation_segments:
             if ts.is_within(segment):
-                segment.translation += ts.text + (self.sep if ts.text else '')
+                if ts.text:
+                    segment.translation += ts.text + self.sep
             elif segment.translation:
                 break
 
@@ -183,9 +186,9 @@ class TokensAlignment:
         else:
             diarization_buffer = ''
             for token in self.new_tokens:
-                if token.is_silence():
+                if isinstance(token, Silence):
                     if self.current_line_tokens:
-                        self.validated_segments.append(Segment().from_tokens(self.current_line_tokens))
+                        self.validated_segments.append(Segment.from_tokens(self.current_line_tokens))
                         self.current_line_tokens = []
 
                     end_silence = token.end if token.has_ended else time() - self.beg_loop
@@ -201,7 +204,7 @@ class TokensAlignment:
             
             segments = list(self.validated_segments)
             if self.current_line_tokens:
-                segments.append(Segment().from_tokens(self.current_line_tokens))
+                segments.append(Segment.from_tokens(self.current_line_tokens))
 
         if current_silence:
             end_silence = current_silence.end if current_silence.has_ended else time() - self.beg_loop

whisperlivekit/voxtral_streaming.py

@@ -0,0 +1,484 @@
+"""
+Voxtral Mini Realtime streaming backend using voxmlx's incremental encode/decode.
+
+Uses model.encode_step() for incremental audio encoding and token-by-token
+autoregressive decoding, matching voxmlx's native streaming pipeline.
+"""
+
+import logging
+import sys
+import time
+from typing import List, Optional, Tuple
+
+import numpy as np
+
+from whisperlivekit.timed_objects import ASRToken, Transcript
+
+logger = logging.getLogger(__name__)
+
+N_LEFT_PAD_TOKENS = 32
+N_RIGHT_PAD_TOKENS = 17
+
+
+class VoxtralStreamingASR:
+    """Voxtral model holder for the streaming pipeline."""
+
+    sep = " "
+
+    def __init__(self, logfile=sys.stderr, **kwargs):
+        from voxmlx import _build_prompt_tokens
+        from voxmlx import load_model as vox_load_model
+
+        self.logfile = logfile
+        self.transcribe_kargs = {}
+
+        lan = kwargs.get("lan", "auto")
+        self.original_language = None if lan == "auto" else lan
+
+        DEFAULT_MODEL = "mlx-community/Voxtral-Mini-4B-Realtime-6bit"
+        model_path = kwargs.get("model_dir") or kwargs.get("model_path")
+        if not model_path:
+            model_size = kwargs.get("model_size", "")
+            # Only use model_size if it looks like a HF repo or a path, not a Whisper size name
+            if model_size and ("/" in model_size or model_size.startswith(".")):
+                model_path = model_size
+            else:
+                model_path = DEFAULT_MODEL
+
+        t = time.time()
+        logger.info(f"Loading Voxtral model '{model_path}' via voxmlx...")
+        self.model, self._tokenizer, self._config = vox_load_model(model_path)
+        self._prompt_tokens, self._n_delay_tokens = _build_prompt_tokens(
+            self._tokenizer
+        )
+        logger.info(f"Voxtral model loaded in {time.time() - t:.2f}s")
+
+        self.backend_choice = "voxtral-mlx"
+        self.tokenizer = None  # sentence tokenizer — not needed for streaming
+
+    def transcribe(self, audio):
+        pass
+
+
+class VoxtralStreamingOnlineProcessor:
+    """
+    Online processor for Voxtral streaming ASR.
+
+    Uses voxmlx's incremental encoding (encode_step) and token-by-token
+    autoregressive decoding. Each decode step corresponds to 80ms of audio.
+    """
+
+    SAMPLING_RATE = 16000
+
+    def __init__(self, asr: VoxtralStreamingASR, logfile=sys.stderr):
+        from mistral_common.tokens.tokenizers.base import SpecialTokenPolicy
+
+        self.asr = asr
+        self.logfile = logfile
+        self.end = 0.0
+        self.buffer = []
+        self.audio_buffer = np.array([], dtype=np.float32)  # for logging compat
+        self._special_token_policy = SpecialTokenPolicy.IGNORE
+        self._reset_state()
+        logger.info(
+            f"[voxtral] Initialized. eos_id={asr._tokenizer.eos_id}, "
+            f"prefix_len={len(asr._prompt_tokens)}, "
+            f"n_delay={asr._n_delay_tokens}"
+        )
+
+    def _reset_state(self):
+        from voxmlx.audio import SAMPLES_PER_TOKEN
+
+        self._samples_per_token = SAMPLES_PER_TOKEN
+
+        # Incremental encoder state
+        self._audio_tail = None
+        self._conv1_tail = None
+        self._conv2_tail = None
+        self._encoder_cache = None
+        self._ds_buf = None
+
+        # Decoder state
+        self._decoder_cache = None
+        self._y = None  # last sampled token (mx.array scalar)
+        self._t_cond = None
+        self._text_embeds = None
+
+        # Audio / decode tracking
+        self._pending_audio = np.zeros(0, dtype=np.float32)
+        self._audio_embeds = None
+        self._n_audio_samples_fed = 0
+        self._n_total_decoded = 0
+        self._first_cycle = True
+        self._prefilled = False
+
+        # Word extraction: accumulate token IDs, full-sequence decode for correct spacing
+        self._output_token_ids: List[int] = []
+        self._token_positions: List[int] = []  # decode position for each token
+        self._n_committed_words = 0
+        self._global_time_offset = 0.0
+        self._y_flushed_to_output = False  # True after start_silence flushes pending _y
+
+    # ── Interface methods (same as SimulStreamingOnlineProcessor) ──
+
+    def insert_audio_chunk(self, audio: np.ndarray, audio_stream_end_time: float):
+        self.end = audio_stream_end_time
+        self._pending_audio = np.append(self._pending_audio, audio)
+        self.audio_buffer = self._pending_audio  # for logging compat
+
+    def process_iter(self, is_last=False) -> Tuple[List[ASRToken], float]:
+        try:
+            return self._process_iter_inner(is_last)
+        except Exception as e:
+            logger.warning(f"[voxtral] process_iter exception: {e}", exc_info=True)
+            return [], self.end
+
+    def _get_full_text(self) -> str:
+        """Decode all accumulated token IDs at once for correct spacing."""
+        if not self._output_token_ids:
+            return ""
+        sp = self.asr._tokenizer
+        return sp.decode(self._output_token_ids, special_token_policy=self._special_token_policy)
+
+    def get_buffer(self) -> Transcript:
+        """Return all uncommitted text as buffer, including pending _y token."""
+        # Temporarily include pending _y for buffer display
+        ids = list(self._output_token_ids)
+        if self._y is not None and not self._y_flushed_to_output:
+            sp = self.asr._tokenizer
+            token_id = self._y.item()
+            if token_id != sp.eos_id:
+                ids.append(token_id)
+        if not ids:
+            return Transcript(start=None, end=None, text="")
+        sp = self.asr._tokenizer
+        full_text = sp.decode(ids, special_token_policy=self._special_token_policy)
+        words = full_text.split()
+        uncommitted = words[self._n_committed_words:]
+        if uncommitted:
+            text = " ".join(uncommitted)
+            return Transcript(start=self.end, end=self.end, text=text)
+        return Transcript(start=None, end=None, text="")
+
+    def start_silence(self) -> Tuple[List[ASRToken], float]:
+        """Flush all uncommitted words when silence starts."""
+        self._flush_last_y()  # Include the pending _y token before flushing
+        words = self._flush_all_pending_words()
+        logger.info(f"[voxtral] start_silence: flushed {len(words)} words")
+        return words, self.end
+
+    def end_silence(self, silence_duration: float, offset: float):
+        self._global_time_offset += silence_duration
+        self.end += silence_duration
+
+    def new_speaker(self, change_speaker):
+        self.start_silence()
+
+    def warmup(self, audio, init_prompt=""):
+        pass
+
+    def finish(self) -> Tuple[List[ASRToken], float]:
+        """Flush remaining audio with right-padding to let the model finish decoding."""
+        right_pad = np.zeros(
+            N_RIGHT_PAD_TOKENS * self._samples_per_token, dtype=np.float32
+        )
+        self._pending_audio = np.append(self._pending_audio, right_pad)
+        self._n_audio_samples_fed += len(right_pad)
+
+        final_words, _ = self._process_iter_inner(is_last=True)
+        # Flush the last pending self._y token (like voxmlx's finally block)
+        self._flush_last_y()
+        final_words.extend(self._flush_all_pending_words())
+        return final_words, self.end
+
+    # ── Word extraction ──
+
+    def _pos_to_time(self, pos: int) -> float:
+        """Convert a decode position to seconds relative to audio start."""
+        SPT = self._samples_per_token
+        return max(0.0, (pos - N_LEFT_PAD_TOKENS) * SPT / self.SAMPLING_RATE)
+
+    def _flush_last_y(self):
+        """Flush the last pending self._y token that hasn't been processed yet."""
+        if self._y is None or self._y_flushed_to_output:
+            return
+        sp = self.asr._tokenizer
+        token_id = self._y.item()
+        if token_id != sp.eos_id:
+            self._output_token_ids.append(token_id)
+            self._token_positions.append(self._n_total_decoded)
+            self._y_flushed_to_output = True
+
+    def _extract_new_words(self) -> List[ASRToken]:
+        """
+        Split accumulated text into words and return new complete words
+        (all but the last, which may still be growing).
+        """
+        if not self._output_token_ids:
+            return []
+
+        full_text = self._get_full_text()
+        words = full_text.split()
+
+        new_words: List[ASRToken] = []
+        n_tokens = len(self._output_token_ids)
+        # All words except the last are guaranteed complete
+        while len(words) > self._n_committed_words + 1:
+            word = words[self._n_committed_words]
+            word_idx = self._n_committed_words
+            n_words_total = len(words)
+            # Approximate: assign token range proportionally
+            tok_start = int(word_idx / n_words_total * n_tokens)
+            tok_end = int((word_idx + 1) / n_words_total * n_tokens)
+            tok_start = min(tok_start, len(self._token_positions) - 1)
+            tok_end = min(tok_end, len(self._token_positions) - 1)
+
+            start_time = self._pos_to_time(self._token_positions[tok_start]) + self._global_time_offset
+            end_time = self._pos_to_time(self._token_positions[tok_end]) + self._global_time_offset
+
+            # Prepend space to match Whisper convention (Segment.from_tokens joins with '')
+            text = word if self._n_committed_words == 0 else " " + word
+            new_words.append(ASRToken(start=start_time, end=end_time, text=text))
+            self._n_committed_words += 1
+
+        return new_words
+
+    def _flush_all_pending_words(self) -> List[ASRToken]:
+        """Flush ALL words including the last partial one."""
+        if not self._output_token_ids:
+            return []
+
+        full_text = self._get_full_text()
+        words = full_text.split()
+
+        new_words: List[ASRToken] = []
+        n_tokens = len(self._output_token_ids)
+        n_words_total = max(len(words), 1)
+
+        while self._n_committed_words < len(words):
+            word = words[self._n_committed_words]
+            word_idx = self._n_committed_words
+
+            tok_start = int(word_idx / n_words_total * n_tokens)
+            tok_end = int((word_idx + 1) / n_words_total * n_tokens)
+            tok_start = min(tok_start, max(len(self._token_positions) - 1, 0))
+            tok_end = min(tok_end, max(len(self._token_positions) - 1, 0))
+
+            if self._token_positions:
+                start_time = self._pos_to_time(self._token_positions[tok_start]) + self._global_time_offset
+                end_time = self._pos_to_time(self._token_positions[tok_end]) + self._global_time_offset
+            else:
+                start_time = self._global_time_offset
+                end_time = self._global_time_offset
+
+            # Prepend space to match Whisper convention (Segment.from_tokens joins with '')
+            text = word if self._n_committed_words == 0 else " " + word
+            new_words.append(ASRToken(start=start_time, end=end_time, text=text))
+            self._n_committed_words += 1
+
+        return new_words
+
+    # ── Core streaming logic ──
+
+    def _process_iter_inner(self, is_last: bool) -> Tuple[List[ASRToken], float]:
+        import mlx.core as mx
+
+        from voxmlx.audio import log_mel_spectrogram_step
+        from voxmlx.cache import RotatingKVCache
+
+        model = self.asr.model
+        sp = self.asr._tokenizer
+        prompt_tokens = self.asr._prompt_tokens
+        prefix_len = len(prompt_tokens)
+        SPT = self._samples_per_token
+
+        # ── Phase 1: Encode new audio ──
+        if self._first_cycle and len(self._pending_audio) >= SPT:
+            left_pad = np.zeros(N_LEFT_PAD_TOKENS * SPT, dtype=np.float32)
+            n_feed = (len(self._pending_audio) // SPT) * SPT
+            chunk = np.concatenate([left_pad, self._pending_audio[:n_feed]])
+            self._pending_audio = self._pending_audio[n_feed:]
+            self._n_audio_samples_fed += n_feed
+
+            mel, self._audio_tail = log_mel_spectrogram_step(
+                chunk, self._audio_tail
+            )
+            (
+                new_embeds,
+                self._conv1_tail,
+                self._conv2_tail,
+                self._encoder_cache,
+                self._ds_buf,
+            ) = model.encode_step(
+                mel,
+                self._conv1_tail,
+                self._conv2_tail,
+                self._encoder_cache,
+                self._ds_buf,
+            )
+            if new_embeds is not None:
+                mx.eval(new_embeds)
+                self._audio_embeds = new_embeds
+                logger.info(f"[voxtral] first encode: {new_embeds.shape[0]} embeds from {n_feed} samples")
+            else:
+                logger.info(f"[voxtral] first encode: no embeds from {n_feed} samples")
+            self._first_cycle = False
+
+        elif not self._first_cycle and len(self._pending_audio) >= SPT:
+            n_feed = (len(self._pending_audio) // SPT) * SPT
+            chunk = self._pending_audio[:n_feed]
+            self._pending_audio = self._pending_audio[n_feed:]
+            self._n_audio_samples_fed += n_feed
+
+            mel, self._audio_tail = log_mel_spectrogram_step(
+                chunk, self._audio_tail
+            )
+            (
+                new_embeds,
+                self._conv1_tail,
+                self._conv2_tail,
+                self._encoder_cache,
+                self._ds_buf,
+            ) = model.encode_step(
+                mel,
+                self._conv1_tail,
+                self._conv2_tail,
+                self._encoder_cache,
+                self._ds_buf,
+            )
+            if new_embeds is not None:
+                mx.eval(new_embeds)
+                if self._audio_embeds is not None:
+                    self._audio_embeds = mx.concatenate(
+                        [self._audio_embeds, new_embeds]
+                    )
+                else:
+                    self._audio_embeds = new_embeds
+
+        self.audio_buffer = self._pending_audio  # for logging compat
+
+        if self._audio_embeds is None:
+            return [], self.end
+
+        # Safety: don't decode ahead of encoded audio
+        safe_total = (
+            N_LEFT_PAD_TOKENS + self._n_audio_samples_fed // SPT
+        )
+        n_decodable = min(
+            self._audio_embeds.shape[0], safe_total - self._n_total_decoded
+        )
+
+        if n_decodable <= 0:
+            return [], self.end
+
+        # ── Phase 2: Prefill (once per utterance) ──
+        if not self._prefilled:
+            if self._n_total_decoded + self._audio_embeds.shape[0] < prefix_len:
+                logger.info(
+                    f"[voxtral] waiting for prefill: have {self._audio_embeds.shape[0]} embeds, need {prefix_len}"
+                )
+                return [], self.end
+
+            n_layers = len(model.language_model.layers)
+            self._decoder_cache = [RotatingKVCache(8192) for _ in range(n_layers)]
+
+            self._t_cond = model.time_embedding(
+                mx.array([self.asr._n_delay_tokens], dtype=mx.float32)
+            )
+
+            prompt_ids = mx.array([prompt_tokens])
+            self._text_embeds = model.language_model.embed(prompt_ids)[0]
+
+            prefix_embeds = (
+                self._text_embeds + self._audio_embeds[:prefix_len]
+            )[None, :, :]
+
+            logits = model.decode(
+                prefix_embeds, self._t_cond, "causal", self._decoder_cache
+            )
+            mx.eval(
+                logits,
+                *[x for c in self._decoder_cache for x in (c.keys, c.values)],
+            )
+
+            self._y = mx.argmax(logits[0, -1:], axis=-1).squeeze()
+            mx.async_eval(self._y)
+
+            self._audio_embeds = self._audio_embeds[prefix_len:]
+            self._n_total_decoded = prefix_len
+            self._prefilled = True
+            logger.info(f"[voxtral] prefill done, first token y={self._y.item()}")
+
+            n_decodable = min(
+                self._audio_embeds.shape[0], safe_total - self._n_total_decoded
+            )
+
+        if n_decodable <= 0:
+            return [], self.end
+
+        # ── Phase 3: Decode new positions ──
+        eos_id = sp.eos_id
+        hit_eos = False
+        n_consumed = 0
+
+        for i in range(n_decodable):
+            token_embed = model.language_model.embed(self._y.reshape(1, 1))[0, 0]
+            step_embed = (self._audio_embeds[i] + token_embed)[None, None, :]
+            logits = model.decode(
+                step_embed, self._t_cond, mask=None, cache=self._decoder_cache
+            )
+            next_y = mx.argmax(logits[0, -1:], axis=-1).squeeze()
+            mx.async_eval(next_y)
+
+            token_id = self._y.item()
+            n_consumed = i + 1
+
+            if token_id == eos_id:
+                hit_eos = True
+                logger.info("[voxtral] hit EOS")
+                break
+
+            # Accumulate token ID — full-sequence decode produces correct spacing
+            # Skip if this _y was already flushed by start_silence()
+            if self._y_flushed_to_output:
+                self._y_flushed_to_output = False
+            else:
+                self._output_token_ids.append(token_id)
+                # Track position for timestamp estimation
+                pos = self._n_total_decoded + i
+                self._token_positions.append(pos)
+
+            if i > 0 and i % 256 == 0:
+                mx.clear_cache()
+
+            self._y = next_y
+
+        self._n_total_decoded += n_consumed
+
+        # Trim consumed embeddings
+        if self._audio_embeds.shape[0] > n_consumed:
+            self._audio_embeds = self._audio_embeds[n_consumed:]
+        else:
+            self._audio_embeds = None
+
+        # Log decode results
+        full_text = self._get_full_text()
+        logger.info(
+            f"[voxtral] decoded {n_consumed} tokens | "
+            f"total_decoded={self._n_total_decoded} | "
+            f"text='{full_text[-80:]}' | "
+            f"n_words={len(full_text.split())} committed={self._n_committed_words}"
+        )
+
+        # Extract complete words from the decoded token sequence
+        new_words = self._extract_new_words()
+
+        if hit_eos:
+            new_words.extend(self._flush_all_pending_words())
+            self._reset_state()
+
+        if new_words:
+            logger.info(f"[voxtral] returning {len(new_words)} words: {[w.text for w in new_words]}")
+
+        self.buffer = []
+        return new_words, self.end

whisperlivekit/whisper/__init__.py

@@ -108,7 +108,7 @@ def available_models() -> List[str]:
 def _infer_dims_from_config(path: str) -> Optional[ModelDimensions]:
     """
     attempt to infer ModelDimensions from a HF style config.json located
-    next to the given checkpoint, usefull for distilled models
+    next to the given checkpoint, usefull for distilled models/MLX models.
     """
     candidates = []
     if os.path.isdir(path):
@@ -122,6 +122,25 @@ def _infer_dims_from_config(path: str) -> Optional[ModelDimensions]:
         with open(candidate, "r", encoding="utf-8") as f:
             config = json.load(f)
 
+        # native Whisper format
+        native_keys = ["n_mels", "n_audio_ctx", "n_audio_state", "n_audio_head",
+                       "n_audio_layer", "n_vocab", "n_text_ctx", "n_text_state",
+                       "n_text_head", "n_text_layer"]
+        if all(k in config for k in native_keys):
+            return ModelDimensions(
+                n_mels=config["n_mels"],
+                n_audio_ctx=config["n_audio_ctx"],
+                n_audio_state=config["n_audio_state"],
+                n_audio_head=config["n_audio_head"],
+                n_audio_layer=config["n_audio_layer"],
+                n_vocab=config["n_vocab"],
+                n_text_ctx=config["n_text_ctx"],
+                n_text_state=config["n_text_state"],
+                n_text_head=config["n_text_head"],
+                n_text_layer=config["n_text_layer"],
+            )
+
+        # HuggingFace format
         try:
             return ModelDimensions(
                 n_mels=config["num_mel_bins"],
@@ -236,6 +255,24 @@ def _convert_hf_state_dict(state_dict: Dict[str, torch.Tensor]) -> Dict[str, tor
     return converted if converted else state_dict
 
 
+def _convert_mlx_state_dict(state_dict: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]:
+    """
+    Converts an mlx whisper checkpoint to a default openai whisper one
+    """
+    if not any("mlp1" in k or "mlp2" in k for k in state_dict):
+        return state_dict
+
+    converted = {}
+    for key, value in state_dict.items():
+        if key == "alignment_heads":
+            continue
+        
+        new_key = key.replace(".mlp1.", ".mlp.0.").replace(".mlp2.", ".mlp.2.")
+        converted[new_key] = value
+
+    return converted
+
+
 def _load_lora_state(lora_path: str):
     safe_path = os.path.join(lora_path, "adapter_model.safetensors")
     bin_path = os.path.join(lora_path, "adapter_model.bin")
@@ -520,7 +557,12 @@ def load_model(
         state_dict = checkpoint["model_state_dict"]
     else:
         state_dict = checkpoint
+    
+    if alignment_heads is None and "alignment_heads" in state_dict:
+        alignment_heads = state_dict["alignment_heads"]
+    
     state_dict = _convert_hf_state_dict(state_dict)
+    state_dict = _convert_mlx_state_dict(state_dict)
     _apply_lora_adapter(state_dict, lora_path)
 
     if dims_cfg is not None:
@@ -546,8 +588,13 @@ def load_model(
     model.load_state_dict(state_dict)
 
     if alignment_heads is not None:
-        model.set_alignment_heads(alignment_heads)
+        if isinstance(alignment_heads, bytes):
-
+            model.set_alignment_heads(alignment_heads)
+        elif isinstance(alignment_heads, torch.Tensor): #for mlx whisper
+            mask = torch.zeros(dims.n_text_layer, dims.n_text_head, dtype=torch.bool)
+            for layer, head in alignment_heads.tolist():
+                mask[layer, head] = True
+            model.register_buffer("alignment_heads", mask.to_sparse(), persistent=False)
     return model.to(device)