lfeats 0.2.1__tar.gz → 0.2.2__tar.gz

{lfeats-0.2.1 → lfeats-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: lfeats
-Version: 0.2.1
+Version: 0.2.2
 Summary: A unified interface to extract hidden representations from speech foundation models
 Project-URL: Documentation, https://takenori-y.github.io/lfeats/stable/
 Project-URL: Source, https://github.com/takenori-y/lfeats
@@ -33,6 +33,7 @@ Requires-Dist: soundfile>=0.10.2
 Requires-Dist: soxr>=0.4.0
 Requires-Dist: torch>=2.6.0
 Requires-Dist: torchaudio>=2.6.0
+Requires-Dist: tqdm>=4.63.0
 Requires-Dist: transformers>=5.3.0
 Provides-Extra: dev
 Requires-Dist: build; extra == 'dev'
@@ -41,7 +42,7 @@ Requires-Dist: mdformat; extra == 'dev'
 Requires-Dist: numpydoc; extra == 'dev'
 Requires-Dist: pkginfo; extra == 'dev'
 Requires-Dist: pydata-sphinx-theme; extra == 'dev'
-Requires-Dist: pyright<=1.1.408; extra == 'dev'
+Requires-Dist: pyright[nodejs]; extra == 'dev'
 Requires-Dist: pytest; extra == 'dev'
 Requires-Dist: pytest-cov; extra == 'dev'
 Requires-Dist: ruff; extra == 'dev'
@@ -148,25 +149,29 @@ pip install git+https://github.com/takenori-y/lfeats.git@master
 
 | Model Name | Model Variant | Hop Size [ms] | Dimension | Paper | Source | Model Hub |
 | :--- | :--- | ---: | ---: | :---: | :---: | :---: |
-| `higgs-audio` | `v2` | 40 | 1024 | | [GitHub](https://github.com/boson-ai/higgs-audio) | [🤗](https://huggingface.co/eustlb/higgs-audio-v2-tokenizer) |
+| `higgs-audio` | `v2` | 40 | 1024 | [Blog](https://github.com/Render-AI-Code/cog-higgs-audio/blob/main/tech_blogs/TOKENIZER_BLOG.md) | [GitHub](https://github.com/boson-ai/higgs-audio) | [🤗](https://huggingface.co/eustlb/higgs-audio-v2-tokenizer) |
+| `x-codec` | `hubert` | 20 | 1024 | [arXiv](https://arxiv.org/abs/2408.17175) | [GitHub](https://github.com/zhenye234/xcodec) | [🤗](https://huggingface.co/hf-audio/xcodec-hubert-librispeech) |
+| | `wavlm` | 20 | 1024 | | | [🤗](https://huggingface.co/hf-audio/xcodec-wavlm-more-data) |
 
 ### Utterance-Level Features
 
-| Model Name | Model Variant | Layers | Dimension | Paper | Source | Model Hub |
-| :--- | :--- | ---: | ---: | :---: | :---: | :---: |
-| `ecapa-tdnn` | `base` | 0 | 192 | [arXiv](https://arxiv.org/abs/2005.07143) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb) |
-| `next-tdnn` | `light` | 0 | 192 | [arXiv](https://arxiv.org/abs/2312.08603) | [GitHub](https://github.com/dmlguq456/NeXt_TDNN_ASV) | |
-| | `base` | 0 | 192 | | | |
-| | `base-v2` | 0 | 192 | | | |
-| `r-vector` | `base` | 0 | 256 | [arXiv](https://arxiv.org/abs/1910.12592) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-resnet-voxceleb) |
-| `redimnet` | `b0` | 0 | 192 | [arXiv](https://arxiv.org/abs/2407.18223) | [GitHub](https://github.com/IDRnD/redimnet) | |
-| | `b1` | 0 | 192 | | | |
-| | `b2` | 0 | 192 | | | |
-| | `b3` | 0 | 192 | | | |
-| | `b4` | 0 | 192 | | | |
-| | `b5` | 0 | 192 | | | |
-| | `b6` | 0 | 192 | | | |
-| `x-vector` | `base` | 0 | 512 | [IEEE](https://ieeexplore.ieee.org/document/8461375) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-xvect-voxceleb) |
+| Model Name | Model Variant | Dimension | Paper | Source | Model Hub |
+| :--- | :--- | ---: | :---: | :---: | :---: |
+| `ecapa-tdnn` | `base` | 192 | [arXiv](https://arxiv.org/abs/2005.07143) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb) |
+| `next-tdnn` | `light` | 192 | [arXiv](https://arxiv.org/abs/2312.08603) | [GitHub](https://github.com/dmlguq456/NeXt_TDNN_ASV) | |
+| | `base` | 192 | | | |
+| | `base-v2` | 192 | | | |
+| `r-vector` | `base` | 256 | [arXiv](https://arxiv.org/abs/1910.12592) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-resnet-voxceleb) |
+| `redimnet` | `b0` | 192 | [arXiv](https://arxiv.org/abs/2407.18223) | [GitHub](https://github.com/IDRnD/redimnet) | |
+| | `b1` | 192 | | | |
+| | `b2` | 192 | | | |
+| | `b3` | 192 | | | |
+| | `b4` | 192 | | | |
+| | `b5` | 192 | | | |
+| | `b6` | 192 | | | |
+| `wavlm-sv` | `base` | 512 | [arXiv](https://arxiv.org/abs/2110.13900) | [GitHub](https://github.com/microsoft/unilm/tree/master/wavlm) | [🤗](https://huggingface.co/microsoft/wavlm-base-sv) |
+| | `base+` | 512 | | | [🤗](https://huggingface.co/microsoft/wavlm-base-plus-sv) |
+| `x-vector` | `base` | 512 | [IEEE](https://ieeexplore.ieee.org/document/8461375) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-xvect-voxceleb) |
 
 > [!IMPORTANT]
 > Users must comply with the respective licenses of the models.

{lfeats-0.2.1 → lfeats-0.2.2}/README.md

@@ -97,25 +97,29 @@ pip install git+https://github.com/takenori-y/lfeats.git@master
 
 | Model Name | Model Variant | Hop Size [ms] | Dimension | Paper | Source | Model Hub |
 | :--- | :--- | ---: | ---: | :---: | :---: | :---: |
-| `higgs-audio` | `v2` | 40 | 1024 | | [GitHub](https://github.com/boson-ai/higgs-audio) | [🤗](https://huggingface.co/eustlb/higgs-audio-v2-tokenizer) |
+| `higgs-audio` | `v2` | 40 | 1024 | [Blog](https://github.com/Render-AI-Code/cog-higgs-audio/blob/main/tech_blogs/TOKENIZER_BLOG.md) | [GitHub](https://github.com/boson-ai/higgs-audio) | [🤗](https://huggingface.co/eustlb/higgs-audio-v2-tokenizer) |
+| `x-codec` | `hubert` | 20 | 1024 | [arXiv](https://arxiv.org/abs/2408.17175) | [GitHub](https://github.com/zhenye234/xcodec) | [🤗](https://huggingface.co/hf-audio/xcodec-hubert-librispeech) |
+| | `wavlm` | 20 | 1024 | | | [🤗](https://huggingface.co/hf-audio/xcodec-wavlm-more-data) |
 
 ### Utterance-Level Features
 
-| Model Name | Model Variant | Layers | Dimension | Paper | Source | Model Hub |
-| :--- | :--- | ---: | ---: | :---: | :---: | :---: |
-| `ecapa-tdnn` | `base` | 0 | 192 | [arXiv](https://arxiv.org/abs/2005.07143) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb) |
-| `next-tdnn` | `light` | 0 | 192 | [arXiv](https://arxiv.org/abs/2312.08603) | [GitHub](https://github.com/dmlguq456/NeXt_TDNN_ASV) | |
-| | `base` | 0 | 192 | | | |
-| | `base-v2` | 0 | 192 | | | |
-| `r-vector` | `base` | 0 | 256 | [arXiv](https://arxiv.org/abs/1910.12592) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-resnet-voxceleb) |
-| `redimnet` | `b0` | 0 | 192 | [arXiv](https://arxiv.org/abs/2407.18223) | [GitHub](https://github.com/IDRnD/redimnet) | |
-| | `b1` | 0 | 192 | | | |
-| | `b2` | 0 | 192 | | | |
-| | `b3` | 0 | 192 | | | |
-| | `b4` | 0 | 192 | | | |
-| | `b5` | 0 | 192 | | | |
-| | `b6` | 0 | 192 | | | |
-| `x-vector` | `base` | 0 | 512 | [IEEE](https://ieeexplore.ieee.org/document/8461375) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-xvect-voxceleb) |
+| Model Name | Model Variant | Dimension | Paper | Source | Model Hub |
+| :--- | :--- | ---: | :---: | :---: | :---: |
+| `ecapa-tdnn` | `base` | 192 | [arXiv](https://arxiv.org/abs/2005.07143) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-ecapa-voxceleb) |
+| `next-tdnn` | `light` | 192 | [arXiv](https://arxiv.org/abs/2312.08603) | [GitHub](https://github.com/dmlguq456/NeXt_TDNN_ASV) | |
+| | `base` | 192 | | | |
+| | `base-v2` | 192 | | | |
+| `r-vector` | `base` | 256 | [arXiv](https://arxiv.org/abs/1910.12592) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-resnet-voxceleb) |
+| `redimnet` | `b0` | 192 | [arXiv](https://arxiv.org/abs/2407.18223) | [GitHub](https://github.com/IDRnD/redimnet) | |
+| | `b1` | 192 | | | |
+| | `b2` | 192 | | | |
+| | `b3` | 192 | | | |
+| | `b4` | 192 | | | |
+| | `b5` | 192 | | | |
+| | `b6` | 192 | | | |
+| `wavlm-sv` | `base` | 512 | [arXiv](https://arxiv.org/abs/2110.13900) | [GitHub](https://github.com/microsoft/unilm/tree/master/wavlm) | [🤗](https://huggingface.co/microsoft/wavlm-base-sv) |
+| | `base+` | 512 | | | [🤗](https://huggingface.co/microsoft/wavlm-base-plus-sv) |
+| `x-vector` | `base` | 512 | [IEEE](https://ieeexplore.ieee.org/document/8461375) | [GitHub](https://github.com/speechbrain/speechbrain) | [🤗](https://huggingface.co/speechbrain/spkrec-xvect-voxceleb) |
 
 > [!IMPORTANT]
 > Users must comply with the respective licenses of the models.

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/cli.py

@@ -12,6 +12,8 @@ import os
 import sys
 from pathlib import Path
 
+from tqdm import tqdm
+
 logger = logging.getLogger("lfeats")
 
 
@@ -125,6 +127,12 @@ def get_arguments() -> argparse.Namespace:
             "'mean', or 'auto'."
         ),
     )
+    parser.add_argument(
+        "--normalize",
+        action=argparse.BooleanOptionalAction,
+        default=False,
+        help="Enable or disable L2 normalization of the extracted features.",
+    )
     parser.add_argument(
         "-q",
         "--quiet",
@@ -163,6 +171,7 @@ def main() -> None:
         for root, _, files in os.walk(args.source):
             for file in files:
                 input_files.append(os.path.join(root, file))
+        input_files = sorted(input_files)
     else:
         raise ValueError(f"Invalid source: {args.source}")
 
@@ -197,12 +206,15 @@ def main() -> None:
         device=args.device,
         cache_dir=args.cache_dir,
     )
+    logger.info(f"Loading {extractor.get_model_id()} model...")
     extractor.load(quiet=args.quiet)
 
     num_errors = 0
 
     # Process each input file and extract features.
-    for input_file in input_files:
+    for input_file in tqdm(
+        input_files, desc="Processing files", unit="file", disable=args.quiet
+    ):
         if not os.path.isfile(input_file):
             logger.error(f"Could not find file: {input_file}. Skipping.")
             num_errors += 1
@@ -244,6 +256,7 @@ def main() -> None:
                 overlap_length_sec=args.overlap_length_sec,
                 upsample_factor=args.upsample_factor,
                 reduction=args.reduction,
+                normalize=args.normalize,
             )
         except Exception as e:
             logger.error(f"Error processing file {input_file}: {e}. Skipping.")

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/interfaces/extractor.py

@@ -102,6 +102,17 @@ class Extractor:
         self.model_manager.to(device)
         self.resampler_manager.to(device)
 
+    def get_model_id(self) -> str:
+        """Return the identifier of the model being used.
+
+        Returns
+        -------
+        out : str
+            The identifier of the model being used.
+
+        """
+        return self.model_manager.get_model().model_id
+
     def __call__(
         self,
         source: np.ndarray | torch.Tensor | Audio,
@@ -113,6 +124,7 @@ class Extractor:
         overlap_length_sec: int = 5,
         upsample_factor: int = 1,
         reduction: Literal["none", "mean", "auto"] = "auto",
+        normalize: bool = False,
     ) -> Features:
         """Extract features from the input waveform.
 
@@ -148,6 +160,9 @@ class Extractor:
             determined based on the feature granularity ('none' for frame-level features
             and 'mean' for utterance-level features).
 
+        normalize : bool, optional
+            If True, the extracted features will be L2-normalized.
+
         Returns
         -------
         out : Features
@@ -188,6 +203,7 @@ class Extractor:
                 chunk_length_sec=chunk_length_sec,
                 overlap_length_sec=overlap_length_sec,
                 reduction=reduction,
+                normalize=normalize,
             )
 
         model = self.model_manager.get_model()
@@ -243,6 +259,7 @@ class Extractor:
         chunk_length_sec: int = 30,
         overlap_length_sec: int = 5,
         reduction: Literal["none", "mean", "auto"] = "auto",
+        normalize: bool = False,
     ) -> Features:
         """Extract features from the input waveform.
 
@@ -275,6 +292,9 @@ class Extractor:
             determined based on the feature granularity ('none' for frame-level features
             and 'mean' for utterance-level features).
 
+        normalize : bool, optional
+            If True, the extracted features will be L2-normalized.
+
         Returns
         -------
         out : Features
@@ -377,6 +397,10 @@ class Extractor:
         ):
             features = features.reduce("mean")
 
+        # Apply normalization if specified.
+        if normalize:
+            features = features.normalize()
+
         return features
 
     @staticmethod

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/interfaces/types.py

@@ -356,6 +356,26 @@ class Features(Container):
 
         return Features(data=reduced_data, source=self.source, layers=self.layers)
 
+    def normalize(self) -> Features:
+        """Normalize the features in the L2 sense along the feature dimension.
+
+        Returns
+        -------
+        out : Features
+            A new Features instance with normalized data.
+
+        """
+        eps = 1e-12
+        if isinstance(self.data, np.ndarray):
+            norm = np.linalg.norm(self.array, ord=2, axis=-1, keepdims=True)
+            normalized_data = self.array / np.maximum(norm, eps)
+        else:
+            normalized_data = torch.nn.functional.normalize(
+                self.tensor, p=2, dim=-1, eps=eps
+            )
+
+        return Features(data=normalized_data, source=self.source, layers=self.layers)
+
     def concat(self, other: Features) -> Features:
         """Concatenate this Features instance with another one along the time dimension.
 

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/__init__.py

@@ -22,7 +22,9 @@ from .sslzip import SSLZipModel
 from .unispeech_sat import UniSpeechSATModel
 from .wav2vec2 import Wav2Vec2Model
 from .wavlm import WavLMModel
+from .wavlm_sv import WavLMSVModel
 from .whisper import WhisperModel
+from .x_codec import XCodecModel
 from .x_vector import XVectorModel
 
 MODEL_MAP = {
@@ -44,11 +46,13 @@ MODEL_MAP = {
     "unispeech-sat": UniSpeechSATModel,
     "wav2vec2": Wav2Vec2Model,
     "wavlm": WavLMModel,
+    "wavlm-sv": WavLMSVModel,
     "whisper": WhisperModel,
+    "x-codec": XCodecModel,
     "x-vector": XVectorModel,
 }
 
 __all__ = [
-    "ModelManager",
     "MODEL_MAP",
+    "ModelManager",
 ]

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/base.py

@@ -281,7 +281,7 @@ class TokenLevelFeatureModel(BaseModel):
             The frame shift in samples.
 
         """
-        return int(40.0 * self.sample_rate / 1000)
+        return int(20.0 * self.sample_rate / 1000)
 
     @property
     def center_offset(self) -> int:

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/higgs_audio.py

@@ -9,7 +9,7 @@ from typing import Any
 import torch
 
 from ..interfaces.types import Audio, Features
-from ..utils.io import silence_transformers
+from ..utils.io import setup_transformers
 from ..utils.validation import validate_enum
 from .base import TokenLevelFeatureModel
 
@@ -73,7 +73,7 @@ class HiggsAudioTokenizerModel(TokenLevelFeatureModel):
 
         from transformers import AutoFeatureExtractor, HiggsAudioV2TokenizerModel
 
-        with silence_transformers(quiet):
+        with setup_transformers(quiet):
             self.feature_extractor = AutoFeatureExtractor.from_pretrained(
                 self.variant.model_name, cache_dir=model_dir
             )
@@ -123,6 +123,18 @@ class HiggsAudioTokenizerModel(TokenLevelFeatureModel):
 
         return Features(data=vectors, source=self.model_id)
 
+    @property
+    def frame_shift(self) -> int:
+        """Get the frame shift of the model.
+
+        Returns
+        -------
+        out : int
+            The frame shift in samples.
+
+        """
+        return int(40.0 * self.sample_rate / 1000)
+
     @property
     def sample_rate(self) -> int:
         """Get the sample rate required by the model.

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/hubert.py

@@ -8,7 +8,7 @@ from enum import Enum
 import torch
 
 from ..interfaces.types import Audio, Features
-from ..utils.io import silence_transformers
+from ..utils.io import setup_transformers
 from ..utils.validation import validate_enum
 from .base import FrameLevelFeatureModel
 
@@ -33,9 +33,7 @@ class HuBERTVariant(str, Enum):
         base = f"facebook/hubert-{self.value}"
         if self.value == "base":
             return f"{base}-ls960"
-        elif self.value == "large":
-            return f"{base}-ll60k"
-        elif self.value == "xlarge":
+        elif self.value == "large" or self.value == "xlarge":
             return f"{base}-ll60k"
         return base
 
@@ -77,7 +75,7 @@ class HuBERTModel(FrameLevelFeatureModel):
 
         from transformers import HubertModel
 
-        with silence_transformers(quiet):
+        with setup_transformers(quiet):
             self.model = HubertModel.from_pretrained(
                 self.variant.model_name, cache_dir=model_dir
             )

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/unispeech_sat.py

@@ -8,7 +8,7 @@ from enum import Enum
 import torch
 
 from ..interfaces.types import Audio, Features
-from ..utils.io import silence_transformers
+from ..utils.io import setup_transformers
 from ..utils.paths import sanitize
 from ..utils.validation import validate_enum
 from .base import FrameLevelFeatureModel
@@ -73,7 +73,7 @@ class UniSpeechSATModel(FrameLevelFeatureModel):
 
         from transformers import UniSpeechSatForPreTraining as _UniSpeechSATModel
 
-        with silence_transformers(quiet):
+        with setup_transformers(quiet):
             self.model = _UniSpeechSATModel.from_pretrained(
                 self.variant.model_name, cache_dir=model_dir
             )

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/wavlm.py

@@ -8,7 +8,7 @@ from enum import Enum
 import torch
 
 from ..interfaces.types import Audio, Features
-from ..utils.io import silence_transformers
+from ..utils.io import setup_transformers
 from ..utils.paths import sanitize
 from ..utils.validation import validate_enum
 from .base import FrameLevelFeatureModel
@@ -71,7 +71,7 @@ class WavLMModel(FrameLevelFeatureModel):
 
         from transformers import WavLMModel as _WavLMModel
 
-        with silence_transformers(quiet):
+        with setup_transformers(quiet):
             self.model = _WavLMModel.from_pretrained(
                 self.variant.model_name, cache_dir=model_dir
             )

lfeats-0.2.2/lfeats/models/wavlm_sv.py

@@ -0,0 +1,120 @@
+# Copyright (c) 2026 Takenori Yoshimura
+# Released under the MIT License.
+
+"""A module for the WavLM model."""
+
+from enum import Enum
+
+import torch
+
+from ..interfaces.types import Audio, Features
+from ..utils.io import setup_transformers
+from ..utils.paths import sanitize
+from ..utils.validation import validate_enum
+from .base import UtteranceLevelFeatureModel
+
+
+class WavLMSVVariant(str, Enum):
+    """Enumeration of supported WavLM-for-speaker-verification model variants."""
+
+    BASE = "base"
+    BASE_PLUS = "base+"
+
+    @property
+    def model_name(self) -> str:
+        """Return the model name corresponding to the variant.
+
+        Returns
+        -------
+        out : str
+            The model name corresponding to the variant.
+
+        """
+        return f"microsoft/wavlm-{sanitize(self.value)}-sv"
+
+
+class WavLMSVModel(UtteranceLevelFeatureModel):
+    """A class for the WavLM-for-speaker-verification model."""
+
+    def __init__(self, variant: str | None = None, device: str = "cpu") -> None:
+        """Initialize the WavLM-for-speaker-verification model.
+
+        Parameters
+        ----------
+        variant : str | None, optional
+            The variant of the model to use.
+
+        device : str, optional
+            The device to run the model on (e.g., 'cpu' or 'cuda').
+
+        """
+        super().__init__(variant, device)
+
+        self.variant = validate_enum(variant, WavLMSVVariant, WavLMSVVariant.BASE_PLUS)
+        self._model_id = f"wavlm-{self.variant.value}"
+
+        self.feature_extractor = None
+
+    def load(self, model_dir: str, quiet: bool = False) -> None:
+        """Load the model from the specified directory.
+
+        Parameters
+        ----------
+        model_dir : str
+            The directory where the model checkpoint will be stored.
+
+        quiet : bool, optional
+            Whether to suppress output during the loading process.
+
+        """
+        if self.model is not None:
+            return
+
+        from transformers import Wav2Vec2FeatureExtractor, WavLMForXVector
+
+        with setup_transformers(quiet):
+            self.feature_extractor = Wav2Vec2FeatureExtractor.from_pretrained(
+                self.variant.model_name, cache_dir=model_dir
+            )
+            self.model = WavLMForXVector.from_pretrained(
+                self.variant.model_name, cache_dir=model_dir
+            )
+            self.model.eval()
+            self.model.to(self.device)  # type: ignore
+
+    def extract_features_impl(self, audio: Audio, layers: list[int]) -> Features:
+        """Extract features from the input audio using the model.
+
+        Parameters
+        ----------
+        audio : Audio
+            The input audio data with shape (B, T).
+
+        layers : list[int]
+            The layer(s) from which to extract features.
+
+        Returns
+        -------
+        out : Features
+            The extracted features.
+
+        Raises
+        ------
+        RuntimeError
+            If the model is not loaded.
+
+        """
+        if self.feature_extractor is None or self.model is None:
+            raise RuntimeError("Model not loaded. Call 'load' method first.")
+
+        with torch.inference_mode():
+            inputs = self.feature_extractor(
+                [x for x in audio.array],
+                sampling_rate=audio.sample_rate,
+                padding=True,
+                return_tensors="pt",
+            ).to(self.device)
+            embeddings = self.model(**inputs).embeddings
+            embeddings = embeddings.unsqueeze(1)
+
+        return Features(data=embeddings, source=self.model_id)

{lfeats-0.2.1 → lfeats-0.2.2}/lfeats/models/whisper.py

@@ -8,7 +8,7 @@ from enum import Enum
 import torch
 
 from ..interfaces.types import Audio, Features
-from ..utils.io import silence_transformers
+from ..utils.io import setup_transformers
 from ..utils.validation import validate_enum
 from .base import FrameLevelFeatureModel
 
@@ -76,7 +76,7 @@ class WhisperModel(FrameLevelFeatureModel):
 
         from transformers import WhisperModel, WhisperProcessor
 
-        with silence_transformers(quiet):
+        with setup_transformers(quiet):
             self.processor = WhisperProcessor.from_pretrained(
                 self.variant.model_name, cache_dir=model_dir
             )