Update feature_extraction_msp_audio.py

feature_extraction_msp_audio.py

@@ -2,7 +2,7 @@ from pathlib import Path
 
 import numpy as np
 import torch
-# from torchcodec.decoders import AudioDecoder
+from torchcodec.decoders import AudioDecoder
 from transformers.feature_extraction_sequence_utils import SequenceFeatureExtractor
 from transformers.feature_extraction_utils import BatchFeature
 from transformers.utils import PaddingStrategy, TensorType, logging
@@ -52,20 +52,20 @@ class MSPAudioFeatureExtractor(SequenceFeatureExtractor):
             normed = [(x - x.mean()) / np.sqrt(x.var() + 1e-7) for x in input_values]
         return normed
 
-    # def _load_audio(
-    #     self,
-    #     src: str | Path | bytes | torch.Tensor,
-    #     start_seconds: float = 0.0,
-    #     stop_seconds: float | None = None,
-    # ) -> np.ndarray:
-    #     """Load audio waveform from file path or bytes as a 1-D numpy array."""
-    #     audio_decoder = AudioDecoder(source=src, sample_rate=self.sampling_rate)
-    #     if stop_seconds is None:
-    #         stop_seconds = audio_decoder.metadata.duration_seconds_from_header
-    #     waveform = audio_decoder.get_samples_played_in_range(
-    #         start_seconds, stop_seconds
-    #     ).data.numpy()
-    #     return waveform.squeeze()  # shape: (T,)
+    def _load_audio(
+        self,
+        src: str | Path | bytes | torch.Tensor,
+        start_seconds: float = 0.0,
+        stop_seconds: float | None = None,
+    ) -> np.ndarray:
+        """Load audio waveform from file path or bytes as a 1-D numpy array."""
+        audio_decoder = AudioDecoder(source=src, sample_rate=self.sampling_rate)
+        if stop_seconds is None:
+            stop_seconds = audio_decoder.metadata.duration_seconds_from_header
+        waveform = audio_decoder.get_samples_played_in_range(
+            start_seconds, stop_seconds
+        ).data.numpy()
+        return waveform.squeeze()  # shape: (T,)
 
     def __call__(
         self,
@@ -112,8 +112,8 @@ class MSPAudioFeatureExtractor(SequenceFeatureExtractor):
             raw_speech = [raw_speech]
 
         # Load from file paths or bytes
-        # if isinstance(raw_speech[0], (str, Path, bytes)):
-        #     raw_speech = [self._load_audio(src) for src in raw_speech]
+        if isinstance(raw_speech[0], (str, Path, bytes)):
+            raw_speech = [self._load_audio(src) for src in raw_speech]
 
         encoded = BatchFeature({"input_values": raw_speech})