chrissewell/whisperX
1
0
Fork 0
mirror of https://github.com/m-bain/whisperX.git synced 2025-07-01 18:17:27 -04:00
Code Issues Packages Projects Releases Wiki Activity

Add torch compile to log mel spectrogram

Browse Source
This commit is contained in:
Simon
2023-05-03 23:17:44 +02:00
parent 2a6830492c
commit 2d59eb9726
2 changed files with 19 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
whisperx/asr.py
View File

@ -181,6 +181,9 @@ class FasterWhisperPipeline(Pipeline):
def preprocess(self, audio):
audio = audio['inputs']
if isinstance(audio, np.ndarray):
audio = torch.from_numpy(audio)
features = log_mel_spectrogram(audio, padding=N_SAMPLES - audio.shape[0])
return {'inputs': features}
@ -253,7 +256,7 @@ class FasterWhisperPipeline(Pipeline):
def detect_language(self, audio: np.ndarray):
if audio.shape[0] < N_SAMPLES:
print("Warning: audio is shorter than 30s, language detection may be inaccurate.")
segment = log_mel_spectrogram(audio[: N_SAMPLES],
segment = log_mel_spectrogram(torch.from_numpy(audio[:N_SAMPLES]),
padding=0 if audio.shape[0] >= N_SAMPLES else N_SAMPLES - audio.shape[0])
encoder_output = self.model.encode(segment)
results = self.model.model.detect_language(encoder_output)