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

Merge pull request #529 from MahmoudAshraf97/main

Browse Source
This commit is contained in:
Max Bain
2023-10-16 10:53:18 -07:00
committed by GitHub
parent a150df4310 b69956d725
commit 66808f6147
1 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
whisperx/alignment.py
View File

@ -194,8 +194,8 @@ def align(
aligned_segments.append(aligned_seg) aligned_segments.append(aligned_seg)
continue continue
if t1 >= MAX_DURATION or t2 - t1 < 0.02: if t1 >= MAX_DURATION:
print("Failed to align segment: original start time longer than audio duration, skipping...") print(f'Failed to align segment ("{segment["text"]}"): original start time longer than audio duration, skipping...')
aligned_segments.append(aligned_seg) aligned_segments.append(aligned_seg)
continue continue
@ -207,17 +207,17 @@ def align(
# TODO: Probably can get some speedup gain with batched inference here # TODO: Probably can get some speedup gain with batched inference here
waveform_segment = audio[:, f1:f2] waveform_segment = audio[:, f1:f2]
# Handle the minimum input length for wav2vec2 models
if waveform_segment.shape[-1] < 400:
lengths = torch.as_tensor([waveform_segment.shape[-1]]).to(device)
waveform_segment = torch.nn.functional.pad(
waveform_segment, (0, 400 - waveform_segment.shape[-1])
)
else:
lengths = None
with torch.inference_mode(): with torch.inference_mode():
if model_type == "torchaudio": if model_type == "torchaudio":
# Handle the minimum input length for torchaudio wav2vec2 models
if waveform_segment.shape[-1] < 400:
lengths = torch.as_tensor([waveform_segment.shape[-1]]).to(device)
waveform_segment = torch.nn.functional.pad(
waveform_segment, (0, 400 - waveform_segment.shape[-1])
)
else:
lengths = None
emissions, _ = model(waveform_segment.to(device), lengths=lengths) emissions, _ = model(waveform_segment.to(device), lengths=lengths)
elif model_type == "huggingface": elif model_type == "huggingface":
emissions = model(waveform_segment.to(device)).logits emissions = model(waveform_segment.to(device)).logits