Open iAladeen opened 1 year ago
Yes I noticed this
8 months later....
There's a bug in line 176 in whisperx/transcribe.py
Patch:
diff --git a/whisperx/transcribe.py b/whisperx/transcribe.py
index edd2764..818179a 100644
--- a/whisperx/transcribe.py
+++ b/whisperx/transcribe.py
@@ -173,7 +173,7 @@ def cli():
audio = load_audio(audio_path)
# >> VAD & ASR
print(">>Performing transcription...")
- result = model.transcribe(audio, batch_size=batch_size, chunk_size=chunk_size, print_progress=print_progress)
+ result = model.transcribe(audio, batch_size=batch_size, chunk_size=chunk_size, print_progress=print_progress, task=task)
results.append((result, audio_path))
# Unload Whisper and VAD
Without it, task only gets set if the language is passed as a command line option. (transcribe.py:170 -> asr.py:294)
So after the new update --task translate command dosen't seem to be translating the text. It just gives the output in the same language as original. It also dosen't align the text with the audio, just gives text in short paragraphs