Jarod/ai-voice-cloning
1
1
Fork 0
forked from mrq/ai-voice-cloning
Code Pull Requests Activity

do not write current whisper.json if there's no changes

Browse Source
This commit is contained in:
mrq 2023-03-22 22:24:07 +00:00
parent 932eaccdf5
commit 19c0854e6a
2 changed files with 9 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
src/utils.py
View File

@ -1271,11 +1271,13 @@ def transcribe_dataset( voice, language=None, skip_existings=False, progress=Non
for file in enumerate_progress(files, desc="Iterating through voice files", progress=progress): for file in enumerate_progress(files, desc="Iterating through voice files", progress=progress):
basename = os.path.basename(file) basename = os.path.basename(file)
modified = False
if basename in results and skip_existings: if basename in results and skip_existings:
print(f"Skipping already parsed file: {basename}") print(f"Skipping already parsed file: {basename}")
else: else:
try: try:
result = whisper_transcribe(file, language=language) result = whisper_transcribe(file, language=language)
modified = True
except Exception as e: except Exception as e:
print("Failed to transcribe:", file) print("Failed to transcribe:", file)
continue continue
@ -1283,8 +1285,9 @@ def transcribe_dataset( voice, language=None, skip_existings=False, progress=Non
try: try:
sanitized = whisper_sanitize(results[basename]) sanitized = whisper_sanitize(results[basename])
if len(sanitized['segments']) > 0 and len(sanitized['segments'] != results[basename]['segments']): if len(sanitized['segments']) > 0 and len(sanitized['segments']) != len(results[basename]['segments']):
results[basename] = sanitized results[basename] = sanitized
modified = True
print("Segments sanizited: ", basename) print("Segments sanizited: ", basename)
except Exception as e: except Exception as e:
print("Failed to sanitize:", basename, e) print("Failed to sanitize:", basename, e)
@ -1298,6 +1301,7 @@ def transcribe_dataset( voice, language=None, skip_existings=False, progress=Non
waveform = waveform[:1] waveform = waveform[:1]
torchaudio.save(f"{indir}/audio/{basename}", waveform, sample_rate, encoding="PCM_S", bits_per_sample=16) torchaudio.save(f"{indir}/audio/{basename}", waveform, sample_rate, encoding="PCM_S", bits_per_sample=16)
if modified:
with open(infile, 'w', encoding="utf-8") as f: with open(infile, 'w', encoding="utf-8") as f:
f.write(json.dumps(results, indent='\t')) f.write(json.dumps(results, indent='\t'))

4
src/webui.py
View File

@ -443,8 +443,8 @@ def setup_gradio():
DATASET_SETTINGS['voice'] = gr.Dropdown( choices=voice_list, label="Dataset Source", type="value", value=voice_list[0] if len(voice_list) > 0 else "" ) DATASET_SETTINGS['voice'] = gr.Dropdown( choices=voice_list, label="Dataset Source", type="value", value=voice_list[0] if len(voice_list) > 0 else "" )
with gr.Row(): with gr.Row():
DATASET_SETTINGS['language'] = gr.Textbox(label="Language", value="en") DATASET_SETTINGS['language'] = gr.Textbox(label="Language", value="en")
DATASET_SETTINGS['validation_text_length'] = gr.Number(label="Validation Text Length Threshold", value=12, precision=0) DATASET_SETTINGS['validation_text_length'] = gr.Number(label="Validation Text Length Threshold", value=12, precision=0, visible=args.tts_backend=="tortoise")
DATASET_SETTINGS['validation_audio_length'] = gr.Number(label="Validation Audio Length Threshold", value=1 ) DATASET_SETTINGS['validation_audio_length'] = gr.Number(label="Validation Audio Length Threshold", value=1, visible=args.tts_backend=="tortoise" )
with gr.Row(): with gr.Row():
DATASET_SETTINGS['skip'] = gr.Checkbox(label="Skip Existing", value=False) DATASET_SETTINGS['skip'] = gr.Checkbox(label="Skip Existing", value=False)
DATASET_SETTINGS['slice'] = gr.Checkbox(label="Slice Segments", value=False) DATASET_SETTINGS['slice'] = gr.Checkbox(label="Slice Segments", value=False)