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

cleaned up brain worms with wrapping around gradio progress by instead just using tqdm directly (slight regressions with some messages not getting pushed)

Browse Source
This commit is contained in:
mrq 2023-05-04 23:40:33 +00:00
parent 09d849a78f
commit 5003bc89d3
3 changed files with 35 additions and 43 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
modules/tortoise-tts

@ -1 +1 @@
Subproject commit 086aad5b49e7ef39d043c6b0e12ac33c20773ab8 Subproject commit c90ee7c5296992ad96c8790b5b7cc3737062e1e6

68
src/utils.py
View File

@ -238,7 +238,7 @@ def generate_bark(**kwargs):
if tts_loading: if tts_loading:
raise Exception("TTS is still initializing...") raise Exception("TTS is still initializing...")
if progress is not None: if progress is not None:
progress(0, "Initializing TTS...") notify_progress("Initializing TTS...", progress=progress)
load_tts() load_tts()
if hasattr(tts, "loading") and tts.loading: if hasattr(tts, "loading") and tts.loading:
raise Exception("TTS is still initializing...") raise Exception("TTS is still initializing...")
@ -339,8 +339,8 @@ def generate_bark(**kwargs):
INFERENCING = True INFERENCING = True
for line, cut_text in enumerate(texts): for line, cut_text in enumerate(texts):
progress.msg_prefix = f'[{str(line+1)}/{str(len(texts))}]' tqdm_prefix = f'[{str(line+1)}/{str(len(texts))}]'
print(f"{progress.msg_prefix} Generating line: {cut_text}") print(f"{tqdm_prefix} Generating line: {cut_text}")
start_time = time.time() start_time = time.time()
# do setting editing # do setting editing
@ -422,12 +422,12 @@ def generate_bark(**kwargs):
if args.voice_fixer: if args.voice_fixer:
if not voicefixer: if not voicefixer:
progress(0, "Loading voicefix...") notify_progress("Loading voicefix...", progress=progress)
load_voicefixer() load_voicefixer()
try: try:
fixed_cache = {} fixed_cache = {}
for name in progress.tqdm(audio_cache, desc="Running voicefix..."): for name in tqdm(audio_cache, desc="Running voicefix..."):
del audio_cache[name]['audio'] del audio_cache[name]['audio']
if 'output' not in audio_cache[name] or not audio_cache[name]['output']: if 'output' not in audio_cache[name] or not audio_cache[name]['output']:
continue continue
@ -467,7 +467,7 @@ def generate_bark(**kwargs):
f.write(json.dumps(audio_cache[name]['settings'], indent='\t') ) f.write(json.dumps(audio_cache[name]['settings'], indent='\t') )
if args.embed_output_metadata: if args.embed_output_metadata:
for name in progress.tqdm(audio_cache, desc="Embedding metadata..."): for name in tqdm(audio_cache, desc="Embedding metadata..."):
if 'pruned' in audio_cache[name] and audio_cache[name]['pruned']: if 'pruned' in audio_cache[name] and audio_cache[name]['pruned']:
continue continue
@ -521,7 +521,7 @@ def generate_valle(**kwargs):
if tts_loading: if tts_loading:
raise Exception("TTS is still initializing...") raise Exception("TTS is still initializing...")
if progress is not None: if progress is not None:
progress(0, "Initializing TTS...") notify_progress("Initializing TTS...", progress=progress)
load_tts() load_tts()
if hasattr(tts, "loading") and tts.loading: if hasattr(tts, "loading") and tts.loading:
raise Exception("TTS is still initializing...") raise Exception("TTS is still initializing...")
@ -630,8 +630,8 @@ def generate_valle(**kwargs):
INFERENCING = True INFERENCING = True
for line, cut_text in enumerate(texts): for line, cut_text in enumerate(texts):
progress.msg_prefix = f'[{str(line+1)}/{str(len(texts))}]' tqdm_prefix = f'[{str(line+1)}/{str(len(texts))}]'
print(f"{progress.msg_prefix} Generating line: {cut_text}") print(f"{tqdm_prefix} Generating line: {cut_text}")
start_time = time.time() start_time = time.time()
# do setting editing # do setting editing
@ -715,12 +715,12 @@ def generate_valle(**kwargs):
if args.voice_fixer: if args.voice_fixer:
if not voicefixer: if not voicefixer:
progress(0, "Loading voicefix...") notify_progress("Loading voicefix...", progress=progress)
load_voicefixer() load_voicefixer()
try: try:
fixed_cache = {} fixed_cache = {}
for name in progress.tqdm(audio_cache, desc="Running voicefix..."): for name in tqdm(audio_cache, desc="Running voicefix..."):
del audio_cache[name]['audio'] del audio_cache[name]['audio']
if 'output' not in audio_cache[name] or not audio_cache[name]['output']: if 'output' not in audio_cache[name] or not audio_cache[name]['output']:
continue continue
@ -760,7 +760,7 @@ def generate_valle(**kwargs):
f.write(json.dumps(audio_cache[name]['settings'], indent='\t') ) f.write(json.dumps(audio_cache[name]['settings'], indent='\t') )
if args.embed_output_metadata: if args.embed_output_metadata:
for name in progress.tqdm(audio_cache, desc="Embedding metadata..."): for name in tqdm(audio_cache, desc="Embedding metadata..."):
if 'pruned' in audio_cache[name] and audio_cache[name]['pruned']: if 'pruned' in audio_cache[name] and audio_cache[name]['pruned']:
continue continue
@ -839,7 +839,7 @@ def generate_tortoise(**kwargs):
voice_samples, conditioning_latents = None, tts.get_random_conditioning_latents() voice_samples, conditioning_latents = None, tts.get_random_conditioning_latents()
else: else:
if progress is not None: if progress is not None:
progress(0, desc=f"Loading voice: {voice}") notify_progress(f"Loading voice: {voice}", progress=progress)
voice_samples, conditioning_latents = load_voice(voice, model_hash=tts.autoregressive_model_hash) voice_samples, conditioning_latents = load_voice(voice, model_hash=tts.autoregressive_model_hash)
@ -1032,8 +1032,8 @@ def generate_tortoise(**kwargs):
elif parameters['emotion'] != "None" and parameters['emotion']: elif parameters['emotion'] != "None" and parameters['emotion']:
cut_text = f"[I am really {parameters['emotion'].lower()},] {cut_text}" cut_text = f"[I am really {parameters['emotion'].lower()},] {cut_text}"
progress.msg_prefix = f'[{str(line+1)}/{str(len(texts))}]' tqdm_prefix = f'[{str(line+1)}/{str(len(texts))}]'
print(f"{progress.msg_prefix} Generating line: {cut_text}") print(f"{tqdm_prefix} Generating line: {cut_text}")
start_time = time.time() start_time = time.time()
# do setting editing # do setting editing
@ -1115,12 +1115,12 @@ def generate_tortoise(**kwargs):
if args.voice_fixer: if args.voice_fixer:
if not voicefixer: if not voicefixer:
progress(0, "Loading voicefix...") notify_progress("Loading voicefix...", progress=progress)
load_voicefixer() load_voicefixer()
try: try:
fixed_cache = {} fixed_cache = {}
for name in progress.tqdm(audio_cache, desc="Running voicefix..."): for name in tqdm(audio_cache, desc="Running voicefix..."):
del audio_cache[name]['audio'] del audio_cache[name]['audio']
if 'output' not in audio_cache[name] or not audio_cache[name]['output']: if 'output' not in audio_cache[name] or not audio_cache[name]['output']:
continue continue
@ -1160,7 +1160,7 @@ def generate_tortoise(**kwargs):
f.write(json.dumps(audio_cache[name]['settings'], indent='\t') ) f.write(json.dumps(audio_cache[name]['settings'], indent='\t') )
if args.embed_output_metadata: if args.embed_output_metadata:
for name in progress.tqdm(audio_cache, desc="Embedding metadata..."): for name in tqdm(audio_cache, desc="Embedding metadata..."):
if 'pruned' in audio_cache[name] and audio_cache[name]['pruned']: if 'pruned' in audio_cache[name] and audio_cache[name]['pruned']:
continue continue
@ -1309,7 +1309,7 @@ def compute_latents(voice=None, voice_samples=None, voice_latents_chunks=0, prog
if voice_samples is None: if voice_samples is None:
return return
conditioning_latents = tts.get_conditioning_latents(voice_samples, return_mels=not args.latents_lean_and_mean, slices=voice_latents_chunks, force_cpu=args.force_cpu_for_conditioning_latents, progress=progress) conditioning_latents = tts.get_conditioning_latents(voice_samples, return_mels=not args.latents_lean_and_mean, slices=voice_latents_chunks, force_cpu=args.force_cpu_for_conditioning_latents)
if len(conditioning_latents) == 4: if len(conditioning_latents) == 4:
conditioning_latents = (conditioning_latents[0], conditioning_latents[1], conditioning_latents[2], None) conditioning_latents = (conditioning_latents[0], conditioning_latents[1], conditioning_latents[2], None)
@ -2117,7 +2117,7 @@ def transcribe_dataset( voice, language=None, skip_existings=False, progress=Non
if os.path.exists(infile): if os.path.exists(infile):
results = json.load(open(infile, 'r', encoding="utf-8")) results = json.load(open(infile, 'r', encoding="utf-8"))
for file in enumerate_progress(files, desc="Iterating through voice files", progress=progress): for file in tqdm(files, desc="Iterating through voice files"):
basename = os.path.basename(file) basename = os.path.basename(file)
if basename in results and skip_existings: if basename in results and skip_existings:
@ -2246,7 +2246,7 @@ def phonemize_txt_file( path ):
reparsed = [] reparsed = []
with open(path.replace(".txt", ".phn.txt"), 'a', encoding='utf-8') as f: with open(path.replace(".txt", ".phn.txt"), 'a', encoding='utf-8') as f:
for line in enumerate_progress(lines, desc='Phonemizing...'): for line in tqdm(lines, desc='Phonemizing...'):
split = line.split("|") split = line.split("|")
audio = split[0] audio = split[0]
text = split[2] text = split[2]
@ -2357,7 +2357,7 @@ def prepare_dataset( voice, use_segments=False, text_length=0, audio_length=0, p
text_length = 0 text_length = 0
audio_length = 0 audio_length = 0
for filename in enumerate_progress(results, desc="Parsing results", progress=progress): for filename in tqdm(results, desc="Parsing results"):
use_segment = use_segments use_segment = use_segments
result = results[filename] result = results[filename]
@ -2438,7 +2438,7 @@ def prepare_dataset( voice, use_segments=False, text_length=0, audio_length=0, p
'phonemize': [[], []], 'phonemize': [[], []],
} }
for file in enumerate_progress(segments, desc="Parsing segments", progress=progress): for file in tqdm(segments, desc="Parsing segments"):
result = segments[file] result = segments[file]
path = f'{indir}/audio/{file}' path = f'{indir}/audio/{file}'
@ -2511,7 +2511,7 @@ def prepare_dataset( voice, use_segments=False, text_length=0, audio_length=0, p
print("Phonemized:", file, normalized, text) print("Phonemized:", file, normalized, text)
""" """
for i in enumerate_progress(range(len(jobs['quantize'][0])), desc="Quantizing", progress=progress): for i in tqdm(range(len(jobs['quantize'][0])), desc="Quantizing"):
qnt_file = jobs['quantize'][0][i] qnt_file = jobs['quantize'][0][i]
waveform, sample_rate = jobs['quantize'][1][i] waveform, sample_rate = jobs['quantize'][1][i]
@ -2519,7 +2519,7 @@ def prepare_dataset( voice, use_segments=False, text_length=0, audio_length=0, p
torch.save(quantized, qnt_file) torch.save(quantized, qnt_file)
print("Quantized:", qnt_file) print("Quantized:", qnt_file)
for i in enumerate_progress(range(len(jobs['phonemize'][0])), desc="Phonemizing", progress=progress): for i in tqdm(range(len(jobs['phonemize'][0])), desc="Phonemizing"):
phn_file = jobs['phonemize'][0][i] phn_file = jobs['phonemize'][0][i]
normalized = jobs['phonemize'][1][i] normalized = jobs['phonemize'][1][i]
@ -2807,7 +2807,7 @@ def import_voices(files, saveAs=None, progress=None):
if not isinstance(files, list): if not isinstance(files, list):
files = [files] files = [files]
for file in enumerate_progress(files, desc="Importing voice files", progress=progress): for file in tqdm(files, desc="Importing voice files"):
j, latents = read_generate_settings(file, read_latents=True) j, latents = read_generate_settings(file, read_latents=True)
if j is not None and saveAs is None: if j is not None and saveAs is None:
@ -3025,21 +3025,13 @@ def check_for_updates( dir = None ):
return False return False
def enumerate_progress(iterable, desc=None, progress=None, verbose=None):
if verbose and desc is not None:
print(desc)
if progress is None:
return tqdm(iterable, disable=False) #not verbose)
return progress.tqdm(iterable, desc=f'{progress.msg_prefix} {desc}' if hasattr(progress, 'msg_prefix') else desc)
def notify_progress(message, progress=None, verbose=True): def notify_progress(message, progress=None, verbose=True):
if verbose: if verbose:
print(message) print(message)
if progress is None: if progress is None:
return tqdm.write( desc=message)
else:
progress(0, desc=message) progress(0, desc=message)
def get_args(): def get_args():
@ -3650,7 +3642,7 @@ def load_whisper_model(language=None, model_name=None, progress=None):
model_name = f'{model_name}.{language}' model_name = f'{model_name}.{language}'
print(f"Loading specialized model for language: {language}") print(f"Loading specialized model for language: {language}")
notify_progress(f"Loading Whisper model: {model_name}", progress) notify_progress(f"Loading Whisper model: {model_name}", progress=progress)
if args.whisper_backend == "openai/whisper": if args.whisper_backend == "openai/whisper":
import whisper import whisper
@ -3733,7 +3725,7 @@ def merge_models( primary_model_name, secondary_model_name, alpha, progress=gr.P
theta_0 = read_model(primary_model_name) theta_0 = read_model(primary_model_name)
theta_1 = read_model(secondary_model_name) theta_1 = read_model(secondary_model_name)
for key in enumerate_progress(theta_0.keys(), desc="Merging...", progress=progress): for key in tqdm(theta_0.keys(), desc="Merging..."):
if key in key_blacklist: if key in key_blacklist:
print("Skipping ignored key:", key) print("Skipping ignored key:", key)
continue continue

6
src/webui.py
View File

@ -200,7 +200,7 @@ def read_generate_settings_proxy(file, saveAs='.temp'):
def slice_dataset_proxy( voice, trim_silence, start_offset, end_offset, progress=gr.Progress(track_tqdm=True) ): def slice_dataset_proxy( voice, trim_silence, start_offset, end_offset, progress=gr.Progress(track_tqdm=True) ):
return slice_dataset( voice, trim_silence=trim_silence, start_offset=start_offset, end_offset=end_offset, results=None, progress=progress ) return slice_dataset( voice, trim_silence=trim_silence, start_offset=start_offset, end_offset=end_offset, results=None, progress=progress )
def diarize_dataset( voice, progress=gr.Progress(track_tqdm=False) ): def diarize_dataset( voice, progress=gr.Progress(track_tqdm=True) ):
from pyannote.audio import Pipeline from pyannote.audio import Pipeline
pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization", use_auth_token=args.hf_token) pipeline = Pipeline.from_pretrained("pyannote/speaker-diarization", use_auth_token=args.hf_token)
@ -215,7 +215,7 @@ def diarize_dataset( voice, progress=gr.Progress(track_tqdm=False) ):
return "\n".join(messages) return "\n".join(messages)
def prepare_all_datasets( language, validation_text_length, validation_audio_length, skip_existings, slice_audio, trim_silence, slice_start_offset, slice_end_offset, progress=gr.Progress(track_tqdm=False) ): def prepare_all_datasets( language, validation_text_length, validation_audio_length, skip_existings, slice_audio, trim_silence, slice_start_offset, slice_end_offset, progress=gr.Progress(track_tqdm=True) ):
kwargs = locals() kwargs = locals()
messages = [] messages = []
@ -239,7 +239,7 @@ def prepare_all_datasets( language, validation_text_length, validation_audio_len
return "\n".join(messages) return "\n".join(messages)
def prepare_dataset_proxy( voice, language, validation_text_length, validation_audio_length, skip_existings, slice_audio, trim_silence, slice_start_offset, slice_end_offset, progress=gr.Progress(track_tqdm=False) ): def prepare_dataset_proxy( voice, language, validation_text_length, validation_audio_length, skip_existings, slice_audio, trim_silence, slice_start_offset, slice_end_offset, progress=gr.Progress(track_tqdm=True) ):
messages = [] messages = []
message = transcribe_dataset( voice=voice, language=language, skip_existings=skip_existings, progress=progress ) message = transcribe_dataset( voice=voice, language=language, skip_existings=skip_existings, progress=progress )