slexiz/to23oise-tts
1
0
Fork 0
forked from mrq/tortoise-tts
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update api constants

Browse Source
This commit is contained in:
James Betker 2022-04-18 09:22:15 -06:00
parent c52cc78632
commit 713281e376
3 changed files with 9 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
api.py
View File

@ -186,7 +186,9 @@ class TextToSpeech:
'high_quality': Use if you want the absolute best. This is not really worth the compute, though.
"""
# Use generally found best tuning knobs for generation.
kwargs.update({'temperature': .8, 'length_penalty': 1.0, 'repetition_penalty': 2.0, 'top_p': .8,
kwargs.update({'temperature': .8, 'length_penalty': 1.0, 'repetition_penalty': 2.0,
#'typical_sampling': True,
'top_p': .8,
'cond_free_k': 2.0, 'diffusion_temperature': 1.0})
# Presets are defined here.
presets = {
@ -202,7 +204,8 @@ class TextToSpeech:
# autoregressive generation parameters follow
num_autoregressive_samples=512, temperature=.8, length_penalty=1, repetition_penalty=2.0, top_p=.8,
# diffusion generation parameters follow
diffusion_iterations=100, cond_free=True, cond_free_k=2, diffusion_temperature=1.0,):
diffusion_iterations=100, cond_free=True, cond_free_k=2, diffusion_temperature=1.0,
**hf_generate_kwargs):
text = torch.IntTensor(self.tokenizer.encode(text)).unsqueeze(0).cuda()
text = F.pad(text, (0, 1)) # This may not be necessary.
@ -228,7 +231,8 @@ class TextToSpeech:
temperature=temperature,
num_return_sequences=self.autoregressive_batch_size,
length_penalty=length_penalty,
repetition_penalty=repetition_penalty)
repetition_penalty=repetition_penalty,
**hf_generate_kwargs)
padding_needed = self.autoregressive.max_mel_tokens - codes.shape[1]
codes = F.pad(codes, (0, padding_needed), value=stop_mel_token)
samples.append(codes)

6
eval_multiple.py
View File

@ -16,7 +16,7 @@ if __name__ == '__main__':
lines = [l.strip().split('\t') for l in f.readlines()]
tts = TextToSpeech()
for k in range(4):
for k in range(3):
outpath = f'{outpath_base}_{k}'
os.makedirs(outpath, exist_ok=True)
recorder = open(os.path.join(outpath, 'transcript.tsv'), 'w', encoding='utf-8')
@ -27,9 +27,7 @@ if __name__ == '__main__':
path = os.path.join(os.path.dirname(fname), line[1])
cond_audio = load_audio(path, 22050)
torchaudio.save(os.path.join(outpath_real, os.path.basename(line[1])), cond_audio, 22050)
sample = tts.tts(transcript, [cond_audio, cond_audio], num_autoregressive_samples=128, k=1,
repetition_penalty=2.0, length_penalty=2, temperature=.5, top_p=.5,
diffusion_temperature=.7, cond_free_k=2, diffusion_iterations=70)
sample = tts.tts_with_preset(transcript, [cond_audio, cond_audio], preset='standard')
down = torchaudio.functional.resample(sample, 24000, 22050)
fout_path = os.path.join(outpath, os.path.basename(line[1]))

0
models/cvvp.py Normal file
View File