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

more cleanup

Browse Source
This commit is contained in:
James Betker 2022-05-01 16:28:39 -06:00
parent f7c8decfdb
commit a8264f5cef
3 changed files with 3 additions and 106 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

38
tortoise/eval_multiple.py
View File

@ -1,38 +0,0 @@
import os
import torchaudio
from api import TextToSpeech
from tortoise.utils.audio import load_audio
if __name__ == '__main__':
fname = 'Y:\\clips\\books2\\subset512-oco.tsv'
stop_after = 128
outpath_base = 'D:\\tmp\\tortoise-tts-eval\\audiobooks'
outpath_real = 'D:\\tmp\\tortoise-tts-eval\\real'
os.makedirs(outpath_real, exist_ok=True)
with open(fname, 'r', encoding='utf-8') as f:
lines = [l.strip().split('\t') for l in f.readlines()]
tts = TextToSpeech()
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')
for e, line in enumerate(lines):
if e >= stop_after:
break
transcript = line[0]
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_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]))
torchaudio.save(fout_path, down.squeeze(0), 22050)
recorder.write(f'{transcript}\t{fout_path}\n')
recorder.flush()
recorder.close()

65
tortoise/sweep.py
View File

@ -1,65 +0,0 @@
import os
from random import shuffle
import torchaudio
from api import TextToSpeech
from tortoise.utils.audio import load_audio
def permutations(args):
res = []
k = next(iter(args.keys()))
vals = args[k]
del args[k]
if not args:
return [{k: v} for v in vals]
lower = permutations(args)
for v in vals:
for l in lower:
lc = l.copy()
lc[k] = v
res.append(lc)
return res
if __name__ == '__main__':
fname = 'Y:\\clips\\books2\\subset512-oco.tsv'
stop_after = 512
outpath_base = 'D:\\tmp\\tortoise-tts-eval\\sweep-2'
outpath_real = 'D:\\tmp\\tortoise-tts-eval\\real'
arg_ranges = {
'top_p': [.8,1],
'temperature': [.8,.9,1],
'diffusion_temperature': [.8,1],
'cond_free_k': [1,2,5,10],
}
cfgs = permutations(arg_ranges)
shuffle(cfgs)
for cfg in cfgs:
cfg_desc = '_'.join([f'{k}-{v}' for k,v in cfg.items()])
outpath = os.path.join(outpath_base, f'{cfg_desc}')
os.makedirs(outpath, exist_ok=True)
os.makedirs(outpath_real, exist_ok=True)
with open(fname, 'r', encoding='utf-8') as f:
lines = [l.strip().split('\t') for l in f.readlines()]
recorder = open(os.path.join(outpath, 'transcript.tsv'), 'w', encoding='utf-8')
tts = TextToSpeech()
for e, line in enumerate(lines):
if e >= stop_after:
break
transcript = line[0]
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=32, repetition_penalty=2.0,
k=1, diffusion_iterations=32, length_penalty=1.0, **cfg)
down = torchaudio.functional.resample(sample, 24000, 22050)
fout_path = os.path.join(outpath, os.path.basename(line[1]))
torchaudio.save(fout_path, down.squeeze(0), 22050)
recorder.write(f'{transcript}\t{fout_path}\n')
recorder.flush()
recorder.close()

6
tortoise/samples_generator.py → tortoise/utils/samples_generator.py
View File

@ -4,7 +4,7 @@ import os
if __name__ == '__main__':
result = "<html><head><title>These words were never spoken.</title></head><body><h1>Handpicked results</h1>"
for fv in os.listdir('../results/favorites'):
for fv in os.listdir('../../results/favorites'):
url = f'https://github.com/neonbjb/tortoise-tts/raw/main/results/favorites/{fv}'
result = result + f'<audio controls="" style="width: 600px;"><source src="{url}" type="audio/mp3"></audio><br>\n'
@ -30,7 +30,7 @@ if __name__ == '__main__':
line = line + f'<td><audio controls="" style="width: 150px;"><source src="{url}" type="audio/mp3"></audio></td>'
line = line + "</tr>"
lines.append(line)
for txt in os.listdir('../results/various/'):
for txt in os.listdir('../../results/various/'):
if 'desktop' in txt:
continue
line = f'<tr><td>{txt}</td>'
@ -42,7 +42,7 @@ if __name__ == '__main__':
result = result + '\n'.join(lines) + "</table>"
result = result + "<h1>Longform result for all voices:</h1>"
for lf in os.listdir('../results/riding_hood'):
for lf in os.listdir('../../results/riding_hood'):
url = f'https://github.com/neonbjb/tortoise-tts/raw/main/results/riding_hood/{lf}'
result = result + f'<audio controls="" style="width: 600px;"><source src="{url}" type="audio/mp3"></audio><br>\n'