more cleanup
Browse files
tortoise/eval_multiple.py
DELETED
@@ -1,38 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
|
3 |
-
import torchaudio
|
4 |
-
|
5 |
-
from api import TextToSpeech
|
6 |
-
from tortoise.utils.audio import load_audio
|
7 |
-
|
8 |
-
if __name__ == '__main__':
|
9 |
-
fname = 'Y:\\clips\\books2\\subset512-oco.tsv'
|
10 |
-
stop_after = 128
|
11 |
-
outpath_base = 'D:\\tmp\\tortoise-tts-eval\\audiobooks'
|
12 |
-
outpath_real = 'D:\\tmp\\tortoise-tts-eval\\real'
|
13 |
-
|
14 |
-
os.makedirs(outpath_real, exist_ok=True)
|
15 |
-
with open(fname, 'r', encoding='utf-8') as f:
|
16 |
-
lines = [l.strip().split('\t') for l in f.readlines()]
|
17 |
-
|
18 |
-
tts = TextToSpeech()
|
19 |
-
for k in range(3):
|
20 |
-
outpath = f'{outpath_base}_{k}'
|
21 |
-
os.makedirs(outpath, exist_ok=True)
|
22 |
-
recorder = open(os.path.join(outpath, 'transcript.tsv'), 'w', encoding='utf-8')
|
23 |
-
for e, line in enumerate(lines):
|
24 |
-
if e >= stop_after:
|
25 |
-
break
|
26 |
-
transcript = line[0]
|
27 |
-
path = os.path.join(os.path.dirname(fname), line[1])
|
28 |
-
cond_audio = load_audio(path, 22050)
|
29 |
-
torchaudio.save(os.path.join(outpath_real, os.path.basename(line[1])), cond_audio, 22050)
|
30 |
-
sample = tts.tts_with_preset(transcript, [cond_audio, cond_audio], preset='standard')
|
31 |
-
|
32 |
-
down = torchaudio.functional.resample(sample, 24000, 22050)
|
33 |
-
fout_path = os.path.join(outpath, os.path.basename(line[1]))
|
34 |
-
torchaudio.save(fout_path, down.squeeze(0), 22050)
|
35 |
-
|
36 |
-
recorder.write(f'{transcript}\t{fout_path}\n')
|
37 |
-
recorder.flush()
|
38 |
-
recorder.close()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tortoise/sweep.py
DELETED
@@ -1,65 +0,0 @@
|
|
1 |
-
import os
|
2 |
-
from random import shuffle
|
3 |
-
|
4 |
-
import torchaudio
|
5 |
-
|
6 |
-
from api import TextToSpeech
|
7 |
-
from tortoise.utils.audio import load_audio
|
8 |
-
|
9 |
-
|
10 |
-
def permutations(args):
|
11 |
-
res = []
|
12 |
-
k = next(iter(args.keys()))
|
13 |
-
vals = args[k]
|
14 |
-
del args[k]
|
15 |
-
if not args:
|
16 |
-
return [{k: v} for v in vals]
|
17 |
-
lower = permutations(args)
|
18 |
-
for v in vals:
|
19 |
-
for l in lower:
|
20 |
-
lc = l.copy()
|
21 |
-
lc[k] = v
|
22 |
-
res.append(lc)
|
23 |
-
return res
|
24 |
-
|
25 |
-
|
26 |
-
if __name__ == '__main__':
|
27 |
-
fname = 'Y:\\clips\\books2\\subset512-oco.tsv'
|
28 |
-
stop_after = 512
|
29 |
-
outpath_base = 'D:\\tmp\\tortoise-tts-eval\\sweep-2'
|
30 |
-
outpath_real = 'D:\\tmp\\tortoise-tts-eval\\real'
|
31 |
-
|
32 |
-
arg_ranges = {
|
33 |
-
'top_p': [.8,1],
|
34 |
-
'temperature': [.8,.9,1],
|
35 |
-
'diffusion_temperature': [.8,1],
|
36 |
-
'cond_free_k': [1,2,5,10],
|
37 |
-
}
|
38 |
-
cfgs = permutations(arg_ranges)
|
39 |
-
shuffle(cfgs)
|
40 |
-
|
41 |
-
for cfg in cfgs:
|
42 |
-
cfg_desc = '_'.join([f'{k}-{v}' for k,v in cfg.items()])
|
43 |
-
outpath = os.path.join(outpath_base, f'{cfg_desc}')
|
44 |
-
os.makedirs(outpath, exist_ok=True)
|
45 |
-
os.makedirs(outpath_real, exist_ok=True)
|
46 |
-
with open(fname, 'r', encoding='utf-8') as f:
|
47 |
-
lines = [l.strip().split('\t') for l in f.readlines()]
|
48 |
-
|
49 |
-
recorder = open(os.path.join(outpath, 'transcript.tsv'), 'w', encoding='utf-8')
|
50 |
-
tts = TextToSpeech()
|
51 |
-
for e, line in enumerate(lines):
|
52 |
-
if e >= stop_after:
|
53 |
-
break
|
54 |
-
transcript = line[0]
|
55 |
-
path = os.path.join(os.path.dirname(fname), line[1])
|
56 |
-
cond_audio = load_audio(path, 22050)
|
57 |
-
torchaudio.save(os.path.join(outpath_real, os.path.basename(line[1])), cond_audio, 22050)
|
58 |
-
sample = tts.tts(transcript, [cond_audio, cond_audio], num_autoregressive_samples=32, repetition_penalty=2.0,
|
59 |
-
k=1, diffusion_iterations=32, length_penalty=1.0, **cfg)
|
60 |
-
down = torchaudio.functional.resample(sample, 24000, 22050)
|
61 |
-
fout_path = os.path.join(outpath, os.path.basename(line[1]))
|
62 |
-
torchaudio.save(fout_path, down.squeeze(0), 22050)
|
63 |
-
recorder.write(f'{transcript}\t{fout_path}\n')
|
64 |
-
recorder.flush()
|
65 |
-
recorder.close()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
tortoise/{samples_generator.py → utils/samples_generator.py}
RENAMED
@@ -4,7 +4,7 @@ import os
|
|
4 |
|
5 |
if __name__ == '__main__':
|
6 |
result = "<html><head><title>These words were never spoken.</title></head><body><h1>Handpicked results</h1>"
|
7 |
-
for fv in os.listdir('
|
8 |
url = f'https://github.com/neonbjb/tortoise-tts/raw/main/results/favorites/{fv}'
|
9 |
result = result + f'<audio controls="" style="width: 600px;"><source src="{url}" type="audio/mp3"></audio><br>\n'
|
10 |
|
@@ -30,7 +30,7 @@ if __name__ == '__main__':
|
|
30 |
line = line + f'<td><audio controls="" style="width: 150px;"><source src="{url}" type="audio/mp3"></audio></td>'
|
31 |
line = line + "</tr>"
|
32 |
lines.append(line)
|
33 |
-
for txt in os.listdir('
|
34 |
if 'desktop' in txt:
|
35 |
continue
|
36 |
line = f'<tr><td>{txt}</td>'
|
@@ -42,7 +42,7 @@ if __name__ == '__main__':
|
|
42 |
result = result + '\n'.join(lines) + "</table>"
|
43 |
|
44 |
result = result + "<h1>Longform result for all voices:</h1>"
|
45 |
-
for lf in os.listdir('
|
46 |
url = f'https://github.com/neonbjb/tortoise-tts/raw/main/results/riding_hood/{lf}'
|
47 |
result = result + f'<audio controls="" style="width: 600px;"><source src="{url}" type="audio/mp3"></audio><br>\n'
|
48 |
|
|
|
4 |
|
5 |
if __name__ == '__main__':
|
6 |
result = "<html><head><title>These words were never spoken.</title></head><body><h1>Handpicked results</h1>"
|
7 |
+
for fv in os.listdir('../../results/favorites'):
|
8 |
url = f'https://github.com/neonbjb/tortoise-tts/raw/main/results/favorites/{fv}'
|
9 |
result = result + f'<audio controls="" style="width: 600px;"><source src="{url}" type="audio/mp3"></audio><br>\n'
|
10 |
|
|
|
30 |
line = line + f'<td><audio controls="" style="width: 150px;"><source src="{url}" type="audio/mp3"></audio></td>'
|
31 |
line = line + "</tr>"
|
32 |
lines.append(line)
|
33 |
+
for txt in os.listdir('../../results/various/'):
|
34 |
if 'desktop' in txt:
|
35 |
continue
|
36 |
line = f'<tr><td>{txt}</td>'
|
|
|
42 |
result = result + '\n'.join(lines) + "</table>"
|
43 |
|
44 |
result = result + "<h1>Longform result for all voices:</h1>"
|
45 |
+
for lf in os.listdir('../../results/riding_hood'):
|
46 |
url = f'https://github.com/neonbjb/tortoise-tts/raw/main/results/riding_hood/{lf}'
|
47 |
result = result + f'<audio controls="" style="width: 600px;"><source src="{url}" type="audio/mp3"></audio><br>\n'
|
48 |
|