-
Notifications
You must be signed in to change notification settings - Fork 3
/
inference.py
193 lines (160 loc) · 7.36 KB
/
inference.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
from models import SynthesizerTrn
import argparse
import utils
from PL_BERT_ja.text.symbols import symbols
import json
from preprocess_ja import get_pl_bert_ja
import torch
import soundcard as sc
import time
import os
import soundfile as sf
from transformers import BertJapaneseTokenizer
import torch
from PL_BERT_ja.text_utils import TextCleaner
from PL_BERT_ja.phonemize import phonemize
import commons
from text import cleaned_text_to_sequence, text_to_sequence
def inference(model_ckpt_path, model_config_path, pl_bert_dir, is_save=True):
with open(model_config_path, "r") as f:
data = f.read()
config = json.loads(data)
hps = utils.HParams(**config)
if hps.model.use_noise_scaled_mas is True :
print("Using noise scaled MAS for VITS2")
use_noise_scaled_mas = True
mas_noise_scale_initial = 0.01
noise_scale_delta = 2e-6
net_g = SynthesizerTrn(
len(symbols)+1,
hps.data.n_mel_channels,
hps.train.segment_size // hps.data.hop_length,
n_speakers=hps.data.n_speakers,
mas_noise_scale_initial=mas_noise_scale_initial,
noise_scale_delta=noise_scale_delta,
**hps.model,
)
pl_bert_model, pl_bert_config, device = get_pl_bert_ja(dir=pl_bert_dir)
pl_bert_cleaner = TextCleaner()
pl_bert_tokenizer = BertJapaneseTokenizer.from_pretrained(pl_bert_config['dataset_params']['tokenizer'])
net_g, _, _, _ = utils.load_checkpoint( model_ckpt_path, net_g, optimizer=None)
# play audio by system default
speaker = sc.get_speaker(sc.default_speaker().name)
# parameter settings
noise_scale = torch.tensor(0.66) # adjust z_p noise
noise_scale_w = torch.tensor(0.8) # adjust SDP noise
length_scale = torch.tensor(1.0) # adjust sound length scale (talk speed)
if is_save is True:
n_save = 0
save_dir = os.path.join("./infer_logs/")
os.makedirs(save_dir, exist_ok=True)
net_g = net_g.to(device)
pl_bert_model = pl_bert_model.to(device)
### Dummy Input ###
with torch.inference_mode():
dummy_text = "色々疲れちまったけど、やっぱ音声合成してるときが一番ワクワクするんだよな。"
# get bert features
bert_features, phonemes = get_bert_features(dummy_text, pl_bert_model, pl_bert_tokenizer, pl_bert_config, pl_bert_cleaner, device, add_blank=hps.data.add_blank)
x = get_text_ids(phonemes=phonemes,
add_blank=hps.data.add_blank)
x = x.unsqueeze(0)
bert_features = bert_features.unsqueeze(0)
x_lengths = torch.LongTensor([x.size(1)])
sid = torch.LongTensor([0])
net_g.infer(x .to(device),
x_lengths .to(device),
bert_features .to(device),
x_lengths .to(device),
sid .to(device),
noise_scale=noise_scale.to(device),
noise_scale_w=noise_scale_w.to(device),
length_scale=length_scale.to(device),
max_len=1000)
while True:
# get text
text = input("Enter text. ==> ")
if text=="":
print("Empty input is detected... Exit...")
break
# measure the execution time
torch.cuda.synchronize()
start = time.time()
# required_grad is False
with torch.inference_mode():
bert_features, phonemes = get_bert_features(text, pl_bert_model, pl_bert_tokenizer, pl_bert_config, pl_bert_cleaner, device, add_blank=hps.data.add_blank)
x = get_text_ids(phonemes=phonemes,
add_blank=hps.data.add_blank).unsqueeze(0)
bert_features = bert_features.unsqueeze(0)
x_lengths = torch.LongTensor([x.size(1)])
sid = torch.LongTensor([0])
y_hat, _, _, _ = net_g.infer(x .to(device),
x_lengths .to(device),
bert_features .to(device),
x_lengths .to(device),
sid .to(device),
noise_scale=noise_scale.to(device),
noise_scale_w=noise_scale_w.to(device),
length_scale=length_scale.to(device),
max_len=1000)
y_hat = y_hat.permute(0,2,1)[0, :, :].cpu().float().numpy().copy()
# measure the execution time
torch.cuda.synchronize()
elapsed_time = time.time() - start
print(f"Gen Time : {elapsed_time}")
# play audio
speaker.play(y_hat, hps.data.sampling_rate)
# save audio
if is_save is True:
n_save += 1
data = y_hat
try:
save_path = os.path.join(save_dir, str(n_save).zfill(3)+f"_{text}.wav")
sf.write(
file=save_path,
data=data,
samplerate=hps.data.sampling_rate,
format="WAV")
except:
save_path = os.path.join(save_dir, str(n_save).zfill(3)+f"_{text[:10]}〜.wav")
sf.write(
file=save_path,
data=data,
samplerate=hps.data.sampling_rate,
format="WAV")
print(f"Audio is saved at : {save_path}")
return 0
def get_text_ids(phonemes, add_blank):
text_norm = cleaned_text_to_sequence(phonemes)
if add_blank:
text_norm = commons.intersperse(text_norm, 0)
text_norm = torch.LongTensor(text_norm)
return text_norm
def get_bert_features(text, pl_bert_model, pl_bert_tokenizer, pl_bert_config, pl_bert_cleaner, device, add_blank):
text = text.replace("\n", "")
hidden_size = pl_bert_config["model_params"]["hidden_size"]
n_layers = pl_bert_config["model_params"]["num_hidden_layers"] + 1
phonemes = ''.join(phonemize(text,pl_bert_tokenizer)["phonemes"])
input_ids = pl_bert_cleaner(phonemes)
with torch.inference_mode():
hidden_stats = pl_bert_model(torch.tensor(input_ids, dtype=torch.int64, device=device).unsqueeze(0))[-1]["hidden_states"]
save_tensor = torch.zeros(size=(n_layers, len(input_ids), hidden_size))
for idx, hidden_stat in enumerate(hidden_stats):
save_tensor[idx, :, :] = hidden_stat
if add_blank is True:
L, T, H = save_tensor.shape
new_data = torch.zeros(size=(L,2*T+1,H), dtype=save_tensor.dtype)
for idx in range(T):
target_idx = idx*2+1
new_data[:, target_idx, :] = save_tensor[:, idx, :]
save_tensor = new_data
return save_tensor, phonemes
def text2input_ids():
return 0
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--model_ckpt_path", default="./logs/AddBlankTrue/G_54000.pth")
parser.add_argument("--model_cnfg_path", default="./logs/AddBlankTrue/config.json")
parser.add_argument("--pl_bert_dir", default="./plb-ja_10000000-steps/")
parser.add_argument("--is_save", default=False)
args = parser.parse_args()
inference(args.model_ckpt_path, args.model_cnfg_path, args.pl_bert_dir, args.is_save)