-
Notifications
You must be signed in to change notification settings - Fork 1
/
inference.py
147 lines (128 loc) · 5.56 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
from model.pl_model import PL_model
import yaml
import torch
from utils.tools import inference_synth_one_samples
from utils.model import get_vocoder
import numpy as np
from text import text_to_sequence
from string import punctuation
import re
import argparse
def read_lexicon(lex_path):
lexicon = {}
with open(lex_path) as f:
for line in f:
temp = re.split(r"\s+", line.strip("\n"))
word = temp[0]
phones = temp[1:]
if word.lower() not in lexicon:
lexicon[word.lower()] = phones
return lexicon
class inference_data_generator():
def __init__(self, preprocess_config):
self.preprocess_config = preprocess_config
def preprocess_english(self, text):
from g2p_en import G2p
text = text.rstrip(punctuation)
lexicon = read_lexicon(self.preprocess_config["path"]["lexicon_path"])
g2p = G2p()
phones = []
words = re.split(r"([,;.\-\?\!\s+])", text)
for w in words:
if w.lower() in lexicon:
phones += lexicon[w.lower()]
else:
phones += list(filter(lambda p: p != " ", g2p(w)))
phones = "{" + "}{".join(phones) + "}"
phones = re.sub(r"\{[^\w\s]?\}", "{sp}", phones)
phones = phones.replace("}{", " ")
print("Raw Text Sequence: {}".format(text))
print("Phoneme Sequence: {}".format(phones))
sequence = np.array(
text_to_sequence(
phones, self.preprocess_config["preprocessing"]["text"]["text_cleaners"]
)
)
return np.array(sequence)
def preprocess_korean(self, text):
from text.korean import tokenize
phones = tokenize(text)
phones = list(map(lambda x: 'pau' if x == ' ' else x, phones))
phones = "{ pau " + " ".join(phones) + " pau }"
print("Raw Text Sequence: {}".format(text))
print("Phoneme Sequence: {}".format(phones))
sequence = np.array(
text_to_sequence(
phones, self.preprocess_config["preprocessing"]["text"]["text_cleaners"]
)
)
return np.array(sequence)
def preprocess_japanese(self, text):
import pyopenjtalk
phones = "{ " + pyopenjtalk.g2p(text) + " pau }"
print("Raw Text Sequence: {}".format(text))
print("Phoneme Sequence: {}".format(phones))
sequence = np.array(
text_to_sequence(
phones, self.preprocess_config["preprocessing"]["text"]["text_cleaners"]
)
)
return np.array(sequence)
def process(self, text: str, lang: str,
speaker: int, ref_mel_path=None,
p_control=1.0, e_control=1.0, d_control=1.0):
if lang == 'kr':
sequence = self.preprocess_korean(text)
elif lang == 'ja':
sequence = self.preprocess_japanese(text)
elif lang == 'en':
sequence = self.preprocess_english(text)
else:
raise NotImplementedError("Wrong Language")
texts = torch.LongTensor(np.array([sequence]))
text_lens = torch.LongTensor(np.array([len(texts[0])]))
max_text_lens = int(max(text_lens))
mel = None
if ref_mel_path:
mel = torch.FloatTensor(np.array([np.load(ref_mel_path)]))
inputs = {
'speakers': torch.LongTensor(speaker),
'texts' : texts,
'text_lens': text_lens,
'max_text_lens' :max_text_lens,
'ref_mels' : mel,
'p_control' : torch.FloatTensor([p_control]),
'e_control' : torch.FloatTensor([e_control]),
'd_control': torch.FloatTensor([d_control]),
}
return inputs
def define_argparser():
p = argparse.ArgumentParser()
p.add_argument('--text', type=str, required=True)
p.add_argument('--basename', type=str, required=True)
p.add_argument("--checkpoint_path", type=str, required=True)
p.add_argument('--p_control', type=float, default=1.0)
p.add_argument('--e_control', type=float, default=1.0)
p.add_argument('--d_control', type=float, default=1.0)
p.add_argument('--speaker', type=int, default=0)
p.add_argument('--ref_mel_path', type=str, default=None)
p.add_argument("-t", '--train_config', default='./config/LJSpeech/train.yaml', type=str)
p.add_argument("-p", '--preprocess_config', default='./config/LJSpeech/preprocess.yaml', type=str)
p.add_argument("-m", '--model_config', default='./config/LJSpeech/model.yaml', type=str)
config = p.parse_args()
return config
def main(args):
train_config = yaml.load(open(args.train_config, "r"), Loader=yaml.FullLoader)
preprocess_config = yaml.load(open(args.preprocess_config, "r"), Loader=yaml.FullLoader)
model_config = yaml.load(open(args.model_config, "r"), Loader=yaml.FullLoader)
model = PL_model(train_config, preprocess_config, model_config).load_from_checkpoint(args.checkpoint_path).cpu()
vocoder = get_vocoder(model_config, 'cpu')
generator = inference_data_generator(preprocess_config)
lang = preprocess_config['preprocessing']['text']['language']
pred_inputs = generator.process(args.text, lang, args.speaker, args.ref_mel_path,
args.p_control, args.e_control, args.d_control)
pred = model.inference(pred_inputs)
inference_synth_one_samples(args.basename, pred[1], pred[9], vocoder, model_config, preprocess_config, './prediction/')
if __name__ == '__main__':
args = define_argparser()
main(args)