-
Notifications
You must be signed in to change notification settings - Fork 6
/
tts.py
232 lines (189 loc) · 9.08 KB
/
tts.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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
import os
from glob import glob
import soundfile as sf
import numpy as np
import yaml
import uuid
import torch
from tensorrt_onnx.run_onnx import ONNXModel
from tensorrt_onnx.run_trt import TRTWrapper
from mykantts.bin.infer_sambert import am_init, am_synthesis
from mykantts.bin.infer_hifigan import hifigan_init
from mykantts.utils.ling_unit.ling_unit import KanTtsLinguisticUnit
import time
import wave
import ttsfrd
import logging
ENG_LANG_MAPPING = {
"PinYin": "zh-cn",
"English": "en-us",
"British": "en-gb",
"ZhHK": "hk_cantonese",
"Sichuan": "sichuan",
"Japanese": "japanese",
"WuuShangHai": "shanghai",
"Indonesian": "indonesian",
"Malay": "malay",
"Filipino": "filipino",
"Vietnamese": "vietnamese",
"Korean": "korean",
"Russian": "russian",
}
def text_to_mit_symbols1(text, fe, speaker):
symbols_lst = []
text = text.strip()
res = fe.gen_tacotron_symbols(text)
res = res.replace("F7", speaker)
sentences = res.split("\n")
for sentence in sentences:
arr = sentence.split("\t")
# skip the empty line
if len(arr) != 2:
continue
sub_index, symbols = sentence.split("\t")
symbol_str = "{}_{}\t{}\n".format(0, sub_index, symbols)
symbols_lst.append(symbol_str)
return symbols_lst
def save_wav(data, out_path):
with wave.open(out_path, 'w') as wav_file:
wav_file.setnchannels(1) # 单声道
wav_file.setsampwidth(2) # 2 字节,16-bit
wav_file.setframerate(16000) # 采样率为 44100 Hz
# 将数据转换为 16-bit 格式
data = (data * 32767).astype(np.int16)
# 写入数据
wav_file.writeframes(data.tobytes())
class TTS():
def __init__(self, basepath, voice, infer_type):
self.infer_type = infer_type
self.resource_dir = f"{basepath}/resource"
self.am_ckpt = f"{basepath}/voices/{voice}/am"
self.voc_ckpt = f"{basepath}/voices/{voice}/voc"
self.se_file = None
self.am_config = os.path.join(self.am_ckpt, "config.yaml")
self.voc_config = os.path.join(self.voc_ckpt, "config.yaml")
self.output_dir = r"./outputs"
with open(self.am_config, "r") as f:
self.config = yaml.load(f, Loader=yaml.Loader)
self.speaker = self.config["linguistic_unit"]["speaker_list"].split(",")[0]
logging.info(f"HifiGAN infer using : {infer_type}.......")
if infer_type == "torch":
self.hifigan_model = hifigan_init(ckpt_path=os.path.join(self.voc_ckpt, "ckpt/checkpoint_0.pth"),
config=self.voc_config)
p = os.path.join(self.voc_ckpt, "ckpt/checkpoint_0.pth")
logging.info(f"Loading HifiGAN checkpoint: {p}")
elif infer_type == "onnx_cpu":
self.hifigan_model = ONNXModel(f"./tensorrt_onnx/model_save/simplify_model_{voice}.onnx", device='cpu')
logging.info(f"Loading HifiGAN checkpoint: ./tensorrt_onnx/model_save/simplify_model_{voice}.onnx")
elif infer_type == "onnx_gpu":
self.hifigan_model = ONNXModel(f"./tensorrt_onnx/model_save/simplify_model_{voice}.onnx", device='gpu')
logging.info(f"Loading HifiGAN checkpoint: ./tensorrt_onnx/model_save/simplify_model_{voice}.onnx")
elif infer_type == "trt":
self.hifigan_model = TRTWrapper(f"./tensorrt_onnx/model_save/simplify_model_{voice}.engine", ['output'])
logging.info(f"Loading HifiGAN checkpoint: ./tensorrt_onnx/model_save/simplify_model_{voice}.engine")
else:
print("Wrong infer type......")
self.am_model = am_init(ckpt=os.path.join(self.am_ckpt, "ckpt/checkpoint_0.pth"), config=self.am_config, infer_type=infer_type, voice=voice)
self.fe = ttsfrd.TtsFrontendEngine()
self.fe.initialize(self.resource_dir)
self.fe.set_lang_type(ENG_LANG_MAPPING["PinYin"])
def concat_process(self, chunked_dir, output_dir):
wav_files = sorted(glob(os.path.join(chunked_dir, "*.wav")))
# print(wav_files)
sentence_sil = 0.28 # seconds
end_sil = 0.05 # seconds
wav_concat, sr = sf.read(wav_files[0])
sentence_sil_samples = int(sentence_sil * sr)
end_sil_samples = int(end_sil * sr)
if len(wav_files) >= 2:
for p in wav_files[1:]:
wav, sr = sf.read(p)
wav_concat = np.concatenate(
(wav_concat, np.zeros(sentence_sil_samples), wav), axis=0
)
wav_concat = np.concatenate((wav_concat, np.zeros(end_sil_samples)), axis=0)
save_wav(wav_concat, os.path.join(output_dir, f"all.wav"))
def infer(self, text, scale=1.0):
# t0 = time.time()
self.output_dir = "./outputs/" + str(uuid.uuid4())
os.makedirs(self.output_dir, exist_ok=True)
os.makedirs(os.path.join(self.output_dir, "res_wavs"), exist_ok=True)
t0 = time.time()
symbols_lst = text_to_mit_symbols1(text, self.fe, self.speaker)
t1 = time.time()
print("文本前端推理时间: ", (t1-t0)*1000, " ms")
symbols_file = os.path.join(self.output_dir, "symbols.lst")
with open(symbols_file, "w") as symbol_data:
for symbol in symbols_lst:
symbol_data.write(symbol)
with open(self.am_config, "r") as f:
config = yaml.load(f, Loader=yaml.Loader)
ling_unit = KanTtsLinguisticUnit(config)
ling_unit_size = ling_unit.get_unit_size()
config["Model"]["KanTtsSAMBERT"]["params"].update(ling_unit_size)
if not torch.cuda.is_available():
device = torch.device("cpu")
else:
torch.backends.cudnn.benchmark = True
device = torch.device("cuda", 0)
with open(symbols_file, encoding="utf-8") as f:
for i, line in enumerate(f):
line = line.strip().split("\t")
# logging.info("Inference sentence: {}".format(line[0]))
# mel_path = "%s/%s_mel.npy" % (results_dir, line[0])
# dur_path = "%s/%s_dur.txt" % (results_dir, line[0])
# f0_path = "%s/%s_f0.txt" % (results_dir, line[0])
# energy_path = "%s/%s_energy.txt" % (results_dir, line[0])
t0 = time.time()
with torch.no_grad():
mel, mel_post, dur, f0, energy = am_synthesis(
line[1], self.am_model, ling_unit, device, se=None, scale=scale
)
t1 = time.time()
print("am infer time: ", (t1-t0)*1000, " ms")
# (T, C) -> (B, C, T)
mel_data = mel_post.transpose(1, 0).unsqueeze(0)
if self.infer_type == "torch":
# mel_data = torch.tensor(mel_post, dtype=torch.float).to(device)
# print("mel_data.shape: ", mel_data.shape)
y = self.hifigan_model(mel_data)
y = y.view(-1).detach().cpu().numpy()
elif self.infer_type in ["onnx_cpu", "onnx_gpu"]:
out_onnx = self.hifigan_model.onnx_session.run([], input_feed={'input': mel_data.cpu().numpy()})
y = np.squeeze(out_onnx[0])
# print("y shape: ", y.shape)
elif self.infer_type == 'trt':
output = self.hifigan_model(dict(input=mel_data.cuda()))
# print(output)
# print(output['output'].shape)
y = output['output'].view(-1).detach().cpu().numpy()
# if hasattr(self.hifigan_model, "pqmf"):
# print("----------------------")
# y = self.hifigan_model.pqmf.synthesis(y)
# y = y.view(-1).detach().cpu().numpy()
# pcm_len += len(y)
t2 = time.time()
print("hifigan infer time: ", (t2-t1)*1000, " ms")
# print(y)
save_wav(y, os.path.join(self.output_dir, f"{i}_gen.wav"))
# t1 = time.time()
# print("文本前端infer time: ", t1-t0)
# t0 = time.time()
# logging.info("AM is infering...")
# # am_forward(symbols_file, os.path.join(am_ckpt, "ckpt/checkpoint_0.pth"), output_dir, se_file, config=am_config)
# am_forward(ckpt=os.path.join(self.am_ckpt, "ckpt/checkpoint_0.pth"), sentence=symbols_file, model=self.am_model,
# output_dir=self.output_dir, se_file=None, config=self.am_config, scale=scale)
# t1 = time.time()
# print("am_forward time: ", t1-t0)
# t0 = time.time()
# logging.info("Vocoder is infering...")
# hifigan_forward(os.path.join(self.output_dir, "feat"), model=self.hifigan_model, output_dir=self.output_dir)
# t1 = time.time()
# print("hifigan_forward time: ", t1-t0)
self.concat_process(self.output_dir, os.path.join(self.output_dir, "res_wavs"))
# t2 = time.time()
# print("concat_process time: ", t2 - t1)
# logging.info("Text to wav finished!")
# t1 = time.time()
# print("infer time: ", t1-t0)
return self.output_dir