/
offline-tts.py
executable file
·237 lines (191 loc) · 6.82 KB
/
offline-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
233
234
235
236
237
#!/usr/bin/env python3
#
# Copyright (c) 2023 Xiaomi Corporation
"""
This file demonstrates how to use sherpa-onnx Python API to generate audio
from text, i.e., text-to-speech.
Different from ./offline-tts-play.py, this file does not play back the
generated audio.
Usage:
Example (1/3)
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-en_US-amy-low.tar.bz2
tar xf vits-piper-en_US-amy-low.tar.bz2
python3 ./python-api-examples/offline-tts.py \
--vits-model=./vits-piper-en_US-amy-low/en_US-amy-low.onnx \
--vits-tokens=./vits-piper-en_US-amy-low/tokens.txt \
--vits-data-dir=./vits-piper-en_US-amy-low/espeak-ng-data \
--output-filename=./generated.wav \
"Today as always, men fall into two groups: slaves and free men. Whoever does not have two-thirds of his day for himself, is a slave, whatever he may be: a statesman, a businessman, an official, or a scholar."
Example (2/3)
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-icefall-zh-aishell3.tar.bz2
tar xvf vits-icefall-zh-aishell3.tar.bz2
python3 ./python-api-examples/offline-tts.py \
--vits-model=./vits-icefall-zh-aishell3/model.onnx \
--vits-lexicon=./vits-icefall-zh-aishell3/lexicon.txt \
--vits-tokens=./vits-icefall-zh-aishell3/tokens.txt \
--tts-rule-fsts='./vits-icefall-zh-aishell3/phone.fst,./vits-icefall-zh-aishell3/date.fst,./vits-icefall-zh-aishell3/number.fst' \
--sid=21 \
--output-filename=./liubei-21.wav \
"勿以恶小而为之,勿以善小而不为。惟贤惟德,能服于人。122334"
Example (3/3)
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/sherpa-onnx-vits-zh-ll.tar.bz2
tar xvf sherpa-onnx-vits-zh-ll.tar.bz2
rm sherpa-onnx-vits-zh-ll.tar.bz2
python3 ./python-api-examples/offline-tts.py \
--vits-model=./sherpa-onnx-vits-zh-ll/model.onnx \
--vits-lexicon=./sherpa-onnx-vits-zh-ll/lexicon.txt \
--vits-tokens=./sherpa-onnx-vits-zh-ll/tokens.txt \
--tts-rule-fsts=./sherpa-onnx-vits-zh-ll/phone.fst,./sherpa-onnx-vits-zh-ll/date.fst,./sherpa-onnx-vits-zh-ll/number.fst \
--vits-dict-dir=./sherpa-onnx-vits-zh-ll/dict \
--sid=2 \
--output-filename=./test-2.wav \
"当夜幕降临,星光点点,伴随着微风拂面,我在静谧中感受着时光的流转,思念如涟漪荡漾,梦境如画卷展开,我与自然融为一体,沉静在这片宁静的美丽之中,感受着生命的奇迹与温柔。2024年5月11号,拨打110或者18920240511。123456块钱。"
You can find more models at
https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models
Please see
https://k2-fsa.github.io/sherpa/onnx/tts/index.html
for details.
"""
import argparse
import time
import sherpa_onnx
import soundfile as sf
def get_args():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter
)
parser.add_argument(
"--vits-model",
type=str,
help="Path to vits model.onnx",
)
parser.add_argument(
"--vits-lexicon",
type=str,
default="",
help="Path to lexicon.txt",
)
parser.add_argument(
"--vits-tokens",
type=str,
default="",
help="Path to tokens.txt",
)
parser.add_argument(
"--vits-data-dir",
type=str,
default="",
help="""Path to the dict directory of espeak-ng. If it is specified,
--vits-lexicon and --vits-tokens are ignored""",
)
parser.add_argument(
"--vits-dict-dir",
type=str,
default="",
help="Path to the dict directory for models using jieba",
)
parser.add_argument(
"--tts-rule-fsts",
type=str,
default="",
help="Path to rule.fst",
)
parser.add_argument(
"--max-num-sentences",
type=int,
default=2,
help="""Max number of sentences in a batch to avoid OOM if the input
text is very long. Set it to -1 to process all the sentences in a
single batch. A smaller value does not mean it is slower compared
to a larger one on CPU.
""",
)
parser.add_argument(
"--output-filename",
type=str,
default="./generated.wav",
help="Path to save generated wave",
)
parser.add_argument(
"--sid",
type=int,
default=0,
help="""Speaker ID. Used only for multi-speaker models, e.g.
models trained using the VCTK dataset. Not used for single-speaker
models, e.g., models trained using the LJ speech dataset.
""",
)
parser.add_argument(
"--debug",
type=bool,
default=False,
help="True to show debug messages",
)
parser.add_argument(
"--provider",
type=str,
default="cpu",
help="valid values: cpu, cuda, coreml",
)
parser.add_argument(
"--num-threads",
type=int,
default=1,
help="Number of threads for neural network computation",
)
parser.add_argument(
"--speed",
type=float,
default=1.0,
help="Speech speed. Larger->faster; smaller->slower",
)
parser.add_argument(
"text",
type=str,
help="The input text to generate audio for",
)
return parser.parse_args()
def main():
args = get_args()
print(args)
tts_config = sherpa_onnx.OfflineTtsConfig(
model=sherpa_onnx.OfflineTtsModelConfig(
vits=sherpa_onnx.OfflineTtsVitsModelConfig(
model=args.vits_model,
lexicon=args.vits_lexicon,
data_dir=args.vits_data_dir,
dict_dir=args.vits_dict_dir,
tokens=args.vits_tokens,
),
provider=args.provider,
debug=args.debug,
num_threads=args.num_threads,
),
rule_fsts=args.tts_rule_fsts,
max_num_sentences=args.max_num_sentences,
)
if not tts_config.validate():
raise ValueError("Please check your config")
tts = sherpa_onnx.OfflineTts(tts_config)
start = time.time()
audio = tts.generate(args.text, sid=args.sid, speed=args.speed)
end = time.time()
if len(audio.samples) == 0:
print("Error in generating audios. Please read previous error messages.")
return
elapsed_seconds = end - start
audio_duration = len(audio.samples) / audio.sample_rate
real_time_factor = elapsed_seconds / audio_duration
sf.write(
args.output_filename,
audio.samples,
samplerate=audio.sample_rate,
subtype="PCM_16",
)
print(f"Saved to {args.output_filename}")
print(f"The text is '{args.text}'")
print(f"Elapsed seconds: {elapsed_seconds:.3f}")
print(f"Audio duration in seconds: {audio_duration:.3f}")
print(f"RTF: {elapsed_seconds:.3f}/{audio_duration:.3f} = {real_time_factor:.3f}")
if __name__ == "__main__":
main()