offline-tts.py
3.6 KB
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
#!/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.
Usage:
1. Download a model
wget https://huggingface.co/csukuangfj/vits-ljs/resolve/main/vits-ljs.onnx
wget https://huggingface.co/csukuangfj/vits-ljs/resolve/main/lexicon.txt
wget https://huggingface.co/csukuangfj/vits-ljs/resolve/main/tokens.txt
python3 ./python-api-examples/offline-tts.py \
--vits-model=./vits-ljs.onnx \
--vits-lexicon=./lexicon.txt \
--vits-tokens=./tokens.txt \
--output-filename=./generated.wav \
'liliana, the most beautiful and lovely assistant of our team!'
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,
help="Path to lexicon.txt",
)
parser.add_argument(
"--vits-tokens",
type=str,
help="Path to tokens.txt",
)
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(
"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,
tokens=args.vits_tokens,
),
provider=args.provider,
debug=args.debug,
num_threads=args.num_threads,
)
)
tts = sherpa_onnx.OfflineTts(tts_config)
start = time.time()
audio = tts.generate(args.text, sid=args.sid)
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()