Spaces:
Running
on
Zero
Running
on
Zero
Delete infer/infer_cli.py
Browse files- infer/infer_cli.py +0 -226
infer/infer_cli.py
DELETED
@@ -1,226 +0,0 @@
|
|
1 |
-
import argparse
|
2 |
-
import codecs
|
3 |
-
import os
|
4 |
-
import re
|
5 |
-
from importlib.resources import files
|
6 |
-
from pathlib import Path
|
7 |
-
|
8 |
-
import numpy as np
|
9 |
-
import soundfile as sf
|
10 |
-
import tomli
|
11 |
-
from cached_path import cached_path
|
12 |
-
|
13 |
-
from infer.utils_infer import (
|
14 |
-
infer_process,
|
15 |
-
load_model,
|
16 |
-
load_vocoder,
|
17 |
-
preprocess_ref_audio_text,
|
18 |
-
remove_silence_for_generated_wav,
|
19 |
-
)
|
20 |
-
from model import DiT, UNetT
|
21 |
-
|
22 |
-
parser = argparse.ArgumentParser(
|
23 |
-
prog="python3 infer-cli.py",
|
24 |
-
description="Commandline interface for E2/F5 TTS with Advanced Batch Processing.",
|
25 |
-
epilog="Specify options above to override one or more settings from config.",
|
26 |
-
)
|
27 |
-
parser.add_argument(
|
28 |
-
"-c",
|
29 |
-
"--config",
|
30 |
-
help="Configuration file. Default=infer/examples/basic/basic.toml",
|
31 |
-
default=os.path.join(files("f5_tts").joinpath("infer/examples/basic"), "basic.toml"),
|
32 |
-
)
|
33 |
-
parser.add_argument(
|
34 |
-
"-m",
|
35 |
-
"--model",
|
36 |
-
help="F5-TTS | E2-TTS",
|
37 |
-
)
|
38 |
-
parser.add_argument(
|
39 |
-
"-p",
|
40 |
-
"--ckpt_file",
|
41 |
-
help="The Checkpoint .pt",
|
42 |
-
)
|
43 |
-
parser.add_argument(
|
44 |
-
"-v",
|
45 |
-
"--vocab_file",
|
46 |
-
help="The vocab .txt",
|
47 |
-
)
|
48 |
-
parser.add_argument("-r", "--ref_audio", type=str, help="Reference audio file < 15 seconds.")
|
49 |
-
parser.add_argument("-s", "--ref_text", type=str, default="666", help="Subtitle for the reference audio.")
|
50 |
-
parser.add_argument(
|
51 |
-
"-t",
|
52 |
-
"--gen_text",
|
53 |
-
type=str,
|
54 |
-
help="Text to generate.",
|
55 |
-
)
|
56 |
-
parser.add_argument(
|
57 |
-
"-f",
|
58 |
-
"--gen_file",
|
59 |
-
type=str,
|
60 |
-
help="File with text to generate. Ignores --gen_text",
|
61 |
-
)
|
62 |
-
parser.add_argument(
|
63 |
-
"-o",
|
64 |
-
"--output_dir",
|
65 |
-
type=str,
|
66 |
-
help="Path to output folder..",
|
67 |
-
)
|
68 |
-
parser.add_argument(
|
69 |
-
"-w",
|
70 |
-
"--output_file",
|
71 |
-
type=str,
|
72 |
-
help="Filename of output file..",
|
73 |
-
)
|
74 |
-
parser.add_argument(
|
75 |
-
"--remove_silence",
|
76 |
-
help="Remove silence.",
|
77 |
-
)
|
78 |
-
parser.add_argument("--vocoder_name", type=str, default="vocos", choices=["vocos", "bigvgan"], help="vocoder name")
|
79 |
-
parser.add_argument(
|
80 |
-
"--load_vocoder_from_local",
|
81 |
-
action="store_true",
|
82 |
-
help="load vocoder from local. Default: ../checkpoints/charactr/vocos-mel-24khz",
|
83 |
-
)
|
84 |
-
parser.add_argument(
|
85 |
-
"--speed",
|
86 |
-
type=float,
|
87 |
-
default=1.0,
|
88 |
-
help="Adjust the speed of the audio generation (default: 1.0)",
|
89 |
-
)
|
90 |
-
args = parser.parse_args()
|
91 |
-
|
92 |
-
config = tomli.load(open(args.config, "rb"))
|
93 |
-
|
94 |
-
ref_audio = args.ref_audio if args.ref_audio else config["ref_audio"]
|
95 |
-
ref_text = args.ref_text if args.ref_text != "666" else config["ref_text"]
|
96 |
-
gen_text = args.gen_text if args.gen_text else config["gen_text"]
|
97 |
-
gen_file = args.gen_file if args.gen_file else config["gen_file"]
|
98 |
-
|
99 |
-
# patches for pip pkg user
|
100 |
-
if "infer/examples/" in ref_audio:
|
101 |
-
ref_audio = str(files("f5_tts").joinpath(f"{ref_audio}"))
|
102 |
-
if "infer/examples/" in gen_file:
|
103 |
-
gen_file = str(files("f5_tts").joinpath(f"{gen_file}"))
|
104 |
-
if "voices" in config:
|
105 |
-
for voice in config["voices"]:
|
106 |
-
voice_ref_audio = config["voices"][voice]["ref_audio"]
|
107 |
-
if "infer/examples/" in voice_ref_audio:
|
108 |
-
config["voices"][voice]["ref_audio"] = str(files("f5_tts").joinpath(f"{voice_ref_audio}"))
|
109 |
-
|
110 |
-
if gen_file:
|
111 |
-
gen_text = codecs.open(gen_file, "r", "utf-8").read()
|
112 |
-
output_dir = args.output_dir if args.output_dir else config["output_dir"]
|
113 |
-
output_file = args.output_file if args.output_file else config["output_file"]
|
114 |
-
model = args.model if args.model else config["model"]
|
115 |
-
ckpt_file = args.ckpt_file if args.ckpt_file else ""
|
116 |
-
vocab_file = args.vocab_file if args.vocab_file else ""
|
117 |
-
remove_silence = args.remove_silence if args.remove_silence else config["remove_silence"]
|
118 |
-
speed = args.speed
|
119 |
-
|
120 |
-
wave_path = Path(output_dir) / output_file
|
121 |
-
# spectrogram_path = Path(output_dir) / "infer_cli_out.png"
|
122 |
-
|
123 |
-
vocoder_name = args.vocoder_name
|
124 |
-
mel_spec_type = args.vocoder_name
|
125 |
-
if vocoder_name == "vocos":
|
126 |
-
vocoder_local_path = "../checkpoints/vocos-mel-24khz"
|
127 |
-
elif vocoder_name == "bigvgan":
|
128 |
-
vocoder_local_path = "../checkpoints/bigvgan_v2_24khz_100band_256x"
|
129 |
-
|
130 |
-
vocoder = load_vocoder(vocoder_name=mel_spec_type, is_local=args.load_vocoder_from_local, local_path=vocoder_local_path)
|
131 |
-
|
132 |
-
|
133 |
-
# load models
|
134 |
-
if model == "F5-TTS":
|
135 |
-
model_cls = DiT
|
136 |
-
model_cfg = dict(dim=1024, depth=22, heads=16, ff_mult=2, text_dim=512, conv_layers=4)
|
137 |
-
if ckpt_file == "":
|
138 |
-
if vocoder_name == "vocos":
|
139 |
-
repo_name = "F5-TTS"
|
140 |
-
exp_name = "F5TTS_Base"
|
141 |
-
ckpt_step = 1200000
|
142 |
-
ckpt_file = str(cached_path(f"hf://SWivid/{repo_name}/{exp_name}/model_{ckpt_step}.safetensors"))
|
143 |
-
# ckpt_file = f"ckpts/{exp_name}/model_{ckpt_step}.pt" # .pt | .safetensors; local path
|
144 |
-
elif vocoder_name == "bigvgan":
|
145 |
-
repo_name = "F5-TTS"
|
146 |
-
exp_name = "F5TTS_Base_bigvgan"
|
147 |
-
ckpt_step = 1250000
|
148 |
-
ckpt_file = str(cached_path(f"hf://SWivid/{repo_name}/{exp_name}/model_{ckpt_step}.pt"))
|
149 |
-
|
150 |
-
elif model == "E2-TTS":
|
151 |
-
assert vocoder_name == "vocos", "E2-TTS only supports vocoder vocos"
|
152 |
-
model_cls = UNetT
|
153 |
-
model_cfg = dict(dim=1024, depth=24, heads=16, ff_mult=4)
|
154 |
-
if ckpt_file == "":
|
155 |
-
repo_name = "E2-TTS"
|
156 |
-
exp_name = "E2TTS_Base"
|
157 |
-
ckpt_step = 1200000
|
158 |
-
ckpt_file = str(cached_path(f"hf://SWivid/{repo_name}/{exp_name}/model_{ckpt_step}.safetensors"))
|
159 |
-
# ckpt_file = f"ckpts/{exp_name}/model_{ckpt_step}.pt" # .pt | .safetensors; local path
|
160 |
-
|
161 |
-
|
162 |
-
print(f"Using {model}...")
|
163 |
-
ema_model = load_model(model_cls, model_cfg, ckpt_file, mel_spec_type=mel_spec_type, vocab_file=vocab_file)
|
164 |
-
|
165 |
-
|
166 |
-
def main_process(ref_audio, ref_text, text_gen, model_obj, mel_spec_type, remove_silence, speed):
|
167 |
-
main_voice = {"ref_audio": ref_audio, "ref_text": ref_text}
|
168 |
-
if "voices" not in config:
|
169 |
-
voices = {"main": main_voice}
|
170 |
-
else:
|
171 |
-
voices = config["voices"]
|
172 |
-
voices["main"] = main_voice
|
173 |
-
for voice in voices:
|
174 |
-
voices[voice]["ref_audio"], voices[voice]["ref_text"] = preprocess_ref_audio_text(
|
175 |
-
voices[voice]["ref_audio"], voices[voice]["ref_text"]
|
176 |
-
)
|
177 |
-
print("Voice:", voice)
|
178 |
-
print("Ref_audio:", voices[voice]["ref_audio"])
|
179 |
-
print("Ref_text:", voices[voice]["ref_text"])
|
180 |
-
|
181 |
-
generated_audio_segments = []
|
182 |
-
reg1 = r"(?=\[\w+\])"
|
183 |
-
chunks = re.split(reg1, text_gen)
|
184 |
-
reg2 = r"\[(\w+)\]"
|
185 |
-
for text in chunks:
|
186 |
-
if not text.strip():
|
187 |
-
continue
|
188 |
-
match = re.match(reg2, text)
|
189 |
-
if match:
|
190 |
-
voice = match[1]
|
191 |
-
else:
|
192 |
-
print("No voice tag found, using main.")
|
193 |
-
voice = "main"
|
194 |
-
if voice not in voices:
|
195 |
-
print(f"Voice {voice} not found, using main.")
|
196 |
-
voice = "main"
|
197 |
-
text = re.sub(reg2, "", text)
|
198 |
-
gen_text = text.strip()
|
199 |
-
ref_audio = voices[voice]["ref_audio"]
|
200 |
-
ref_text = voices[voice]["ref_text"]
|
201 |
-
print(f"Voice: {voice}")
|
202 |
-
audio, final_sample_rate, spectragram = infer_process(
|
203 |
-
ref_audio, ref_text, gen_text, model_obj, vocoder, mel_spec_type=mel_spec_type, speed=speed
|
204 |
-
)
|
205 |
-
generated_audio_segments.append(audio)
|
206 |
-
|
207 |
-
if generated_audio_segments:
|
208 |
-
final_wave = np.concatenate(generated_audio_segments)
|
209 |
-
|
210 |
-
if not os.path.exists(output_dir):
|
211 |
-
os.makedirs(output_dir)
|
212 |
-
|
213 |
-
with open(wave_path, "wb") as f:
|
214 |
-
sf.write(f.name, final_wave, final_sample_rate)
|
215 |
-
# Remove silence
|
216 |
-
if remove_silence:
|
217 |
-
remove_silence_for_generated_wav(f.name)
|
218 |
-
print(f.name)
|
219 |
-
|
220 |
-
|
221 |
-
def main():
|
222 |
-
main_process(ref_audio, ref_text, gen_text, ema_model, mel_spec_type, remove_silence, speed)
|
223 |
-
|
224 |
-
|
225 |
-
if __name__ == "__main__":
|
226 |
-
main()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|