File size: 15,144 Bytes
0e18fe8 2878e41 0e18fe8 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 |
import re
import os
import numpy as np
import torch
from torch import no_grad, LongTensor
import argparse
import commons
from mel_processing import spectrogram_torch
import utils
from models import SynthesizerTrn
import gradio as gr
import librosa
import webbrowser
from text import text_to_sequence, _clean_text
device = "cuda:0" if torch.cuda.is_available() else "cpu"
language_marks = {
"Japanese": "",
"日本語": "[JA]",
"简体中文": "[ZH]",
"English": "[EN]",
"Mix": "",
}
def get_text(text, hps, is_symbol):
text_norm = text_to_sequence(
text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
if hps.data.add_blank:
text_norm = commons.intersperse(text_norm, 0)
text_norm = LongTensor(text_norm)
return text_norm
def create_tts_fn(model, hps, speaker_ids):
def tts_fn(text, speaker, language, ns, nsw, speed, is_symbol):
if language is not None:
text = language_marks[language] + text + language_marks[language]
speaker_id = speaker_ids[speaker]
stn_tst = get_text(text, hps, is_symbol)
with no_grad():
x_tst = stn_tst.unsqueeze(0).to(device)
x_tst_lengths = LongTensor([stn_tst.size(0)]).to(device)
sid = LongTensor([speaker_id]).to(device)
audio = model.infer(x_tst, x_tst_lengths, sid=sid, noise_scale=ns, noise_scale_w=nsw,
length_scale=1.0 / speed)[0][0, 0].data.cpu().float().numpy()
del stn_tst, x_tst, x_tst_lengths, sid
return "Success", (hps.data.sampling_rate, audio)
return tts_fn
def create_vc_fn(model, hps, speaker_ids):
def vc_fn(original_speaker, target_speaker, record_audio, upload_audio):
input_audio = record_audio if record_audio is not None else upload_audio
if input_audio is None:
return "You need to record or upload an audio", None
sampling_rate, audio = input_audio
original_speaker_id = speaker_ids[original_speaker]
target_speaker_id = speaker_ids[target_speaker]
audio = (audio / np.iinfo(audio.dtype).max).astype(np.float32)
if len(audio.shape) > 1:
audio = librosa.to_mono(audio.transpose(1, 0))
if sampling_rate != hps.data.sampling_rate:
audio = librosa.resample(
audio, orig_sr=sampling_rate, target_sr=hps.data.sampling_rate)
with no_grad():
y = torch.FloatTensor(audio)
y = y / max(-y.min(), y.max()) / 0.99
y = y.to(device)
y = y.unsqueeze(0)
spec = spectrogram_torch(y, hps.data.filter_length,
hps.data.sampling_rate, hps.data.hop_length, hps.data.win_length,
center=False).to(device)
spec_lengths = LongTensor([spec.size(-1)]).to(device)
sid_src = LongTensor([original_speaker_id]).to(device)
sid_tgt = LongTensor([target_speaker_id]).to(device)
audio = model.voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt)[0][
0, 0].data.cpu().float().numpy()
del y, spec, spec_lengths, sid_src, sid_tgt
return "Success", (hps.data.sampling_rate, audio)
return vc_fn
def get_text(text, hps, is_symbol):
text_norm = text_to_sequence(
text, hps.symbols, [] if is_symbol else hps.data.text_cleaners)
if hps.data.add_blank:
text_norm = commons.intersperse(text_norm, 0)
text_norm = LongTensor(text_norm)
return text_norm
def create_to_symbol_fn(hps):
def to_symbol_fn(is_symbol_input, input_text, temp_text):
return (_clean_text(input_text, hps.data.text_cleaners), input_text) if is_symbol_input \
else (temp_text, temp_text)
return to_symbol_fn
models_info = [
{
"languages": ['日本語', '简体中文', 'English', 'Mix'],
"description": """
这个模型包含Blue Archive的142名角色，能合成中日英三语。\n\n
中英效果肯定没有日语好。\n\n
若需要在同一个句子中混合多种语言，使用相应的语言标记包裹句子。 （日语用[JA], 中文用[ZH], 英文用[EN]），参考Examples中的示例。
""",
"model_path": "./G_15100.pth",
"config_path": "./config.json",
"examples": [['メイドのアリスに何でもお任せください。', '爱丽丝（女仆）', '日本語', 1, False],
['ちゃーんといい子でお留守番してたよ。', '未花', '日本語', 1, False],
['老师，欢迎。今天也由我来保护老师吧。', '阿露', '简体中文', 1, False],
['Can you tell me how much the shirt is?',
'日富美', 'English', 1, False],
['[EN]Excuse me?[EN][JA]お帰りなさい，お兄様！[JA]', '优香（体操服）', 'Mix', 1, False]],
}
]
models_tts = []
models_vc = []
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--share", action="store_true",
default=False, help="share gradio app")
args = parser.parse_args()
categories = ["Blue Archive"]
others = {
"Princess Connect! Re:Dive": "https://huggingface.co/spaces/FrankZxShen/vits-fast-finetuning-pcr",
"Umamusume": "https://huggingface.co/spaces/FrankZxShen/vits-fast-finetuning-umamusume",
}
for info in models_info:
lang = info['languages']
examples = info['examples']
config_path = info['config_path']
model_path = info['model_path']
description = info['description']
hps = utils.get_hparams_from_file(config_path)
net_g = SynthesizerTrn(
len(hps.symbols),
hps.data.filter_length // 2 + 1,
hps.train.segment_size // hps.data.hop_length,
n_speakers=hps.data.n_speakers,
**hps.model).to(device)
_ = net_g.eval()
_ = utils.load_checkpoint(model_path, net_g, None)
speaker_ids = hps.speakers
speakers = list(hps.speakers.keys())
models_tts.append((description, speakers, lang, examples,
hps.symbols, create_tts_fn(net_g, hps, speaker_ids),
create_to_symbol_fn(hps)))
models_vc.append(
(description, speakers, create_vc_fn(net_g, hps, speaker_ids)))
app = gr.Blocks()
with app:
gr.Markdown(
"# <center> vits-fast-fineturning-models-ba\n"
"## <center> Please do not generate content that could infringe upon the rights or cause harm to individuals or organizations.\n"
"## <center> 请不要生成会对个人以及组织造成侵害的内容\n\n"
"[](https://colab.research.google.com/drive/1pn1xnFfdLK63gVXDwV4zCXfVeo8c-I-0?usp=sharing)\n\n"
"[](https://huggingface.co/spaces/FrankZxShen/vits-fast-finetuning-ba?duplicate=true)\n\n"
"[](https://github.com/Plachtaa/VITS-fast-fine-tuning)"
)
gr.Markdown("# TTS&Voice Conversion for Blue Archive\n\n"
)
with gr.Tabs():
for category in categories:
with gr.TabItem(category):
with gr.Tab("TTS"):
for i, (description, speakers, lang, example, symbols, tts_fn, to_symbol_fn) in enumerate(
models_tts):
gr.Markdown(description)
with gr.Row():
with gr.Column():
textbox = gr.TextArea(label="Text",
placeholder="Type your sentence here ",
value="よーし、私もがんばらないと！", elem_id=f"tts-input")
with gr.Accordion(label="Phoneme Input", open=False):
temp_text_var = gr.Variable()
symbol_input = gr.Checkbox(
value=False, label="Symbol input")
symbol_list = gr.Dataset(label="Symbol list", components=[textbox],
samples=[[x]
for x in symbols],
elem_id=f"symbol-list")
symbol_list_json = gr.Json(
value=symbols, visible=False)
symbol_input.change(to_symbol_fn,
[symbol_input, textbox,
temp_text_var],
[textbox, temp_text_var])
symbol_list.click(None, [symbol_list, symbol_list_json], textbox,
_js=f"""
(i, symbols, text) => {{
let root = document.querySelector("body > gradio-app");
if (root.shadowRoot != null)
root = root.shadowRoot;
let text_input = root.querySelector("#tts-input").querySelector("textarea");
let startPos = text_input.selectionStart;
let endPos = text_input.selectionEnd;
let oldTxt = text_input.value;
let result = oldTxt.substring(0, startPos) + symbols[i] + oldTxt.substring(endPos);
text_input.value = result;
let x = window.scrollX, y = window.scrollY;
text_input.focus();
text_input.selectionStart = startPos + symbols[i].length;
text_input.selectionEnd = startPos + symbols[i].length;
text_input.blur();
window.scrollTo(x, y);
text = text_input.value;
return text;
}}""")
# select character
char_dropdown = gr.Dropdown(
choices=speakers, value=speakers[0], label='character')
language_dropdown = gr.Dropdown(
choices=lang, value=lang[0], label='language')
ns = gr.Slider(
label="noise_scale", minimum=0.1, maximum=1.0, step=0.1, value=0.6, interactive=True)
nsw = gr.Slider(label="noise_scale_w", minimum=0.1,
maximum=1.0, step=0.1, value=0.668, interactive=True)
duration_slider = gr.Slider(minimum=0.1, maximum=5, value=1, step=0.1,
label='速度 Speed')
with gr.Column():
text_output = gr.Textbox(label="Message")
audio_output = gr.Audio(
label="Output Audio", elem_id="tts-audio")
btn = gr.Button("Generate!")
btn.click(tts_fn,
inputs=[textbox, char_dropdown, language_dropdown, ns, nsw, duration_slider,
symbol_input],
outputs=[text_output, audio_output])
gr.Examples(
examples=example,
inputs=[textbox, char_dropdown, language_dropdown,
duration_slider, symbol_input],
outputs=[text_output, audio_output],
fn=tts_fn
)
with gr.Tab("Voice Conversion"):
for i, (description, speakers, vc_fn) in enumerate(
models_vc):
gr.Markdown("""
录制或上传声音，并选择要转换的音色。
""")
with gr.Column():
record_audio = gr.Audio(
label="record your voice", source="microphone")
upload_audio = gr.Audio(
label="or upload audio here", source="upload")
source_speaker = gr.Dropdown(
choices=speakers, value=speakers[0], label="source speaker")
target_speaker = gr.Dropdown(
choices=speakers, value=speakers[0], label="target speaker")
with gr.Column():
message_box = gr.Textbox(label="Message")
converted_audio = gr.Audio(
label='converted audio')
btn = gr.Button("Convert!")
btn.click(vc_fn, inputs=[source_speaker, target_speaker, record_audio, upload_audio],
outputs=[message_box, converted_audio])
for category, link in others.items():
with gr.TabItem(category):
gr.Markdown(
f'''
<center>
<h2>Click to Go</h2>
<a href="{link}">
<img src="https://huggingface.co/datasets/huggingface/badges/raw/main/open-in-hf-spaces-xl-dark.svg"
</a>
</center>
'''
)
app.queue(concurrency_count=3).launch(show_api=False, share=args.share)
|