File size: 22,934 Bytes
ebb199f |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 |
import os
import glob
import json
import argparse
import traceback
import logging
import gradio as gr
import numpy as np
import librosa
import torch
import asyncio
import edge_tts
import yt_dlp
import ffmpeg
import subprocess
import sys
import io
import wave
from datetime import datetime
from fairseq import checkpoint_utils
from lib.infer_pack.models import (
SynthesizerTrnMs256NSFsid,
SynthesizerTrnMs256NSFsid_nono,
SynthesizerTrnMs768NSFsid,
SynthesizerTrnMs768NSFsid_nono,
)
from vc_infer_pipeline import VC
from config import Config
config = Config()
logging.getLogger("numba").setLevel(logging.WARNING)
limitation = os.getenv("SYSTEM") == "spaces" # limit audio length in huggingface spaces
audio_mode = []
f0method_mode = []
f0method_info = ""
if limitation is True:
audio_mode = ["Upload audio", "TTS Audio"]
f0method_mode = ["pm", "harvest"]
f0method_info = "PM is fast, Harvest is good but extremely slow, Rvmpe is alternative to harvest (might be better). (Default: PM)"
else:
audio_mode = ["Input path", "Upload audio", "Youtube", "TTS Audio"]
f0method_mode = ["pm", "harvest", "crepe"]
f0method_info = "PM is fast, Harvest is good but extremely slow, Rvmpe is alternative to harvest (might be better), and Crepe effect is good but requires GPU (Default: PM)"
if os.path.isfile("rmvpe.pt"):
f0method_mode.insert(2, "rmvpe")
def create_vc_fn(model_name, tgt_sr, net_g, vc, if_f0, version, file_index):
def vc_fn(
vc_audio_mode,
vc_input,
vc_upload,
tts_text,
tts_voice,
f0_up_key,
f0_method,
index_rate,
filter_radius,
resample_sr,
rms_mix_rate,
protect
):
try:
print(f"Converting using {model_name}...")
if vc_audio_mode == "Input path" or "Youtube" and vc_input != "":
audio, sr = librosa.load(vc_input, sr=16000, mono=True)
elif vc_audio_mode == "Upload audio":
if vc_upload is None:
return "You need to upload an audio", None
sampling_rate, audio = vc_upload
duration = audio.shape[0] / sampling_rate
if duration > 20 and limitation:
return "Please upload an audio file that is less than 20 seconds. If you need to generate a longer audio file, please use Colab.", None
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 != 16000:
audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
elif vc_audio_mode == "TTS Audio":
if len(tts_text) > 100 and limitation:
return "Text is too long", None
if tts_text is None or tts_voice is None:
return "You need to enter text and select a voice", None
asyncio.run(edge_tts.Communicate(tts_text, "-".join(tts_voice.split('-')[:-1])).save("tts.mp3"))
audio, sr = librosa.load("tts.mp3", sr=16000, mono=True)
vc_input = "tts.mp3"
times = [0, 0, 0]
f0_up_key = int(f0_up_key)
audio_opt = vc.pipeline(
hubert_model,
net_g,
0,
audio,
vc_input,
times,
f0_up_key,
f0_method,
file_index,
# file_big_npy,
index_rate,
if_f0,
filter_radius,
tgt_sr,
resample_sr,
rms_mix_rate,
version,
protect,
f0_file=None,
)
info = f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s"
print(info)
return info, (tgt_sr, audio_opt)
except:
info = traceback.format_exc()
print(f"{model_name} | {info}")
return info, (tgt_sr, audio_opt)
return vc_fn
def load_model():
models = []
with open(f"weights/model_info.json", "r", encoding="utf-8") as f:
models_info = json.load(f)
for character_name, info in models_info.items():
if not info['enable']:
continue
model_title = info['title']
model_name = info['model_path']
model_author = info.get("author", None)
model_cover = f"weights/{character_name}/{info['cover']}"
model_index = f"weights/{character_name}/{info['feature_retrieval_library']}"
cpt = torch.load(f"weights/{character_name}/{model_name}", map_location="cpu")
tgt_sr = cpt["config"][-1]
cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0] # n_spk
if_f0 = cpt.get("f0", 1)
version = cpt.get("version", "v1")
if version == "v1":
if if_f0 == 1:
net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
else:
net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
model_version = "V1"
elif version == "v2":
if if_f0 == 1:
net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=config.is_half)
else:
net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
model_version = "V2"
del net_g.enc_q
print(net_g.load_state_dict(cpt["weight"], strict=False))
net_g.eval().to(config.device)
if config.is_half:
net_g = net_g.half()
else:
net_g = net_g.float()
vc = VC(tgt_sr, config)
print(f"Model loaded: {character_name} / {info['feature_retrieval_library']} | ({model_version})")
models.append((character_name, model_title, model_author, model_cover, model_version, create_vc_fn(model_title, tgt_sr, net_g, vc, if_f0, version, model_index)))
return models
def cut_vocal_and_inst(url, audio_provider, split_model):
if url != "":
if not os.path.exists("dl_audio"):
os.mkdir("dl_audio")
if audio_provider == "Youtube":
ydl_opts = {
'noplaylist': True,
'format': 'bestaudio/best',
'postprocessors': [{
'key': 'FFmpegExtractAudio',
'preferredcodec': 'wav',
}],
"outtmpl": 'dl_audio/youtube_audio',
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([url])
audio_path = "dl_audio/youtube_audio.wav"
if split_model == "htdemucs":
command = f"demucs --two-stems=vocals {audio_path} -o output"
result = subprocess.run(command.split(), stdout=subprocess.PIPE)
print(result.stdout.decode())
return "output/htdemucs/youtube_audio/vocals.wav", "output/htdemucs/youtube_audio/no_vocals.wav", audio_path, "output/htdemucs/youtube_audio/vocals.wav"
else:
command = f"demucs --two-stems=vocals -n mdx_extra_q {audio_path} -o output"
result = subprocess.run(command.split(), stdout=subprocess.PIPE)
print(result.stdout.decode())
return "output/mdx_extra_q/youtube_audio/vocals.wav", "output/mdx_extra_q/youtube_audio/no_vocals.wav", audio_path, "output/mdx_extra_q/youtube_audio/vocals.wav"
else:
raise gr.Error("URL Required!")
return None, None, None, None
def combine_vocal_and_inst(audio_data, audio_volume, split_model):
if not os.path.exists("output/result"):
os.mkdir("output/result")
vocal_path = "output/result/output.wav"
output_path = "output/result/combine.mp3"
if split_model == "htdemucs":
inst_path = "output/htdemucs/youtube_audio/no_vocals.wav"
else:
inst_path = "output/mdx_extra_q/youtube_audio/no_vocals.wav"
with wave.open(vocal_path, "w") as wave_file:
wave_file.setnchannels(1)
wave_file.setsampwidth(2)
wave_file.setframerate(audio_data[0])
wave_file.writeframes(audio_data[1].tobytes())
command = f'ffmpeg -y -i {inst_path} -i {vocal_path} -filter_complex [1:a]volume={audio_volume}dB[v];[0:a][v]amix=inputs=2:duration=longest -b:a 320k -c:a libmp3lame {output_path}'
result = subprocess.run(command.split(), stdout=subprocess.PIPE)
print(result.stdout.decode())
return output_path
def load_hubert():
global hubert_model
models, _, _ = checkpoint_utils.load_model_ensemble_and_task(
["hubert_base.pt"],
suffix="",
)
hubert_model = models[0]
hubert_model = hubert_model.to(config.device)
if config.is_half:
hubert_model = hubert_model.half()
else:
hubert_model = hubert_model.float()
hubert_model.eval()
def change_audio_mode(vc_audio_mode):
if vc_audio_mode == "Input path":
return (
# Input & Upload
gr.Textbox.update(visible=True),
gr.Checkbox.update(visible=False),
gr.Audio.update(visible=False),
# Youtube
gr.Dropdown.update(visible=False),
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False),
gr.Button.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Slider.update(visible=False),
gr.Audio.update(visible=False),
gr.Button.update(visible=False),
# TTS
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False)
)
elif vc_audio_mode == "Upload audio":
return (
# Input & Upload
gr.Textbox.update(visible=False),
gr.Checkbox.update(visible=True),
gr.Audio.update(visible=True),
# Youtube
gr.Dropdown.update(visible=False),
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False),
gr.Button.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Slider.update(visible=False),
gr.Audio.update(visible=False),
gr.Button.update(visible=False),
# TTS
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False)
)
elif vc_audio_mode == "Youtube":
return (
# Input & Upload
gr.Textbox.update(visible=False),
gr.Checkbox.update(visible=False),
gr.Audio.update(visible=False),
# Youtube
gr.Dropdown.update(visible=True),
gr.Textbox.update(visible=True),
gr.Dropdown.update(visible=True),
gr.Button.update(visible=True),
gr.Audio.update(visible=True),
gr.Audio.update(visible=True),
gr.Audio.update(visible=True),
gr.Slider.update(visible=True),
gr.Audio.update(visible=True),
gr.Button.update(visible=True),
# TTS
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False)
)
elif vc_audio_mode == "TTS Audio":
return (
# Input & Upload
gr.Textbox.update(visible=False),
gr.Checkbox.update(visible=False),
gr.Audio.update(visible=False),
# Youtube
gr.Dropdown.update(visible=False),
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False),
gr.Button.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Slider.update(visible=False),
gr.Audio.update(visible=False),
gr.Button.update(visible=False),
# TTS
gr.Textbox.update(visible=True),
gr.Dropdown.update(visible=True)
)
else:
return (
# Input & Upload
gr.Textbox.update(visible=False),
gr.Checkbox.update(visible=True),
gr.Audio.update(visible=True),
# Youtube
gr.Dropdown.update(visible=False),
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False),
gr.Button.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Audio.update(visible=False),
gr.Slider.update(visible=False),
gr.Audio.update(visible=False),
gr.Button.update(visible=False),
# TTS
gr.Textbox.update(visible=False),
gr.Dropdown.update(visible=False)
)
def use_microphone(microphone):
if microphone == True:
return gr.Audio.update(source="microphone")
else:
return gr.Audio.update(source="upload")
if __name__ == '__main__':
load_hubert()
models = load_model()
tts_voice_list = asyncio.new_event_loop().run_until_complete(edge_tts.list_voices())
voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list]
with gr.Blocks() as app:
gr.Markdown(
"# <center> Combined Genshin Impact RVC Models\n"
"## <center> The input audio should be clean and pure voice without background music.\n"
"### <center> It is recommended to use google colab for more features. \n"
"[](https://colab.research.google.com/drive/1Tgr6q9kKiB5P37rUitrB3CsNl8JP9iQZ?usp=sharing)\n\n"
"[](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI)"
)
with gr.Tabs():
for (name, title, author, cover, model_version, vc_fn) in models:
with gr.TabItem(name):
with gr.Row():
gr.Markdown(
'<div align="center">'
f'<div>{title}</div>\n'+
f'<div>RVC {model_version} Model</div>\n'+
(f'<div>Model author: {author}</div>' if author else "")+
(f'<img style="width:auto;height:300px;" src="file/{cover}">' if cover else "")+
'</div>'
)
with gr.Row():
with gr.Column():
vc_audio_mode = gr.Dropdown(label="Input voice", choices=audio_mode, allow_custom_value=False, value="Upload audio")
# Input
vc_input = gr.Textbox(label="Input audio path", visible=False)
# Upload
vc_microphone_mode = gr.Checkbox(label="Use Microphone", value=False, visible=True, interactive=True)
vc_upload = gr.Audio(label="Upload audio file", source="upload", visible=True, interactive=True)
# Youtube
vc_download_audio = gr.Dropdown(label="Provider", choices=["Youtube"], allow_custom_value=False, visible=False, value="Youtube", info="Select provider (Default: Youtube)")
vc_link = gr.Textbox(label="Youtube URL", visible=False, info="Example: https://www.youtube.com/watch?v=Nc0sB1Bmf-A", placeholder="https://www.youtube.com/watch?v=...")
vc_split_model = gr.Dropdown(label="Splitter Model", choices=["htdemucs", "mdx_extra_q"], allow_custom_value=False, visible=False, value="htdemucs", info="Select the splitter model (Default: htdemucs)")
vc_split = gr.Button("Split Audio", variant="primary", visible=False)
vc_vocal_preview = gr.Audio(label="Vocal Preview", visible=False)
vc_inst_preview = gr.Audio(label="Instrumental Preview", visible=False)
vc_audio_preview = gr.Audio(label="Audio Preview", visible=False)
# TTS
tts_text = gr.Textbox(visible=False, label="TTS text", info="Text to speech input")
tts_voice = gr.Dropdown(label="Edge-tts speaker", choices=voices, visible=False, allow_custom_value=False, value="en-US-AnaNeural-Female")
with gr.Column():
vc_transform0 = gr.Number(label="Transpose", value=0, info='Type "12" to change from male to female voice. Type "-12" to change female to male voice')
f0method0 = gr.Radio(
label="Pitch extraction algorithm",
info=f0method_info,
choices=f0method_mode,
value="pm",
interactive=True
)
index_rate1 = gr.Slider(
minimum=0,
maximum=1,
label="Retrieval feature ratio",
info="(Default: 0.7)",
value=0.7,
interactive=True,
)
filter_radius0 = gr.Slider(
minimum=0,
maximum=7,
label="Apply Median Filtering",
info="The value represents the filter radius and can reduce breathiness.",
value=3,
step=1,
interactive=True,
)
resample_sr0 = gr.Slider(
minimum=0,
maximum=48000,
label="Resample the output audio",
info="Resample the output audio in post-processing to the final sample rate. Set to 0 for no resampling",
value=0,
step=1,
interactive=True,
)
rms_mix_rate0 = gr.Slider(
minimum=0,
maximum=1,
label="Volume Envelope",
info="Use the volume envelope of the input to replace or mix with the volume envelope of the output. The closer the ratio is to 1, the more the output envelope is used",
value=1,
interactive=True,
)
protect0 = gr.Slider(
minimum=0,
maximum=0.5,
label="Voice Protection",
info="Protect voiceless consonants and breath sounds to prevent artifacts such as tearing in electronic music. Set to 0.5 to disable. Decrease the value to increase protection, but it may reduce indexing accuracy",
value=0.5,
step=0.01,
interactive=True,
)
with gr.Column():
vc_log = gr.Textbox(label="Output Information", interactive=False)
vc_output = gr.Audio(label="Output Audio", interactive=False)
vc_convert = gr.Button("Convert", variant="primary")
vc_volume = gr.Slider(
minimum=0,
maximum=10,
label="Vocal volume",
value=4,
interactive=True,
step=1,
info="Adjust vocal volume (Default: 4}",
visible=False
)
vc_combined_output = gr.Audio(label="Output Combined Audio", visible=False)
vc_combine = gr.Button("Combine",variant="primary", visible=False)
vc_convert.click(
fn=vc_fn,
inputs=[
vc_audio_mode,
vc_input,
vc_upload,
tts_text,
tts_voice,
vc_transform0,
f0method0,
index_rate1,
filter_radius0,
resample_sr0,
rms_mix_rate0,
protect0,
],
outputs=[vc_log ,vc_output]
)
vc_split.click(
fn=cut_vocal_and_inst,
inputs=[vc_link, vc_download_audio, vc_split_model],
outputs=[vc_vocal_preview, vc_inst_preview, vc_audio_preview, vc_input]
)
vc_combine.click(
fn=combine_vocal_and_inst,
inputs=[vc_output, vc_volume, vc_split_model],
outputs=[vc_combined_output]
)
vc_microphone_mode.change(
fn=use_microphone,
inputs=vc_microphone_mode,
outputs=vc_upload
)
vc_audio_mode.change(
fn=change_audio_mode,
inputs=[vc_audio_mode],
outputs=[
vc_input,
vc_microphone_mode,
vc_upload,
vc_download_audio,
vc_link,
vc_split_model,
vc_split,
vc_vocal_preview,
vc_inst_preview,
vc_audio_preview,
vc_volume,
vc_combined_output,
vc_combine,
tts_text,
tts_voice
]
)
app.queue(concurrency_count=1, max_size=20, api_open=config.api).launch(share=config.colab) |