File size: 12,049 Bytes
1cb796d b4f6525 1cb796d b4f6525 1cb796d b4f6525 1cb796d b4f6525 1cb796d |
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 |
import os
import json
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 infer_pack.models import SynthesizerTrnMs256NSFsid, SynthesizerTrnMs256NSFsid_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
def create_vc_fn(tgt_sr, net_g, vc, if_f0, file_index):
def vc_fn(
input_audio,
f0_up_key,
f0_method,
index_rate,
tts_mode,
tts_text,
tts_voice
):
try:
if tts_mode:
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)
else:
if config.files:
audio, sr = librosa.load(input_audio, sr=16000, mono=True)
else:
if input_audio is None:
return "You need to upload an audio", None
sampling_rate, audio = input_audio
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)
times = [0, 0, 0]
f0_up_key = int(f0_up_key)
audio_opt = vc.pipeline(
hubert_model,
net_g,
0,
audio,
times,
f0_up_key,
f0_method,
file_index,
index_rate,
if_f0,
f0_file=None,
)
print(
f"[{datetime.now().strftime('%Y-%m-%d %H:%M')}]: npy: {times[0]}, f0: {times[1]}s, infer: {times[2]}s"
)
return "Success", (tgt_sr, audio_opt)
except:
info = traceback.format_exc()
print(info)
return info, (None, None)
return vc_fn
def cut_vocal_and_inst(yt_url):
if yt_url != "":
if not os.path.exists("/content/youtube_audio"):
os.mkdir("/content/youtube_audio")
ydl_opts = {
'format': 'bestaudio/best',
'postprocessors': [{
'key': 'FFmpegExtractAudio',
'preferredcodec': 'wav',
}],
"outtmpl": '/content/youtube_audio/audio',
}
with yt_dlp.YoutubeDL(ydl_opts) as ydl:
ydl.download([yt_url])
yt_audio_path = "/content/youtube_audio/audio.wav"
command = f"demucs --two-stems=vocals {yt_audio_path}"
result = subprocess.run(command.split(), stdout=subprocess.PIPE)
print(result.stdout.decode())
return ("/content/rvc-models/separated/htdemucs/audio/vocals.wav", "/content/rvc-models/separated/htdemucs/audio/no_vocals.wav", yt_audio_path, "/content/rvc-models/separated/htdemucs/audio/vocals.wav")
def combine_vocal_and_inst(audio_data, audio_volume):
print(audio_data)
if not os.path.exists("/content/result"):
os.mkdir("/content/result")
vocal_path = "/content/result/output.wav"
inst_path = "/content/rvc-models/separated/htdemucs/audio/no_vocals.wav"
output_path = "/content/result/combine.mp3"
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)
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_to_tts_mode(tts_mode):
if tts_mode:
return gr.Audio.update(visible=False), gr.Textbox.update(visible=True), gr.Dropdown.update(visible=True)
else:
return gr.Audio.update(visible=True), gr.Textbox.update(visible=False), gr.Dropdown.update(visible=False)
'''
def change_audio_to_path_mode(audio_mode):
if audio_mode:
return gr.Audio.update(visible=False), gr.Textbox.update(visible=True)
else:
return gr.Audio.update(visible=True), gr.Textbox.update(visible=False)
'''
if __name__ == '__main__':
load_hubert()
models = []
tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
voices = [f"{v['ShortName']}-{v['Gender']}" for v in tts_voice_list]
folder_path = "weights"
for name in os.listdir(folder_path):
print("check folder: " + name)
if name.startswith("."): break
cover_path = glob.glob(f"{folder_path}/{name}/*.png") + glob.glob(f"{folder_path}/{name}/*.jpg")
index_path = glob.glob(f"{folder_path}/{name}/*.index")
checkpoint_path = glob.glob(f"{folder_path}/{name}/*.pth")
title = name
cover = cover_path[0]
index = index_path[0]
cpt = torch.load(checkpoint_path[0], 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)
if if_f0 == 1:
net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=config.is_half)
else:
net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
del net_g.enc_q
print(net_g.load_state_dict(cpt["weight"], strict=False)) # 不加这一行清不干净, 真奇葩
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)
models.append((name, title, author, cover, create_vc_fn(tgt_sr, net_g, vc, if_f0, index)))
with gr.Blocks() as app:
gr.Markdown(
"# <center> RVC Models (Latest Update)\n"
"## <center> The input audio should be clean and pure voice without background music.\n"
"### <center> More feature will be added soon... \n"
"<center> [](https://colab.research.google.com/drive/1hx6kKvIuv5XNY1Gai2PEuZhpO5z6xpVh?usp=sharing)\n\n"
"<center> [](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI)"
)
with gr.Tabs():
for (name, title, author, cover, vc_fn) in models:
with gr.TabItem(name):
with gr.Row():
gr.Markdown(
'<div align="center">'
f'<div>{title}</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():
if config.files:
with gr.Column():
vc_youtube = gr.Textbox(label="Youtube URL")
vc_convert = gr.Button("Convert", variant="primary")
vc_vocal_preview = gr.Audio(label="Vocal Preview")
vc_inst_preview = gr.Audio(label="Instrumental Preview")
vc_audio_preview = gr.Audio(label="Audio Preview")
with gr.Column():
# vc_audio_mode = gr.Checkbox(label="Use audio path", value=False)
if config.files:
vc_input = gr.Textbox(label="Input audio path")
else:
vc_input = gr.Audio(label="Input audio"+' (less than 20 seconds)' if limitation else '')
vc_transpose = gr.Number(label="Transpose", value=0)
vc_f0method = gr.Radio(
label="Pitch extraction algorithm, PM is fast but Harvest is better for low frequencies",
choices=["pm", "harvest"],
value="pm",
interactive=True,
)
vc_index_ratio = gr.Slider(
minimum=0,
maximum=1,
label="Retrieval feature ratio",
value=0.6,
interactive=True,
)
tts_mode = gr.Checkbox(label="tts (use edge-tts as input)", value=False)
tts_text = gr.Textbox(visible=False,label="TTS text (100 words limitation)" if limitation else "TTS text")
tts_voice = gr.Dropdown(label="Edge-tts speaker", choices=voices, visible=False, allow_custom_value=False, value="en-US-AnaNeural-Female")
vc_submit = gr.Button("Generate", variant="primary")
vc_output1 = gr.Textbox(label="Output Message")
vc_output2 = gr.Audio(label="Output Audio")
if config.files:
with gr.Column():
vc_volume = gr.Slider(
minimum=0,
maximum=10,
label="Vocal volume",
value=5,
interactive=True,
step=1
)
vc_outputCombine = gr.Audio(label="Output Combined Audio")
vc_combine = gr.Button("Combine",variant="primary")
vc_submit.click(vc_fn, [vc_input, vc_transpose, vc_f0method, vc_index_ratio, tts_mode, tts_text, tts_voice], [vc_output1, vc_output2])
tts_mode.change(change_to_tts_mode, [tts_mode], [vc_input, tts_text, tts_voice])
# vc_audio_input(change_audio_to_path_mode, [vc_audio_mode])
if config.files:
vc_convert.click(cut_vocal_and_inst, vc_youtube, [vc_vocal_preview, vc_inst_preview, vc_audio_preview, vc_input])
vc_combine.click(combine_vocal_and_inst, [vc_output2, vc_volume], vc_outputCombine)
app.queue(concurrency_count=1, max_size=20, api_open=config.api).launch(share=config.share) |