2023-04-11 00:11:14 +08:00
|
|
|
import PySimpleGUI as sg
|
|
|
|
import sounddevice as sd
|
2023-04-12 14:28:28 +08:00
|
|
|
import noisereduce as nr
|
2023-04-11 00:11:14 +08:00
|
|
|
import numpy as np
|
|
|
|
from fairseq import checkpoint_utils
|
2023-04-13 10:15:11 +08:00
|
|
|
import librosa,torch,parselmouth,faiss,time,threading,math
|
2023-04-11 00:11:14 +08:00
|
|
|
import torch.nn.functional as F
|
2023-04-13 10:15:11 +08:00
|
|
|
import torchaudio.transforms as tat
|
|
|
|
|
2023-04-11 00:11:14 +08:00
|
|
|
#import matplotlib.pyplot as plt
|
|
|
|
from infer_pack.models import SynthesizerTrnMs256NSFsid, SynthesizerTrnMs256NSFsid_nono
|
2023-04-12 10:48:39 +08:00
|
|
|
from webui_locale import I18nAuto
|
|
|
|
i18n = I18nAuto()
|
2023-04-11 00:11:14 +08:00
|
|
|
|
|
|
|
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
|
|
|
|
|
|
|
class RVC:
|
|
|
|
def __init__(self,key,pth_path,index_path,npy_path) -> None:
|
|
|
|
'''
|
|
|
|
初始化
|
|
|
|
'''
|
|
|
|
self.f0_up_key=key
|
|
|
|
self.time_step = 160 / 16000 * 1000
|
|
|
|
self.f0_min = 50
|
|
|
|
self.f0_max = 1100
|
|
|
|
self.f0_mel_min = 1127 * np.log(1 + self.f0_min / 700)
|
|
|
|
self.f0_mel_max = 1127 * np.log(1 + self.f0_max / 700)
|
|
|
|
self.index=faiss.read_index(index_path)
|
|
|
|
self.big_npy=np.load(npy_path)
|
|
|
|
model_path = "TEMP\\hubert_base.pt"
|
|
|
|
print("load model(s) from {}".format(model_path))
|
|
|
|
models, saved_cfg, task = checkpoint_utils.load_model_ensemble_and_task(
|
|
|
|
[model_path],
|
|
|
|
suffix="",
|
|
|
|
)
|
|
|
|
self.model = models[0]
|
|
|
|
self.model = self.model.to(device)
|
|
|
|
self.model = self.model.half()
|
|
|
|
self.model.eval()
|
|
|
|
cpt = torch.load(pth_path, 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):
|
|
|
|
self.net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=True)
|
|
|
|
else:
|
|
|
|
self.net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
|
|
|
|
del self.net_g.enc_q
|
|
|
|
print(self.net_g.load_state_dict(cpt["weight"], strict=False))
|
|
|
|
self.net_g.eval().to(device)
|
|
|
|
self.net_g.half()
|
|
|
|
|
|
|
|
|
|
|
|
def get_f0_coarse(self,f0):
|
|
|
|
f0_mel = 1127 * np.log(1 + f0 / 700)
|
|
|
|
f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - self.f0_mel_min) * 254 / (self.f0_mel_max - self.f0_mel_min) + 1
|
|
|
|
f0_mel[f0_mel <= 1] = 1
|
|
|
|
f0_mel[f0_mel > 255] = 255
|
|
|
|
# f0_mel[f0_mel > 188] = 188
|
|
|
|
f0_coarse = np.rint(f0_mel).astype(np.int)
|
|
|
|
return f0_coarse
|
|
|
|
|
|
|
|
def get_f0(self,x, p_len,f0_up_key=0):
|
|
|
|
f0 = parselmouth.Sound(x, 16000).to_pitch_ac(
|
|
|
|
time_step=self.time_step / 1000, voicing_threshold=0.6,
|
|
|
|
pitch_floor=self.f0_min, pitch_ceiling=self.f0_max).selected_array['frequency']
|
|
|
|
|
|
|
|
pad_size=(p_len - len(f0) + 1) // 2
|
|
|
|
if(pad_size>0 or p_len - len(f0) - pad_size>0):
|
|
|
|
f0 = np.pad(f0,[[pad_size,p_len - len(f0) - pad_size]], mode='constant')
|
|
|
|
f0 *= pow(2, f0_up_key / 12)
|
|
|
|
# f0=suofang(f0)
|
|
|
|
f0bak = f0.copy()
|
|
|
|
f0_coarse=self.get_f0_coarse(f0)
|
|
|
|
return f0_coarse, f0bak
|
|
|
|
|
|
|
|
|
|
|
|
def infer(self,audio:np.ndarray,sampling_rate:int) -> np.ndarray:
|
|
|
|
'''
|
|
|
|
推理函数。
|
|
|
|
:param audio: ndarray(n,2)
|
|
|
|
:sampling_rate: 采样率
|
|
|
|
'''
|
|
|
|
|
|
|
|
# f0_up_key=12
|
|
|
|
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)
|
2023-04-13 10:15:11 +08:00
|
|
|
#print('test:audio:'+str(audio.shape))
|
2023-04-11 00:11:14 +08:00
|
|
|
'''padding'''
|
|
|
|
|
|
|
|
|
|
|
|
feats = torch.from_numpy(audio).float()
|
|
|
|
if feats.dim() == 2: # double channels
|
|
|
|
feats = feats.mean(-1)
|
|
|
|
assert feats.dim() == 1, feats.dim()
|
|
|
|
feats = feats.view(1, -1)
|
|
|
|
padding_mask = torch.BoolTensor(feats.shape).fill_(False)
|
|
|
|
inputs = {
|
|
|
|
"source": feats.half().to(device),
|
|
|
|
"padding_mask": padding_mask.to(device),
|
|
|
|
"output_layer": 9, # layer 9
|
|
|
|
}
|
|
|
|
torch.cuda.synchronize()
|
|
|
|
with torch.no_grad():
|
|
|
|
logits = self.model.extract_features(**inputs)
|
|
|
|
feats = self.model.final_proj(logits[0])
|
|
|
|
|
|
|
|
####索引优化
|
|
|
|
npy = feats[0].cpu().numpy().astype("float32")
|
|
|
|
D, I = self.index.search(npy, 1)
|
|
|
|
# feats = torch.from_numpy(big_npy[I.squeeze()].astype("float16")).unsqueeze(0).to(device)
|
|
|
|
index_rate=0.5
|
|
|
|
feats = torch.from_numpy(npy).unsqueeze(0).to(device) * index_rate + (1 - index_rate) * feats
|
|
|
|
feats=feats.half()
|
|
|
|
|
|
|
|
feats=F.interpolate(feats.permute(0,2,1),scale_factor=2).permute(0,2,1)
|
|
|
|
torch.cuda.synchronize()
|
|
|
|
# p_len = min(feats.shape[1],10000,pitch.shape[0])#太大了爆显存
|
|
|
|
p_len = min(feats.shape[1],12000)#
|
|
|
|
pitch, pitchf = self.get_f0(audio, p_len,self.f0_up_key)
|
|
|
|
p_len = min(feats.shape[1],12000,pitch.shape[0])#太大了爆显存
|
|
|
|
torch.cuda.synchronize()
|
|
|
|
# print(feats.shape,pitch.shape)
|
|
|
|
feats = feats[:,:p_len, :]
|
|
|
|
pitch = pitch[:p_len]
|
|
|
|
pitchf = pitchf[:p_len]
|
|
|
|
p_len = torch.LongTensor([p_len]).to(device)
|
|
|
|
pitch = torch.LongTensor(pitch).unsqueeze(0).to(device)
|
|
|
|
pitchf = torch.FloatTensor(pitchf).unsqueeze(0).to(device)
|
|
|
|
ii=0#sid
|
|
|
|
sid=torch.LongTensor([ii]).to(device)
|
|
|
|
with torch.no_grad():
|
|
|
|
audio = self.net_g.infer(feats, p_len,pitch,pitchf,sid)[0][0, 0].data.cpu().float().numpy()#nsf
|
|
|
|
torch.cuda.synchronize()
|
|
|
|
return audio
|
|
|
|
|
|
|
|
|
|
|
|
class Config:
|
|
|
|
def __init__(self) -> None:
|
|
|
|
self.pth_path:str=''
|
|
|
|
self.index_path:str=''
|
|
|
|
self.npy_path:str=''
|
|
|
|
self.pitch:int=12
|
|
|
|
self.samplerate:int=44100
|
|
|
|
self.block_time:float=1.0#s
|
|
|
|
self.buffer_num:int=1
|
|
|
|
self.threhold:int=-30
|
|
|
|
self.crossfade_time:float=0.08
|
|
|
|
self.extra_time:float=0.04
|
2023-04-13 10:15:11 +08:00
|
|
|
self.I_noise_reduce=False
|
|
|
|
self.O_noise_reduce=False
|
2023-04-11 00:11:14 +08:00
|
|
|
|
|
|
|
class GUI:
|
|
|
|
def __init__(self) -> None:
|
|
|
|
self.config=Config()
|
|
|
|
self.flag_vc=False
|
|
|
|
|
|
|
|
self.launcher()
|
|
|
|
|
|
|
|
def launcher(self):
|
|
|
|
sg.theme('LightBlue3')
|
|
|
|
input_devices,output_devices,_, _=self.get_devices()
|
|
|
|
layout=[
|
|
|
|
[
|
2023-04-12 16:53:50 +08:00
|
|
|
sg.Frame(title=i18n('加载模型'),layout=[
|
2023-04-13 10:15:11 +08:00
|
|
|
[sg.Input(default_text='TEMP\\hubert_base.pt',key='hubert_path'),sg.FileBrowse(i18n('Hubert File'))],
|
2023-04-12 16:53:50 +08:00
|
|
|
[sg.Input(default_text='TEMP\\atri.pth',key='pth_path'),sg.FileBrowse(i18n('选择.pth文件'))],
|
|
|
|
[sg.Input(default_text='TEMP\\added_IVF512_Flat_atri_baseline_src_feat.index',key='index_path'),sg.FileBrowse(i18n('选择.index文件'))],
|
|
|
|
[sg.Input(default_text='TEMP\\big_src_feature_atri.npy',key='npy_path'),sg.FileBrowse(i18n('选择.npy文件'))]
|
2023-04-11 00:11:14 +08:00
|
|
|
])
|
|
|
|
],
|
|
|
|
[
|
|
|
|
sg.Frame(layout=[
|
2023-04-12 16:53:50 +08:00
|
|
|
[sg.Text(i18n("输入设备")),sg.Combo(input_devices,key='sg_input_device',default_value=input_devices[sd.default.device[0]])],
|
|
|
|
[sg.Text(i18n("输出设备")),sg.Combo(output_devices,key='sg_output_device',default_value=output_devices[sd.default.device[1]])]
|
|
|
|
],title=i18n("音频设备(请使用同种类驱动)"))
|
2023-04-11 00:11:14 +08:00
|
|
|
],
|
|
|
|
[
|
|
|
|
sg.Frame(layout=[
|
2023-04-12 16:53:50 +08:00
|
|
|
[sg.Text(i18n("响应阈值")),sg.Slider(range=(-60,0),key='threhold',resolution=1,orientation='h',default_value=-30)],
|
|
|
|
[sg.Text(i18n("音调设置")),sg.Slider(range=(-24,24),key='pitch',resolution=1,orientation='h',default_value=12)]
|
2023-04-11 00:11:14 +08:00
|
|
|
|
2023-04-12 16:53:50 +08:00
|
|
|
],title=i18n("常规设置")),
|
2023-04-11 00:11:14 +08:00
|
|
|
sg.Frame(layout=[
|
2023-04-12 16:53:50 +08:00
|
|
|
[sg.Text(i18n("采样长度")),sg.Slider(range=(0.1,3.0),key='block_time',resolution=0.1,orientation='h',default_value=1.0)],
|
|
|
|
[sg.Text(i18n("淡入淡出长度")),sg.Slider(range=(0.01,0.15),key='crossfade_length',resolution=0.01,orientation='h',default_value=0.08)],
|
|
|
|
[sg.Text(i18n("额外推理时长")),sg.Slider(range=(0.05,3.00),key='extra_time',resolution=0.01,orientation='h',default_value=0.05)],
|
2023-04-13 10:15:11 +08:00
|
|
|
[sg.Checkbox(i18n('Input Noisereduce'),key='I_noise_reduce'),sg.Checkbox(i18n('Output Noisereduce'),key='O_noise_reduce')]
|
2023-04-12 16:53:50 +08:00
|
|
|
],title=i18n("性能设置"))
|
2023-04-11 00:11:14 +08:00
|
|
|
],
|
2023-04-13 10:15:11 +08:00
|
|
|
[sg.Button(i18n("开始音频转换"),key='start_vc'),sg.Button(i18n("停止音频转换"),key='stop_vc'),sg.Text(i18n("Infer Time(ms):")),sg.Text("0",key='infer_time')]
|
2023-04-11 00:11:14 +08:00
|
|
|
]
|
|
|
|
|
2023-04-12 10:48:39 +08:00
|
|
|
self.window=sg.Window("RVC - GUI",layout=layout)
|
2023-04-11 00:11:14 +08:00
|
|
|
self.event_handler()
|
|
|
|
|
|
|
|
def event_handler(self):
|
|
|
|
while True:
|
|
|
|
event, values = self.window.read()
|
|
|
|
if event ==sg.WINDOW_CLOSED:
|
|
|
|
self.flag_vc=False
|
|
|
|
exit()
|
|
|
|
if event == 'start_vc' and self.flag_vc==False:
|
|
|
|
self.set_values(values)
|
|
|
|
print('pth_path:'+self.config.pth_path)
|
|
|
|
print('index_path:'+self.config.index_path)
|
|
|
|
print('npy_path:'+self.config.npy_path)
|
|
|
|
print('using_cuda:'+str(torch.cuda.is_available()))
|
|
|
|
self.start_vc()
|
|
|
|
if event=='stop_vc'and self.flag_vc==True:
|
|
|
|
self.flag_vc = False
|
|
|
|
|
|
|
|
|
|
|
|
def set_values(self,values):
|
|
|
|
self.set_devices(values["sg_input_device"],values['sg_output_device'])
|
|
|
|
self.config.pth_path=values['pth_path']
|
|
|
|
self.config.index_path=values['index_path']
|
|
|
|
self.config.npy_path=values['npy_path']
|
|
|
|
self.config.threhold=values['threhold']
|
|
|
|
self.config.pitch=values['pitch']
|
|
|
|
self.config.block_time=values['block_time']
|
|
|
|
self.config.crossfade_time=values['crossfade_length']
|
|
|
|
self.config.extra_time=values['extra_time']
|
2023-04-13 10:15:11 +08:00
|
|
|
self.config.I_noise_reduce=values['I_noise_reduce']
|
|
|
|
self.config.O_noise_reduce=values['O_noise_reduce']
|
2023-04-11 00:11:14 +08:00
|
|
|
|
|
|
|
def start_vc(self):
|
|
|
|
torch.cuda.empty_cache()
|
|
|
|
self.flag_vc=True
|
2023-04-13 10:15:11 +08:00
|
|
|
self.RMS_threhold=math.e**(float(self.config.threhold)/10)
|
2023-04-11 00:11:14 +08:00
|
|
|
self.block_frame=int(self.config.block_time*self.config.samplerate)
|
|
|
|
self.crossfade_frame=int(self.config.crossfade_time*self.config.samplerate)
|
|
|
|
self.sola_search_frame=int(0.012*self.config.samplerate)
|
|
|
|
self.delay_frame=int(0.02*self.config.samplerate)#往前预留0.02s
|
|
|
|
self.extra_frame=int(self.config.extra_time*self.config.samplerate)#往后预留0.04s
|
|
|
|
self.rvc=None
|
|
|
|
self.rvc=RVC(self.config.pitch,self.config.pth_path,self.config.index_path,self.config.npy_path)
|
2023-04-13 10:15:11 +08:00
|
|
|
self.input_wav:np.ndarray=np.zeros(self.extra_frame+self.crossfade_frame+self.sola_search_frame+self.block_frame,dtype='float32')
|
|
|
|
self.output_wav:torch.Tensor=torch.zeros(self.block_frame,device=device,dtype=torch.float32)
|
|
|
|
#self.sola_buffer:np.ndarray=np.zeros(self.crossfade_frame,dtype='float32')
|
|
|
|
self.sola_buffer:torch.Tensor=torch.zeros(self.crossfade_frame,device=device,dtype=torch.float32)
|
|
|
|
#self.fade_in_window:np.ndarray = np.linspace(0, 1, self.crossfade_frame)
|
|
|
|
self.fade_in_window:torch.Tensor=torch.linspace(0.0,1.0,steps=self.crossfade_frame,device=device,dtype=torch.float32)
|
|
|
|
self.fade_out_window:torch.Tensor = 1 - self.fade_in_window
|
|
|
|
self.resampler=tat.Resample(orig_freq=40000,new_freq=self.config.samplerate,dtype=torch.float32)
|
|
|
|
self.RMS=lambda y:torch.sqrt(torch.mean(torch.square(y))).item()#RMS calculator
|
2023-04-11 00:11:14 +08:00
|
|
|
thread_vc=threading.Thread(target=self.soundinput)
|
|
|
|
thread_vc.start()
|
|
|
|
|
|
|
|
|
|
|
|
def soundinput(self):
|
|
|
|
'''
|
|
|
|
接受音频输入
|
|
|
|
'''
|
|
|
|
with sd.Stream(callback=self.audio_callback, blocksize=self.block_frame,samplerate=self.config.samplerate,dtype='float32'):
|
|
|
|
while self.flag_vc:
|
|
|
|
time.sleep(self.config.block_time)
|
|
|
|
print('Audio block passed.')
|
|
|
|
print('ENDing VC')
|
|
|
|
|
|
|
|
|
|
|
|
def audio_callback(self,indata:np.ndarray,outdata:np.ndarray, frames, times, status):
|
|
|
|
'''
|
|
|
|
音频处理
|
|
|
|
'''
|
|
|
|
start_time=time.perf_counter()
|
|
|
|
indata=librosa.to_mono(indata.T)
|
2023-04-13 10:15:11 +08:00
|
|
|
if self.config.I_noise_reduce:
|
|
|
|
indata[:]=nr.reduce_noise(y=indata,sr=self.config.samplerate)
|
|
|
|
|
2023-04-11 00:11:14 +08:00
|
|
|
'''noise gate'''
|
2023-04-13 10:15:11 +08:00
|
|
|
frame_length=2048
|
|
|
|
hop_length=1024
|
2023-04-11 00:11:14 +08:00
|
|
|
rms=librosa.feature.rms(y=indata,frame_length=frame_length,hop_length=hop_length)
|
|
|
|
db_threhold=librosa.amplitude_to_db(rms,ref=1.0)[0]<self.config.threhold
|
|
|
|
#print(rms.shape,db.shape,db)
|
|
|
|
for i in range(db_threhold.shape[0]):
|
|
|
|
if db_threhold[i]:
|
|
|
|
indata[i*hop_length:(i+1)*hop_length]=0
|
2023-04-13 10:15:11 +08:00
|
|
|
self.input_wav[:]=np.append(self.input_wav[self.block_frame:],indata)
|
2023-04-11 00:11:14 +08:00
|
|
|
|
|
|
|
#infer
|
|
|
|
print('input_wav:'+str(self.input_wav.shape))
|
2023-04-13 10:15:11 +08:00
|
|
|
#print('infered_wav:'+str(infer_wav.shape))
|
|
|
|
infer_wav:torch.Tensor=self.resampler(torch.from_numpy(self.rvc.infer(self.input_wav,self.config.samplerate)))[-self.crossfade_frame-self.sola_search_frame-self.block_frame:].to(device)
|
|
|
|
print('infer_wav:'+str(infer_wav.shape))
|
|
|
|
|
2023-04-11 00:11:14 +08:00
|
|
|
# SOLA algorithm from https://github.com/yxlllc/DDSP-SVC
|
2023-04-13 10:15:11 +08:00
|
|
|
cor_nom=F.conv1d(infer_wav[None,None,:self.crossfade_frame + self.sola_search_frame],self.sola_buffer[None,None,:])
|
|
|
|
cor_den=torch.sqrt(F.conv1d(infer_wav[None,None,:self.crossfade_frame + self.sola_search_frame]**2,torch.ones(1, 1,self.crossfade_frame,device=device))+1e-8)
|
|
|
|
sola_offset = torch.argmax( cor_nom[0, 0] / cor_den[0, 0])
|
|
|
|
print('sola offset: ' + str(int(sola_offset)))
|
2023-04-11 00:11:14 +08:00
|
|
|
|
|
|
|
# crossfade
|
|
|
|
self.output_wav[:]=infer_wav[sola_offset : sola_offset + self.block_frame]
|
|
|
|
self.output_wav[:self.crossfade_frame] *= self.fade_in_window
|
|
|
|
self.output_wav[:self.crossfade_frame] += self.sola_buffer[:]
|
|
|
|
if sola_offset < self.sola_search_frame:
|
|
|
|
self.sola_buffer[:] = infer_wav[-self.sola_search_frame - self.crossfade_frame + sola_offset: -self.sola_search_frame + sola_offset]* self.fade_out_window
|
|
|
|
else:
|
|
|
|
self.sola_buffer[:] = infer_wav[- self.crossfade_frame :]* self.fade_out_window
|
|
|
|
|
2023-04-13 10:15:11 +08:00
|
|
|
if self.config.O_noise_reduce:
|
|
|
|
outdata[:]=np.tile(nr.reduce_noise(y=self.output_wav[:].cpu().numpy(),sr=self.config.samplerate),(2,1)).T
|
|
|
|
else:
|
|
|
|
outdata[:]=self.output_wav[:].repeat(2, 1).t().cpu().numpy()
|
|
|
|
total_time=time.perf_counter()-start_time
|
|
|
|
print('infer time:'+str(total_time))
|
|
|
|
self.window['infer_time'].update(int(total_time*1000))
|
2023-04-11 00:11:14 +08:00
|
|
|
|
|
|
|
def get_devices(self,update: bool = True):
|
|
|
|
'''获取设备列表'''
|
|
|
|
if update:
|
|
|
|
sd._terminate()
|
|
|
|
sd._initialize()
|
|
|
|
devices = sd.query_devices()
|
|
|
|
hostapis = sd.query_hostapis()
|
|
|
|
for hostapi in hostapis:
|
|
|
|
for device_idx in hostapi["devices"]:
|
|
|
|
devices[device_idx]["hostapi_name"] = hostapi["name"]
|
|
|
|
input_devices = [
|
|
|
|
f"{d['name']} ({d['hostapi_name']})"
|
|
|
|
for d in devices
|
|
|
|
if d["max_input_channels"] > 0
|
|
|
|
]
|
|
|
|
output_devices = [
|
|
|
|
f"{d['name']} ({d['hostapi_name']})"
|
|
|
|
for d in devices
|
|
|
|
if d["max_output_channels"] > 0
|
|
|
|
]
|
|
|
|
input_devices_indices = [d["index"] for d in devices if d["max_input_channels"] > 0]
|
|
|
|
output_devices_indices = [
|
|
|
|
d["index"] for d in devices if d["max_output_channels"] > 0
|
|
|
|
]
|
|
|
|
return input_devices, output_devices, input_devices_indices, output_devices_indices
|
|
|
|
|
|
|
|
def set_devices(self,input_device,output_device):
|
|
|
|
'''设置输出设备'''
|
|
|
|
input_devices,output_devices,input_device_indices, output_device_indices=self.get_devices()
|
|
|
|
sd.default.device[0]=input_device_indices[input_devices.index(input_device)]
|
|
|
|
sd.default.device[1]=output_device_indices[output_devices.index(output_device)]
|
|
|
|
print("input device:"+str(sd.default.device[0])+":"+str(input_device))
|
2023-04-12 14:28:28 +08:00
|
|
|
print("output device:"+str(sd.default.device[1])+":"+str(output_device))
|
|
|
|
|
|
|
|
gui=GUI()
|