OPENGIT/Retrieval-based-Voice-Conversion-WebUI
1
0
Fork 0
mirror of https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI.git synced 2025-03-10 15:54:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

Format code (#366)

Browse Source 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
This commit is contained in:
github-actions[bot] 2023-05-28 16:06:11 +00:00 committed by GitHub
parent e569477457
commit e435b3bb8a
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 262 additions and 170 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

121
MDXNet.py
View File

@ -8,9 +8,12 @@ import torch
import torch.nn as nn import torch.nn as nn
dim_c = 4 dim_c = 4
class Conv_TDF_net_trim():
def __init__(self, device, model_name, target_name,
L, dim_f, dim_t, n_fft, hop=1024): class Conv_TDF_net_trim:
def __init__(
self, device, model_name, target_name, L, dim_f, dim_t, n_fft, hop=1024
):
super(Conv_TDF_net_trim, self).__init__() super(Conv_TDF_net_trim, self).__init__()
self.dim_f = dim_f self.dim_f = dim_f
@ -19,57 +22,90 @@ class Conv_TDF_net_trim():
self.hop = hop self.hop = hop
self.n_bins = self.n_fft // 2 + 1 self.n_bins = self.n_fft // 2 + 1
self.chunk_size = hop * (self.dim_t - 1) self.chunk_size = hop * (self.dim_t - 1)
self.window = torch.hann_window(window_length=self.n_fft, periodic=True).to(device) self.window = torch.hann_window(window_length=self.n_fft, periodic=True).to(
device
)
self.target_name = target_name self.target_name = target_name
self.blender = 'blender' in model_name self.blender = "blender" in model_name
out_c = dim_c * 4 if target_name == '*' else dim_c out_c = dim_c * 4 if target_name == "*" else dim_c
self.freq_pad = torch.zeros([1, out_c, self.n_bins - self.dim_f, self.dim_t]).to(device) self.freq_pad = torch.zeros(
[1, out_c, self.n_bins - self.dim_f, self.dim_t]
).to(device)
self.n = L // 2 self.n = L // 2
def stft(self, x): def stft(self, x):
x = x.reshape([-1, self.chunk_size]) x = x.reshape([-1, self.chunk_size])
x = torch.stft(x, n_fft=self.n_fft, hop_length=self.hop, window=self.window, center=True, return_complex=True) x = torch.stft(
x,
n_fft=self.n_fft,
hop_length=self.hop,
window=self.window,
center=True,
return_complex=True,
)
x = torch.view_as_real(x) x = torch.view_as_real(x)
x = x.permute([0, 3, 1, 2]) x = x.permute([0, 3, 1, 2])
x = x.reshape([-1, 2, 2, self.n_bins, self.dim_t]).reshape([-1, dim_c, self.n_bins, self.dim_t]) x = x.reshape([-1, 2, 2, self.n_bins, self.dim_t]).reshape(
[-1, dim_c, self.n_bins, self.dim_t]
)
return x[:, :, : self.dim_f] return x[:, :, : self.dim_f]
def istft(self, x, freq_pad=None): def istft(self, x, freq_pad=None):
freq_pad = self.freq_pad.repeat([x.shape[0], 1, 1, 1]) if freq_pad is None else freq_pad freq_pad = (
self.freq_pad.repeat([x.shape[0], 1, 1, 1])
if freq_pad is None
else freq_pad
)
x = torch.cat([x, freq_pad], -2) x = torch.cat([x, freq_pad], -2)
c = 4 * 2 if self.target_name == '*' else 2 c = 4 * 2 if self.target_name == "*" else 2
x = x.reshape([-1, c, 2, self.n_bins, self.dim_t]).reshape([-1, 2, self.n_bins, self.dim_t]) x = x.reshape([-1, c, 2, self.n_bins, self.dim_t]).reshape(
[-1, 2, self.n_bins, self.dim_t]
)
x = x.permute([0, 2, 3, 1]) x = x.permute([0, 2, 3, 1])
x = x.contiguous() x = x.contiguous()
x = torch.view_as_complex(x) x = torch.view_as_complex(x)
x = torch.istft(x, n_fft=self.n_fft, hop_length=self.hop, window=self.window, center=True) x = torch.istft(
x, n_fft=self.n_fft, hop_length=self.hop, window=self.window, center=True
)
return x.reshape([-1, c, self.chunk_size]) return x.reshape([-1, c, self.chunk_size])
def get_models(device, dim_f, dim_t, n_fft): def get_models(device, dim_f, dim_t, n_fft):
return Conv_TDF_net_trim( return Conv_TDF_net_trim(
device=device, device=device,
model_name='Conv-TDF', target_name='vocals', model_name="Conv-TDF",
target_name="vocals",
L=11, L=11,
dim_f=dim_f, dim_t=dim_t, dim_f=dim_f,
n_fft=n_fft dim_t=dim_t,
n_fft=n_fft,
) )
warnings.filterwarnings("ignore") warnings.filterwarnings("ignore")
cpu = torch.device('cpu') cpu = torch.device("cpu")
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu') device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
class Predictor: class Predictor:
def __init__(self, args): def __init__(self, args):
self.args = args self.args = args
self.model_ = get_models(device=cpu, dim_f=args.dim_f, dim_t=args.dim_t, n_fft=args.n_fft) self.model_ = get_models(
self.model = ort.InferenceSession(os.path.join(args.onnx,self.model_.target_name+'.onnx'), providers=['CUDAExecutionProvider', 'CPUExecutionProvider']) device=cpu, dim_f=args.dim_f, dim_t=args.dim_t, n_fft=args.n_fft
print('onnx load done') )
self.model = ort.InferenceSession(
os.path.join(args.onnx, self.model_.target_name + ".onnx"),
providers=["CUDAExecutionProvider", "CPUExecutionProvider"],
)
print("onnx load done")
def demix(self, mix): def demix(self, mix):
samples = mix.shape[-1] samples = mix.shape[-1]
margin = self.args.margin margin = self.args.margin
chunk_size = self.args.chunks * 44100 chunk_size = self.args.chunks * 44100
assert not margin == 0, 'margin cannot be zero!' assert not margin == 0, "margin cannot be zero!"
if margin > chunk_size: if margin > chunk_size:
margin = chunk_size margin = chunk_size
@ -92,12 +128,13 @@ class Predictor:
break break
sources = self.demix_base(segmented_mix, margin_size=margin) sources = self.demix_base(segmented_mix, margin_size=margin)
''' """
mix:(2,big_sample) mix:(2,big_sample)
segmented_mix:offset->(2,small_sample) segmented_mix:offset->(2,small_sample)
sources:(1,2,big_sample) sources:(1,2,big_sample)
''' """
return sources return sources
def demix_base(self, mixes, margin_size): def demix_base(self, mixes, margin_size):
chunked_sources = [] chunked_sources = []
progress_bar = tqdm(total=len(mixes)) progress_bar = tqdm(total=len(mixes))
@ -110,7 +147,9 @@ class Predictor:
trim = model.n_fft // 2 trim = model.n_fft // 2
gen_size = model.chunk_size - 2 * trim gen_size = model.chunk_size - 2 * trim
pad = gen_size - n_sample % gen_size pad = gen_size - n_sample % gen_size
mix_p = np.concatenate((np.zeros((2,trim)), cmix, np.zeros((2,pad)), np.zeros((2,trim))), 1) mix_p = np.concatenate(
(np.zeros((2, trim)), cmix, np.zeros((2, pad)), np.zeros((2, trim))), 1
)
mix_waves = [] mix_waves = []
i = 0 i = 0
while i < n_sample + pad: while i < n_sample + pad:
@ -122,11 +161,21 @@ class Predictor:
_ort = self.model _ort = self.model
spek = model.stft(mix_waves) spek = model.stft(mix_waves)
if self.args.denoise: if self.args.denoise:
spec_pred = -_ort.run(None, {'input': -spek.cpu().numpy()})[0]*0.5+_ort.run(None, {'input': spek.cpu().numpy()})[0]*0.5 spec_pred = (
-_ort.run(None, {"input": -spek.cpu().numpy()})[0] * 0.5
+ _ort.run(None, {"input": spek.cpu().numpy()})[0] * 0.5
)
tar_waves = model.istft(torch.tensor(spec_pred)) tar_waves = model.istft(torch.tensor(spec_pred))
else: else:
tar_waves = model.istft(torch.tensor(_ort.run(None, {'input': spek.cpu().numpy()})[0])) tar_waves = model.istft(
tar_signal = tar_waves[:,:,trim:-trim].transpose(0,1).reshape(2, -1).numpy()[:, :-pad] torch.tensor(_ort.run(None, {"input": spek.cpu().numpy()})[0])
)
tar_signal = (
tar_waves[:, :, trim:-trim]
.transpose(0, 1)
.reshape(2, -1)
.numpy()[:, :-pad]
)
start = 0 if mix == 0 else margin_size start = 0 if mix == 0 else margin_size
end = None if mix == list(mixes.keys())[::-1][0] else -margin_size end = None if mix == list(mixes.keys())[::-1][0] else -margin_size
@ -141,6 +190,7 @@ class Predictor:
# del self.model # del self.model
progress_bar.close() progress_bar.close()
return _sources return _sources
def prediction(self, m, vocal_root, others_root, format): def prediction(self, m, vocal_root, others_root, format):
os.makedirs(vocal_root, exist_ok=True) os.makedirs(vocal_root, exist_ok=True)
os.makedirs(others_root, exist_ok=True) os.makedirs(others_root, exist_ok=True)
@ -151,10 +201,13 @@ class Predictor:
mix = mix.T mix = mix.T
sources = self.demix(mix.T) sources = self.demix(mix.T)
opt = sources[0].T opt = sources[0].T
sf.write("%s/%s_main_vocal.%s"%(vocal_root,basename,format), mix-opt, rate) sf.write(
"%s/%s_main_vocal.%s" % (vocal_root, basename, format), mix - opt, rate
)
sf.write("%s/%s_others.%s" % (others_root, basename, format), opt, rate) sf.write("%s/%s_others.%s" % (others_root, basename, format), opt, rate)
class MDXNetDereverb():
class MDXNetDereverb:
def __init__(self, chunks): def __init__(self, chunks):
self.onnx = "uvr5_weights/onnx_dereverb_By_FoxJoy" self.onnx = "uvr5_weights/onnx_dereverb_By_FoxJoy"
self.shifts = 10 #'Predict with randomised equivariant stabilisation' self.shifts = 10 #'Predict with randomised equivariant stabilisation'
@ -170,9 +223,11 @@ class MDXNetDereverb():
def _path_audio_(self, input, vocal_root, others_root, format): def _path_audio_(self, input, vocal_root, others_root, format):
self.pred.prediction(input, vocal_root, others_root, format) self.pred.prediction(input, vocal_root, others_root, format)
if __name__ == '__main__':
if __name__ == "__main__":
dereverb = MDXNetDereverb(15) dereverb = MDXNetDereverb(15)
from time import time as ttime from time import time as ttime
t0 = ttime() t0 = ttime()
dereverb._path_audio_( dereverb._path_audio_(
"雪雪伴奏对消HP5.wav", "雪雪伴奏对消HP5.wav",
@ -183,7 +238,7 @@ if __name__ == '__main__':
print(t1 - t0) print(t1 - t0)
''' """
runtime\python.exe MDXNet.py runtime\python.exe MDXNet.py
@ -195,4 +250,4 @@ runtime\python.exe MDXNet.py
half15:0.7G->6.6G,22.69s half15:0.7G->6.6G,22.69s
fp32-15:0.7G->6.6G,20.85s fp32-15:0.7G->6.6G,20.85s
''' """

48
infer-web.py
View File

@ -151,7 +151,7 @@ def vc_single(
filter_radius, filter_radius,
resample_sr, resample_sr,
rms_mix_rate, rms_mix_rate,
protect protect,
): # spk_item, input_audio0, vc_transform0,f0_file,f0method0 ): # spk_item, input_audio0, vc_transform0,f0_file,f0method0
global tgt_sr, net_g, vc, hubert_model, version global tgt_sr, net_g, vc, hubert_model, version
if input_audio_path is None: if input_audio_path is None:
@ -236,7 +236,7 @@ def vc_multi(
resample_sr, resample_sr,
rms_mix_rate, rms_mix_rate,
protect, protect,
format1 format1,
): ):
try: try:
dir_path = ( dir_path = (
@ -267,13 +267,15 @@ def vc_multi(
filter_radius, filter_radius,
resample_sr, resample_sr,
rms_mix_rate, rms_mix_rate,
protect protect,
) )
if "Success" in info: if "Success" in info:
try: try:
tgt_sr, audio_opt = opt tgt_sr, audio_opt = opt
sf.write( sf.write(
"%s/%s.%s" % (opt_root, os.path.basename(path),format1), audio_opt,tgt_sr "%s/%s.%s" % (opt_root, os.path.basename(path), format1),
audio_opt,
tgt_sr,
) )
except: except:
info += traceback.format_exc() info += traceback.format_exc()
@ -294,7 +296,7 @@ def uvr(model_name, inp_root, save_root_vocal, paths, save_root_ins, agg,format0
save_root_ins = ( save_root_ins = (
save_root_ins.strip(" ").strip('"').strip("\n").strip('"').strip(" ") save_root_ins.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
) )
if(model_name=="onnx_dereverb_By_FoxJoy"): if model_name == "onnx_dereverb_By_FoxJoy":
pre_fun = MDXNetDereverb(15) pre_fun = MDXNetDereverb(15)
else: else:
func = _audio_pre_ if "DeEcho" not in model_name else _audio_pre_new func = _audio_pre_ if "DeEcho" not in model_name else _audio_pre_new
@ -319,7 +321,9 @@ def uvr(model_name, inp_root, save_root_vocal, paths, save_root_ins, agg,format0
and info["streams"][0]["sample_rate"] == "44100" and info["streams"][0]["sample_rate"] == "44100"
): ):
need_reformat = 0 need_reformat = 0
pre_fun._path_audio_(inp_path, save_root_ins, save_root_vocal,format0) pre_fun._path_audio_(
inp_path, save_root_ins, save_root_vocal, format0
)
done = 1 done = 1
except: except:
need_reformat = 1 need_reformat = 1
@ -333,7 +337,9 @@ def uvr(model_name, inp_root, save_root_vocal, paths, save_root_ins, agg,format0
inp_path = tmp_path inp_path = tmp_path
try: try:
if done == 0: if done == 0:
pre_fun._path_audio_(inp_path, save_root_ins, save_root_vocal,format0) pre_fun._path_audio_(
inp_path, save_root_ins, save_root_vocal, format0
)
infos.append("%s->Success" % (os.path.basename(inp_path))) infos.append("%s->Success" % (os.path.basename(inp_path)))
yield "\n".join(infos) yield "\n".join(infos)
except: except:
@ -346,7 +352,7 @@ def uvr(model_name, inp_root, save_root_vocal, paths, save_root_ins, agg,format0
yield "\n".join(infos) yield "\n".join(infos)
finally: finally:
try: try:
if (model_name == "onnx_dereverb_By_FoxJoy"): if model_name == "onnx_dereverb_By_FoxJoy":
del pre_fun.pred.model del pre_fun.pred.model
del pre_fun.pred.model_ del pre_fun.pred.model_
else: else:
@ -1175,7 +1181,9 @@ with gr.Blocks() as app:
value="E:\\codes\\py39\\test-20230416b\\todo-songs\\冬之花clip1.wav", value="E:\\codes\\py39\\test-20230416b\\todo-songs\\冬之花clip1.wav",
) )
f0method0 = gr.Radio( f0method0 = gr.Radio(
label=i18n("选择音高提取算法,输入歌声可用pm提速,harvest低音好但巨慢无比,crepe效果好但吃GPU"), label=i18n(
"选择音高提取算法,输入歌声可用pm提速,harvest低音好但巨慢无比,crepe效果好但吃GPU"
),
choices=["pm", "harvest", "crepe"], choices=["pm", "harvest", "crepe"],
value="pm", value="pm",
interactive=True, interactive=True,
@ -1233,7 +1241,9 @@ with gr.Blocks() as app:
protect0 = gr.Slider( protect0 = gr.Slider(
minimum=0, minimum=0,
maximum=0.5, maximum=0.5,
label=i18n("保护清辅音和呼吸声，防止电音撕裂等artifact，拉满0.5不开启，调低加大保护力度但可能降低索引效果"), label=i18n(
"保护清辅音和呼吸声，防止电音撕裂等artifact，拉满0.5不开启，调低加大保护力度但可能降低索引效果"
),
value=0.33, value=0.33,
step=0.01, step=0.01,
interactive=True, interactive=True,
@ -1258,7 +1268,7 @@ with gr.Blocks() as app:
filter_radius0, filter_radius0,
resample_sr0, resample_sr0,
rms_mix_rate0, rms_mix_rate0,
protect0 protect0,
], ],
[vc_output1, vc_output2], [vc_output1, vc_output2],
) )
@ -1273,7 +1283,9 @@ with gr.Blocks() as app:
) )
opt_input = gr.Textbox(label=i18n("指定输出文件夹"), value="opt") opt_input = gr.Textbox(label=i18n("指定输出文件夹"), value="opt")
f0method1 = gr.Radio( f0method1 = gr.Radio(
label=i18n("选择音高提取算法,输入歌声可用pm提速,harvest低音好但巨慢无比,crepe效果好但吃GPU"), label=i18n(
"选择音高提取算法,输入歌声可用pm提速,harvest低音好但巨慢无比,crepe效果好但吃GPU"
),
choices=["pm", "harvest", "crepe"], choices=["pm", "harvest", "crepe"],
value="pm", value="pm",
interactive=True, interactive=True,
@ -1328,7 +1340,9 @@ with gr.Blocks() as app:
protect1 = gr.Slider( protect1 = gr.Slider(
minimum=0, minimum=0,
maximum=0.5, maximum=0.5,
label=i18n("保护清辅音和呼吸声，防止电音撕裂等artifact，拉满0.5不开启，调低加大保护力度但可能降低索引效果"), label=i18n(
"保护清辅音和呼吸声，防止电音撕裂等artifact，拉满0.5不开启，调低加大保护力度但可能降低索引效果"
),
value=0.33, value=0.33,
step=0.01, step=0.01,
interactive=True, interactive=True,
@ -1367,7 +1381,7 @@ with gr.Blocks() as app:
resample_sr1, resample_sr1,
rms_mix_rate1, rms_mix_rate1,
protect1, protect1,
format1 format1,
], ],
[vc_output3], [vc_output3],
) )
@ -1412,7 +1426,9 @@ with gr.Blocks() as app:
opt_vocal_root = gr.Textbox( opt_vocal_root = gr.Textbox(
label=i18n("指定输出主人声文件夹"), value="opt" label=i18n("指定输出主人声文件夹"), value="opt"
) )
opt_ins_root = gr.Textbox(label=i18n("指定输出非主人声文件夹"), value="opt") opt_ins_root = gr.Textbox(
label=i18n("指定输出非主人声文件夹"), value="opt"
)
format0 = gr.Radio( format0 = gr.Radio(
label=i18n("导出文件格式"), label=i18n("导出文件格式"),
choices=["wav", "flac", "mp3", "m4a"], choices=["wav", "flac", "mp3", "m4a"],
@ -1430,7 +1446,7 @@ with gr.Blocks() as app:
wav_inputs, wav_inputs,
opt_ins_root, opt_ins_root,
agg, agg,
format0 format0,
], ],
[vc_output4], [vc_output4],
) )

26
infer_uvr5.py
View File

@ -1,7 +1,9 @@
import os, sys, torch, warnings, pdb import os, sys, torch, warnings, pdb
now_dir = os.getcwd() now_dir = os.getcwd()
sys.path.append(now_dir) sys.path.append(now_dir)
from json import load as ll from json import load as ll
warnings.filterwarnings("ignore") warnings.filterwarnings("ignore")
import librosa import librosa
import importlib import importlib
@ -15,6 +17,7 @@ import soundfile as sf
from uvr5_pack.lib_v5.nets_new import CascadedNet from uvr5_pack.lib_v5.nets_new import CascadedNet
from uvr5_pack.lib_v5 import nets_61968KB as nets from uvr5_pack.lib_v5 import nets_61968KB as nets
class _audio_pre_: class _audio_pre_:
def __init__(self, agg, model_path, device, is_half): def __init__(self, agg, model_path, device, is_half):
self.model_path = model_path self.model_path = model_path
@ -122,9 +125,11 @@ class _audio_pre_:
print("%s instruments done" % name) print("%s instruments done" % name)
sf.write( sf.write(
os.path.join( os.path.join(
ins_root, "instrument_{}_{}.{}".format(name, self.data["agg"],format) ins_root,
"instrument_{}_{}.{}".format(name, self.data["agg"], format),
), ),
(np.array(wav_instrument) * 32768).astype("int16"), self.mp.param["sr"], (np.array(wav_instrument) * 32768).astype("int16"),
self.mp.param["sr"],
) # ) #
if vocal_root is not None: if vocal_root is not None:
if self.data["high_end_process"].startswith("mirroring"): if self.data["high_end_process"].startswith("mirroring"):
@ -141,9 +146,11 @@ class _audio_pre_:
os.path.join( os.path.join(
vocal_root, "vocal_{}_{}.{}".format(name, self.data["agg"], format) vocal_root, "vocal_{}_{}.{}".format(name, self.data["agg"], format)
), ),
(np.array(wav_vocals) * 32768).astype("int16"), self.mp.param["sr"], (np.array(wav_vocals) * 32768).astype("int16"),
self.mp.param["sr"],
) )
class _audio_pre_new: class _audio_pre_new:
def __init__(self, agg, model_path, device, is_half): def __init__(self, agg, model_path, device, is_half):
self.model_path = model_path self.model_path = model_path
@ -171,7 +178,9 @@ class _audio_pre_new:
self.mp = mp self.mp = mp
self.model = model self.model = model
def _path_audio_(self, music_file, vocal_root=None, ins_root=None,format="flac"):#3个VR模型vocal和ins是反的 def _path_audio_(
self, music_file, vocal_root=None, ins_root=None, format="flac"
): # 3个VR模型vocal和ins是反的
if ins_root is None and vocal_root is None: if ins_root is None and vocal_root is None:
return "No save root." return "No save root."
name = os.path.basename(music_file) name = os.path.basename(music_file)
@ -252,9 +261,11 @@ class _audio_pre_new:
print("%s instruments done" % name) print("%s instruments done" % name)
sf.write( sf.write(
os.path.join( os.path.join(
ins_root, "main_vocal_{}_{}.{}".format(name, self.data["agg"],format) ins_root,
"main_vocal_{}_{}.{}".format(name, self.data["agg"], format),
), ),
(np.array(wav_instrument) * 32768).astype("int16"),self.mp.param["sr"], (np.array(wav_instrument) * 32768).astype("int16"),
self.mp.param["sr"],
) # ) #
if vocal_root is not None: if vocal_root is not None:
if self.data["high_end_process"].startswith("mirroring"): if self.data["high_end_process"].startswith("mirroring"):
@ -271,7 +282,8 @@ class _audio_pre_new:
os.path.join( os.path.join(
vocal_root, "others_{}_{}.{}".format(name, self.data["agg"], format) vocal_root, "others_{}_{}.{}".format(name, self.data["agg"], format)
), ),
(np.array(wav_vocals) * 32768).astype("int16"),self.mp.param["sr"], (np.array(wav_vocals) * 32768).astype("int16"),
self.mp.param["sr"],
) )

35
uvr5_pack/lib_v5/layers_new.py
View File

@ -4,27 +4,29 @@ import torch.nn.functional as F
from uvr5_pack.lib_v5 import spec_utils from uvr5_pack.lib_v5 import spec_utils
class Conv2DBNActiv(nn.Module):
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU): def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__() super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential( self.conv = nn.Sequential(
nn.Conv2d( nn.Conv2d(
nin, nout, nin,
nout,
kernel_size=ksize, kernel_size=ksize,
stride=stride, stride=stride,
padding=pad, padding=pad,
dilation=dilation, dilation=dilation,
bias=False), bias=False,
),
nn.BatchNorm2d(nout), nn.BatchNorm2d(nout),
activ() activ(),
) )
def __call__(self, x): def __call__(self, x):
return self.conv(x) return self.conv(x)
class Encoder(nn.Module):
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU): def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__() super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, stride, pad, activ=activ) self.conv1 = Conv2DBNActiv(nin, nout, ksize, stride, pad, activ=activ)
@ -38,15 +40,16 @@ class Encoder(nn.Module):
class Decoder(nn.Module): class Decoder(nn.Module):
def __init__(
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False): self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__() super(Decoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ) self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
# self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ) # self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None): def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode='bilinear', align_corners=True) x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None: if skip is not None:
skip = spec_utils.crop_center(skip, x) skip = spec_utils.crop_center(skip, x)
@ -62,12 +65,11 @@ class Decoder(nn.Module):
class ASPPModule(nn.Module): class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 12), activ=nn.ReLU, dropout=False): def __init__(self, nin, nout, dilations=(4, 8, 12), activ=nn.ReLU, dropout=False):
super(ASPPModule, self).__init__() super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential( self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)), nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ) Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ),
) )
self.conv2 = Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ) self.conv2 = Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ)
self.conv3 = Conv2DBNActiv( self.conv3 = Conv2DBNActiv(
@ -84,7 +86,9 @@ class ASPPModule(nn.Module):
def forward(self, x): def forward(self, x):
_, _, h, w = x.size() _, _, h, w = x.size()
feat1 = F.interpolate(self.conv1(x), size=(h, w), mode='bilinear', align_corners=True) feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x) feat2 = self.conv2(x)
feat3 = self.conv3(x) feat3 = self.conv3(x)
feat4 = self.conv4(x) feat4 = self.conv4(x)
@ -99,19 +103,14 @@ class ASPPModule(nn.Module):
class LSTMModule(nn.Module): class LSTMModule(nn.Module):
def __init__(self, nin_conv, nin_lstm, nout_lstm): def __init__(self, nin_conv, nin_lstm, nout_lstm):
super(LSTMModule, self).__init__() super(LSTMModule, self).__init__()
self.conv = Conv2DBNActiv(nin_conv, 1, 1, 1, 0) self.conv = Conv2DBNActiv(nin_conv, 1, 1, 1, 0)
self.lstm = nn.LSTM( self.lstm = nn.LSTM(
input_size=nin_lstm, input_size=nin_lstm, hidden_size=nout_lstm // 2, bidirectional=True
hidden_size=nout_lstm // 2,
bidirectional=True
) )
self.dense = nn.Sequential( self.dense = nn.Sequential(
nn.Linear(nout_lstm, nin_lstm), nn.Linear(nout_lstm, nin_lstm), nn.BatchNorm1d(nin_lstm), nn.ReLU()
nn.BatchNorm1d(nin_lstm),
nn.ReLU()
) )
def forward(self, x): def forward(self, x):

28
uvr5_pack/lib_v5/nets_new.py
View File

@ -3,9 +3,11 @@ from torch import nn
import torch.nn.functional as F import torch.nn.functional as F
from uvr5_pack.lib_v5 import layers_new as layers from uvr5_pack.lib_v5 import layers_new as layers
class BaseNet(nn.Module):
def __init__(self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))): class BaseNet(nn.Module):
def __init__(
self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))
):
super(BaseNet, self).__init__() super(BaseNet, self).__init__()
self.enc1 = layers.Conv2DBNActiv(nin, nout, 3, 1, 1) self.enc1 = layers.Conv2DBNActiv(nin, nout, 3, 1, 1)
self.enc2 = layers.Encoder(nout, nout * 2, 3, 2, 1) self.enc2 = layers.Encoder(nout, nout * 2, 3, 2, 1)
@ -38,8 +40,8 @@ class BaseNet(nn.Module):
return h return h
class CascadedNet(nn.Module):
class CascadedNet(nn.Module):
def __init__(self, n_fft, nout=32, nout_lstm=128): def __init__(self, n_fft, nout=32, nout_lstm=128):
super(CascadedNet, self).__init__() super(CascadedNet, self).__init__()
@ -50,18 +52,24 @@ class CascadedNet(nn.Module):
self.stg1_low_band_net = nn.Sequential( self.stg1_low_band_net = nn.Sequential(
BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm), BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm),
layers.Conv2DBNActiv(nout // 2, nout // 4, 1, 1, 0) layers.Conv2DBNActiv(nout // 2, nout // 4, 1, 1, 0),
) )
self.stg1_high_band_net = BaseNet(2, nout // 4, self.nin_lstm // 2, nout_lstm // 2) self.stg1_high_band_net = BaseNet(
2, nout // 4, self.nin_lstm // 2, nout_lstm // 2
)
self.stg2_low_band_net = nn.Sequential( self.stg2_low_band_net = nn.Sequential(
BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm), BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm),
layers.Conv2DBNActiv(nout, nout // 2, 1, 1, 0) layers.Conv2DBNActiv(nout, nout // 2, 1, 1, 0),
)
self.stg2_high_band_net = BaseNet(
nout // 4 + 2, nout // 2, self.nin_lstm // 2, nout_lstm // 2
) )
self.stg2_high_band_net = BaseNet(nout // 4 + 2, nout // 2, self.nin_lstm // 2, nout_lstm // 2)
self.stg3_full_band_net = BaseNet(3 * nout // 4 + 2, nout, self.nin_lstm, nout_lstm) self.stg3_full_band_net = BaseNet(
3 * nout // 4 + 2, nout, self.nin_lstm, nout_lstm
)
self.out = nn.Conv2d(nout, 2, 1, bias=False) self.out = nn.Conv2d(nout, 2, 1, bias=False)
self.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False) self.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False)
@ -89,7 +97,7 @@ class CascadedNet(nn.Module):
mask = F.pad( mask = F.pad(
input=mask, input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]), pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode='replicate' mode="replicate",
) )
if self.training: if self.training:
@ -98,7 +106,7 @@ class CascadedNet(nn.Module):
aux = F.pad( aux = F.pad(
input=aux, input=aux,
pad=(0, 0, 0, self.output_bin - aux.size()[2]), pad=(0, 0, 0, self.output_bin - aux.size()[2]),
mode='replicate' mode="replicate",
) )
return mask, aux return mask, aux
else: else:

20
vc_infer_pipeline.py
View File

@ -162,7 +162,7 @@ class VC(object):
big_npy, big_npy,
index_rate, index_rate,
version, version,
protect protect,
): # ,file_index,file_big_npy ): # ,file_index,file_big_npy
feats = torch.from_numpy(audio0) feats = torch.from_numpy(audio0)
if self.is_half: if self.is_half:
@ -184,7 +184,7 @@ class VC(object):
with torch.no_grad(): with torch.no_grad():
logits = model.extract_features(**inputs) logits = model.extract_features(**inputs)
feats = model.final_proj(logits[0]) if version == "v1" else logits[0] feats = model.final_proj(logits[0]) if version == "v1" else logits[0]
if(protect<0.5): if protect < 0.5:
feats0 = feats.clone() feats0 = feats.clone()
if ( if (
isinstance(index, type(None)) == False isinstance(index, type(None)) == False
@ -211,8 +211,10 @@ class VC(object):
) )
feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1) feats = F.interpolate(feats.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1)
if(protect<0.5): if protect < 0.5:
feats0 = F.interpolate(feats0.permute(0, 2, 1), scale_factor=2).permute(0, 2, 1) feats0 = F.interpolate(feats0.permute(0, 2, 1), scale_factor=2).permute(
0, 2, 1
)
t1 = ttime() t1 = ttime()
p_len = audio0.shape[0] // self.window p_len = audio0.shape[0] // self.window
if feats.shape[1] < p_len: if feats.shape[1] < p_len:
@ -221,7 +223,7 @@ class VC(object):
pitch = pitch[:, :p_len] pitch = pitch[:, :p_len]
pitchf = pitchf[:, :p_len] pitchf = pitchf[:, :p_len]
if(protect<0.5): if protect < 0.5:
pitchff = pitchf.clone() pitchff = pitchf.clone()
pitchff[pitchf > 0] = 1 pitchff[pitchf > 0] = 1
pitchff[pitchf < 1] = protect pitchff[pitchf < 1] = protect
@ -356,7 +358,7 @@ class VC(object):
big_npy, big_npy,
index_rate, index_rate,
version, version,
protect protect,
)[self.t_pad_tgt : -self.t_pad_tgt] )[self.t_pad_tgt : -self.t_pad_tgt]
) )
else: else:
@ -373,7 +375,7 @@ class VC(object):
big_npy, big_npy,
index_rate, index_rate,
version, version,
protect protect,
)[self.t_pad_tgt : -self.t_pad_tgt] )[self.t_pad_tgt : -self.t_pad_tgt]
) )
s = t s = t
@ -391,7 +393,7 @@ class VC(object):
big_npy, big_npy,
index_rate, index_rate,
version, version,
protect protect,
)[self.t_pad_tgt : -self.t_pad_tgt] )[self.t_pad_tgt : -self.t_pad_tgt]
) )
else: else:
@ -408,7 +410,7 @@ class VC(object):
big_npy, big_npy,
index_rate, index_rate,
version, version,
protect protect,
)[self.t_pad_tgt : -self.t_pad_tgt] )[self.t_pad_tgt : -self.t_pad_tgt]
) )
audio_opt = np.concatenate(audio_opt) audio_opt = np.concatenate(audio_opt)