blackteay/GPT-SoVITS
1
0
Fork 0
mirror of https://github.com/RVC-Boss/GPT-SoVITS.git synced 2025-04-06 03:57:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add files via upload

Browse Source
This commit is contained in:
RVC-Boss 2024-01-16 17:33:36 +08:00 committed by GitHub
parent 7425d0f780
commit 6b7a98038f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
46 changed files with 4848 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

183
tools/uvr5/lib/lib_v5/dataset.py Normal file
View File

@ -0,0 +1,183 @@
import os
import random
import numpy as np
import torch
import torch.utils.data
from tqdm import tqdm
from . import spec_utils
class VocalRemoverValidationSet(torch.utils.data.Dataset):
def __init__(self, patch_list):
self.patch_list = patch_list
def __len__(self):
return len(self.patch_list)
def __getitem__(self, idx):
path = self.patch_list[idx]
data = np.load(path)
X, y = data["X"], data["y"]
X_mag = np.abs(X)
y_mag = np.abs(y)
return X_mag, y_mag
def make_pair(mix_dir, inst_dir):
input_exts = [".wav", ".m4a", ".mp3", ".mp4", ".flac"]
X_list = sorted(
[
os.path.join(mix_dir, fname)
for fname in os.listdir(mix_dir)
if os.path.splitext(fname)[1] in input_exts
]
)
y_list = sorted(
[
os.path.join(inst_dir, fname)
for fname in os.listdir(inst_dir)
if os.path.splitext(fname)[1] in input_exts
]
)
filelist = list(zip(X_list, y_list))
return filelist
def train_val_split(dataset_dir, split_mode, val_rate, val_filelist):
if split_mode == "random":
filelist = make_pair(
os.path.join(dataset_dir, "mixtures"),
os.path.join(dataset_dir, "instruments"),
)
random.shuffle(filelist)
if len(val_filelist) == 0:
val_size = int(len(filelist) * val_rate)
train_filelist = filelist[:-val_size]
val_filelist = filelist[-val_size:]
else:
train_filelist = [
pair for pair in filelist if list(pair) not in val_filelist
]
elif split_mode == "subdirs":
if len(val_filelist) != 0:
raise ValueError(
"The `val_filelist` option is not available in `subdirs` mode"
)
train_filelist = make_pair(
os.path.join(dataset_dir, "training/mixtures"),
os.path.join(dataset_dir, "training/instruments"),
)
val_filelist = make_pair(
os.path.join(dataset_dir, "validation/mixtures"),
os.path.join(dataset_dir, "validation/instruments"),
)
return train_filelist, val_filelist
def augment(X, y, reduction_rate, reduction_mask, mixup_rate, mixup_alpha):
perm = np.random.permutation(len(X))
for i, idx in enumerate(tqdm(perm)):
if np.random.uniform() < reduction_rate:
y[idx] = spec_utils.reduce_vocal_aggressively(
X[idx], y[idx], reduction_mask
)
if np.random.uniform() < 0.5:
# swap channel
X[idx] = X[idx, ::-1]
y[idx] = y[idx, ::-1]
if np.random.uniform() < 0.02:
# mono
X[idx] = X[idx].mean(axis=0, keepdims=True)
y[idx] = y[idx].mean(axis=0, keepdims=True)
if np.random.uniform() < 0.02:
# inst
X[idx] = y[idx]
if np.random.uniform() < mixup_rate and i < len(perm) - 1:
lam = np.random.beta(mixup_alpha, mixup_alpha)
X[idx] = lam * X[idx] + (1 - lam) * X[perm[i + 1]]
y[idx] = lam * y[idx] + (1 - lam) * y[perm[i + 1]]
return X, y
def make_padding(width, cropsize, offset):
left = offset
roi_size = cropsize - left * 2
if roi_size == 0:
roi_size = cropsize
right = roi_size - (width % roi_size) + left
return left, right, roi_size
def make_training_set(filelist, cropsize, patches, sr, hop_length, n_fft, offset):
len_dataset = patches * len(filelist)
X_dataset = np.zeros((len_dataset, 2, n_fft // 2 + 1, cropsize), dtype=np.complex64)
y_dataset = np.zeros((len_dataset, 2, n_fft // 2 + 1, cropsize), dtype=np.complex64)
for i, (X_path, y_path) in enumerate(tqdm(filelist)):
X, y = spec_utils.cache_or_load(X_path, y_path, sr, hop_length, n_fft)
coef = np.max([np.abs(X).max(), np.abs(y).max()])
X, y = X / coef, y / coef
l, r, roi_size = make_padding(X.shape[2], cropsize, offset)
X_pad = np.pad(X, ((0, 0), (0, 0), (l, r)), mode="constant")
y_pad = np.pad(y, ((0, 0), (0, 0), (l, r)), mode="constant")
starts = np.random.randint(0, X_pad.shape[2] - cropsize, patches)
ends = starts + cropsize
for j in range(patches):
idx = i * patches + j
X_dataset[idx] = X_pad[:, :, starts[j] : ends[j]]
y_dataset[idx] = y_pad[:, :, starts[j] : ends[j]]
return X_dataset, y_dataset
def make_validation_set(filelist, cropsize, sr, hop_length, n_fft, offset):
patch_list = []
patch_dir = "cs{}_sr{}_hl{}_nf{}_of{}".format(
cropsize, sr, hop_length, n_fft, offset
)
os.makedirs(patch_dir, exist_ok=True)
for i, (X_path, y_path) in enumerate(tqdm(filelist)):
basename = os.path.splitext(os.path.basename(X_path))[0]
X, y = spec_utils.cache_or_load(X_path, y_path, sr, hop_length, n_fft)
coef = np.max([np.abs(X).max(), np.abs(y).max()])
X, y = X / coef, y / coef
l, r, roi_size = make_padding(X.shape[2], cropsize, offset)
X_pad = np.pad(X, ((0, 0), (0, 0), (l, r)), mode="constant")
y_pad = np.pad(y, ((0, 0), (0, 0), (l, r)), mode="constant")
len_dataset = int(np.ceil(X.shape[2] / roi_size))
for j in range(len_dataset):
outpath = os.path.join(patch_dir, "{}_p{}.npz".format(basename, j))
start = j * roi_size
if not os.path.exists(outpath):
np.savez(
outpath,
X=X_pad[:, :, start : start + cropsize],
y=y_pad[:, :, start : start + cropsize],
)
patch_list.append(outpath)
return VocalRemoverValidationSet(patch_list)

118
tools/uvr5/lib/lib_v5/layers.py Normal file
View File

@ -0,0 +1,118 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class SeperableConv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(SeperableConv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nin,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
groups=nin,
bias=False,
),
nn.Conv2d(nin, nout, kernel_size=1, bias=False),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
def __call__(self, x):
skip = self.conv1(x)
h = self.conv2(skip)
return h, skip
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv(x)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
self.conv3 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = nn.Sequential(
Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
)
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
bottle = self.bottleneck(out)
return bottle

118
tools/uvr5/lib/lib_v5/layers_123812KB.py Normal file
View File

@ -0,0 +1,118 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class SeperableConv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(SeperableConv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nin,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
groups=nin,
bias=False,
),
nn.Conv2d(nin, nout, kernel_size=1, bias=False),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
def __call__(self, x):
skip = self.conv1(x)
h = self.conv2(skip)
return h, skip
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv(x)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
self.conv3 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = nn.Sequential(
Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
)
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
bottle = self.bottleneck(out)
return bottle

118
tools/uvr5/lib/lib_v5/layers_123821KB.py Normal file
View File

@ -0,0 +1,118 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class SeperableConv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(SeperableConv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nin,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
groups=nin,
bias=False,
),
nn.Conv2d(nin, nout, kernel_size=1, bias=False),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
def __call__(self, x):
skip = self.conv1(x)
h = self.conv2(skip)
return h, skip
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv(x)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 16), activ=nn.ReLU):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
self.conv3 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = nn.Sequential(
Conv2DBNActiv(nin * 5, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
)
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
bottle = self.bottleneck(out)
return bottle

126
tools/uvr5/lib/lib_v5/layers_33966KB.py Normal file
View File

@ -0,0 +1,126 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class SeperableConv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(SeperableConv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nin,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
groups=nin,
bias=False,
),
nn.Conv2d(nin, nout, kernel_size=1, bias=False),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
def __call__(self, x):
skip = self.conv1(x)
h = self.conv2(skip)
return h, skip
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv(x)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
self.conv3 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.conv6 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.conv7 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = nn.Sequential(
Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
)
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
feat6 = self.conv6(x)
feat7 = self.conv7(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
bottle = self.bottleneck(out)
return bottle

126
tools/uvr5/lib/lib_v5/layers_537227KB.py Normal file
View File

@ -0,0 +1,126 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class SeperableConv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(SeperableConv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nin,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
groups=nin,
bias=False,
),
nn.Conv2d(nin, nout, kernel_size=1, bias=False),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
def __call__(self, x):
skip = self.conv1(x)
h = self.conv2(skip)
return h, skip
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv(x)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
self.conv3 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.conv6 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.conv7 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = nn.Sequential(
Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
)
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
feat6 = self.conv6(x)
feat7 = self.conv7(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
bottle = self.bottleneck(out)
return bottle

126
tools/uvr5/lib/lib_v5/layers_537238KB.py Normal file
View File

@ -0,0 +1,126 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class SeperableConv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(SeperableConv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nin,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
groups=nin,
bias=False,
),
nn.Conv2d(nin, nout, kernel_size=1, bias=False),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, stride, pad, activ=activ)
def __call__(self, x):
skip = self.conv1(x)
h = self.conv2(skip)
return h, skip
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv = Conv2DBNActiv(nin, nout, ksize, 1, pad, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv(x)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 16, 32, 64), activ=nn.ReLU):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nin, 1, 1, 0, activ=activ)
self.conv3 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.conv6 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.conv7 = SeperableConv2DBNActiv(
nin, nin, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = nn.Sequential(
Conv2DBNActiv(nin * 7, nout, 1, 1, 0, activ=activ), nn.Dropout2d(0.1)
)
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
feat6 = self.conv6(x)
feat7 = self.conv7(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5, feat6, feat7), dim=1)
bottle = self.bottleneck(out)
return bottle

125
tools/uvr5/lib/lib_v5/layers_new.py Normal file
View File

@ -0,0 +1,125 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class Conv2DBNActiv(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, dilation=1, activ=nn.ReLU):
super(Conv2DBNActiv, self).__init__()
self.conv = nn.Sequential(
nn.Conv2d(
nin,
nout,
kernel_size=ksize,
stride=stride,
padding=pad,
dilation=dilation,
bias=False,
),
nn.BatchNorm2d(nout),
activ(),
)
def __call__(self, x):
return self.conv(x)
class Encoder(nn.Module):
def __init__(self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.LeakyReLU):
super(Encoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, nout, ksize, stride, pad, activ=activ)
self.conv2 = Conv2DBNActiv(nout, nout, ksize, 1, pad, activ=activ)
def __call__(self, x):
h = self.conv1(x)
h = self.conv2(h)
return h
class Decoder(nn.Module):
def __init__(
self, nin, nout, ksize=3, stride=1, pad=1, activ=nn.ReLU, dropout=False
):
super(Decoder, self).__init__()
self.conv1 = Conv2DBNActiv(nin, 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
def __call__(self, x, skip=None):
x = F.interpolate(x, scale_factor=2, mode="bilinear", align_corners=True)
if skip is not None:
skip = spec_utils.crop_center(skip, x)
x = torch.cat([x, skip], dim=1)
h = self.conv1(x)
# h = self.conv2(h)
if self.dropout is not None:
h = self.dropout(h)
return h
class ASPPModule(nn.Module):
def __init__(self, nin, nout, dilations=(4, 8, 12), activ=nn.ReLU, dropout=False):
super(ASPPModule, self).__init__()
self.conv1 = nn.Sequential(
nn.AdaptiveAvgPool2d((1, None)),
Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ),
)
self.conv2 = Conv2DBNActiv(nin, nout, 1, 1, 0, activ=activ)
self.conv3 = Conv2DBNActiv(
nin, nout, 3, 1, dilations[0], dilations[0], activ=activ
)
self.conv4 = Conv2DBNActiv(
nin, nout, 3, 1, dilations[1], dilations[1], activ=activ
)
self.conv5 = Conv2DBNActiv(
nin, nout, 3, 1, dilations[2], dilations[2], activ=activ
)
self.bottleneck = Conv2DBNActiv(nout * 5, nout, 1, 1, 0, activ=activ)
self.dropout = nn.Dropout2d(0.1) if dropout else None
def forward(self, x):
_, _, h, w = x.size()
feat1 = F.interpolate(
self.conv1(x), size=(h, w), mode="bilinear", align_corners=True
)
feat2 = self.conv2(x)
feat3 = self.conv3(x)
feat4 = self.conv4(x)
feat5 = self.conv5(x)
out = torch.cat((feat1, feat2, feat3, feat4, feat5), dim=1)
out = self.bottleneck(out)
if self.dropout is not None:
out = self.dropout(out)
return out
class LSTMModule(nn.Module):
def __init__(self, nin_conv, nin_lstm, nout_lstm):
super(LSTMModule, self).__init__()
self.conv = Conv2DBNActiv(nin_conv, 1, 1, 1, 0)
self.lstm = nn.LSTM(
input_size=nin_lstm, hidden_size=nout_lstm // 2, bidirectional=True
)
self.dense = nn.Sequential(
nn.Linear(nout_lstm, nin_lstm), nn.BatchNorm1d(nin_lstm), nn.ReLU()
)
def forward(self, x):
N, _, nbins, nframes = x.size()
h = self.conv(x)[:, 0] # N, nbins, nframes
h = h.permute(2, 0, 1) # nframes, N, nbins
h, _ = self.lstm(h)
h = self.dense(h.reshape(-1, h.size()[-1])) # nframes * N, nbins
h = h.reshape(nframes, N, 1, nbins)
h = h.permute(1, 2, 3, 0)
return h

69
tools/uvr5/lib/lib_v5/model_param_init.py Normal file
View File

@ -0,0 +1,69 @@
import json
import os
import pathlib
default_param = {}
default_param["bins"] = 768
default_param["unstable_bins"] = 9 # training only
default_param["reduction_bins"] = 762 # training only
default_param["sr"] = 44100
default_param["pre_filter_start"] = 757
default_param["pre_filter_stop"] = 768
default_param["band"] = {}
default_param["band"][1] = {
"sr": 11025,
"hl": 128,
"n_fft": 960,
"crop_start": 0,
"crop_stop": 245,
"lpf_start": 61, # inference only
"res_type": "polyphase",
}
default_param["band"][2] = {
"sr": 44100,
"hl": 512,
"n_fft": 1536,
"crop_start": 24,
"crop_stop": 547,
"hpf_start": 81, # inference only
"res_type": "sinc_best",
}
def int_keys(d):
r = {}
for k, v in d:
if k.isdigit():
k = int(k)
r[k] = v
return r
class ModelParameters(object):
def __init__(self, config_path=""):
if ".pth" == pathlib.Path(config_path).suffix:
import zipfile
with zipfile.ZipFile(config_path, "r") as zip:
self.param = json.loads(
zip.read("param.json"), object_pairs_hook=int_keys
)
elif ".json" == pathlib.Path(config_path).suffix:
with open(config_path, "r") as f:
self.param = json.loads(f.read(), object_pairs_hook=int_keys)
else:
self.param = default_param
for k in [
"mid_side",
"mid_side_b",
"mid_side_b2",
"stereo_w",
"stereo_n",
"reverse",
]:
if not k in self.param:
self.param[k] = False

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr16000_hl512.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 1024,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 16000,
"hl": 512,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 1024,
"hpf_start": -1,
"res_type": "sinc_best"
}
},
"sr": 16000,
"pre_filter_start": 1023,
"pre_filter_stop": 1024
}

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr32000_hl512.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 1024,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 32000,
"hl": 512,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 1024,
"hpf_start": -1,
"res_type": "kaiser_fast"
}
},
"sr": 32000,
"pre_filter_start": 1000,
"pre_filter_stop": 1021
}

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr33075_hl384.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 1024,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 33075,
"hl": 384,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 1024,
"hpf_start": -1,
"res_type": "sinc_best"
}
},
"sr": 33075,
"pre_filter_start": 1000,
"pre_filter_stop": 1021
}

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr44100_hl1024.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 1024,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 44100,
"hl": 1024,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 1024,
"hpf_start": -1,
"res_type": "sinc_best"
}
},
"sr": 44100,
"pre_filter_start": 1023,
"pre_filter_stop": 1024
}

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr44100_hl256.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 256,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 44100,
"hl": 256,
"n_fft": 512,
"crop_start": 0,
"crop_stop": 256,
"hpf_start": -1,
"res_type": "sinc_best"
}
},
"sr": 44100,
"pre_filter_start": 256,
"pre_filter_stop": 256
}

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr44100_hl512.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 1024,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 44100,
"hl": 512,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 1024,
"hpf_start": -1,
"res_type": "sinc_best"
}
},
"sr": 44100,
"pre_filter_start": 1023,
"pre_filter_stop": 1024
}

19
tools/uvr5/lib/lib_v5/modelparams/1band_sr44100_hl512_cut.json Normal file
View File

@ -0,0 +1,19 @@
{
"bins": 1024,
"unstable_bins": 0,
"reduction_bins": 0,
"band": {
"1": {
"sr": 44100,
"hl": 512,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 700,
"hpf_start": -1,
"res_type": "sinc_best"
}
},
"sr": 44100,
"pre_filter_start": 1023,
"pre_filter_stop": 700
}

30
tools/uvr5/lib/lib_v5/modelparams/2band_32000.json Normal file
View File

@ -0,0 +1,30 @@
{
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 705,
"band": {
"1": {
"sr": 6000,
"hl": 66,
"n_fft": 512,
"crop_start": 0,
"crop_stop": 240,
"lpf_start": 60,
"lpf_stop": 118,
"res_type": "sinc_fastest"
},
"2": {
"sr": 32000,
"hl": 352,
"n_fft": 1024,
"crop_start": 22,
"crop_stop": 505,
"hpf_start": 44,
"hpf_stop": 23,
"res_type": "sinc_medium"
}
},
"sr": 32000,
"pre_filter_start": 710,
"pre_filter_stop": 731
}

30
tools/uvr5/lib/lib_v5/modelparams/2band_44100_lofi.json Normal file
View File

@ -0,0 +1,30 @@
{
"bins": 512,
"unstable_bins": 7,
"reduction_bins": 510,
"band": {
"1": {
"sr": 11025,
"hl": 160,
"n_fft": 768,
"crop_start": 0,
"crop_stop": 192,
"lpf_start": 41,
"lpf_stop": 139,
"res_type": "sinc_fastest"
},
"2": {
"sr": 44100,
"hl": 640,
"n_fft": 1024,
"crop_start": 10,
"crop_stop": 320,
"hpf_start": 47,
"hpf_stop": 15,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 510,
"pre_filter_stop": 512
}

30
tools/uvr5/lib/lib_v5/modelparams/2band_48000.json Normal file
View File

@ -0,0 +1,30 @@
{
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 705,
"band": {
"1": {
"sr": 6000,
"hl": 66,
"n_fft": 512,
"crop_start": 0,
"crop_stop": 240,
"lpf_start": 60,
"lpf_stop": 240,
"res_type": "sinc_fastest"
},
"2": {
"sr": 48000,
"hl": 528,
"n_fft": 1536,
"crop_start": 22,
"crop_stop": 505,
"hpf_start": 82,
"hpf_stop": 22,
"res_type": "sinc_medium"
}
},
"sr": 48000,
"pre_filter_start": 710,
"pre_filter_stop": 731
}

42
tools/uvr5/lib/lib_v5/modelparams/3band_44100.json Normal file
View File

@ -0,0 +1,42 @@
{
"bins": 768,
"unstable_bins": 5,
"reduction_bins": 733,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 768,
"crop_start": 0,
"crop_stop": 278,
"lpf_start": 28,
"lpf_stop": 140,
"res_type": "polyphase"
},
"2": {
"sr": 22050,
"hl": 256,
"n_fft": 768,
"crop_start": 14,
"crop_stop": 322,
"hpf_start": 70,
"hpf_stop": 14,
"lpf_start": 283,
"lpf_stop": 314,
"res_type": "polyphase"
},
"3": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 131,
"crop_stop": 313,
"hpf_start": 154,
"hpf_stop": 141,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 757,
"pre_filter_stop": 768
}

43
tools/uvr5/lib/lib_v5/modelparams/3band_44100_mid.json Normal file
View File

@ -0,0 +1,43 @@
{
"mid_side": true,
"bins": 768,
"unstable_bins": 5,
"reduction_bins": 733,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 768,
"crop_start": 0,
"crop_stop": 278,
"lpf_start": 28,
"lpf_stop": 140,
"res_type": "polyphase"
},
"2": {
"sr": 22050,
"hl": 256,
"n_fft": 768,
"crop_start": 14,
"crop_stop": 322,
"hpf_start": 70,
"hpf_stop": 14,
"lpf_start": 283,
"lpf_stop": 314,
"res_type": "polyphase"
},
"3": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 131,
"crop_stop": 313,
"hpf_start": 154,
"hpf_stop": 141,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 757,
"pre_filter_stop": 768
}

43
tools/uvr5/lib/lib_v5/modelparams/3band_44100_msb2.json Normal file
View File

@ -0,0 +1,43 @@
{
"mid_side_b2": true,
"bins": 640,
"unstable_bins": 7,
"reduction_bins": 565,
"band": {
"1": {
"sr": 11025,
"hl": 108,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 187,
"lpf_start": 92,
"lpf_stop": 186,
"res_type": "polyphase"
},
"2": {
"sr": 22050,
"hl": 216,
"n_fft": 768,
"crop_start": 0,
"crop_stop": 212,
"hpf_start": 68,
"hpf_stop": 34,
"lpf_start": 174,
"lpf_stop": 209,
"res_type": "polyphase"
},
"3": {
"sr": 44100,
"hl": 432,
"n_fft": 640,
"crop_start": 66,
"crop_stop": 307,
"hpf_start": 86,
"hpf_stop": 72,
"res_type": "kaiser_fast"
}
},
"sr": 44100,
"pre_filter_start": 639,
"pre_filter_stop": 640
}

54
tools/uvr5/lib/lib_v5/modelparams/4band_44100.json Normal file
View File

@ -0,0 +1,54 @@
{
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 668,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 186,
"lpf_start": 37,
"lpf_stop": 73,
"res_type": "polyphase"
},
"2": {
"sr": 11025,
"hl": 128,
"n_fft": 512,
"crop_start": 4,
"crop_stop": 185,
"hpf_start": 36,
"hpf_stop": 18,
"lpf_start": 93,
"lpf_stop": 185,
"res_type": "polyphase"
},
"3": {
"sr": 22050,
"hl": 256,
"n_fft": 512,
"crop_start": 46,
"crop_stop": 186,
"hpf_start": 93,
"hpf_stop": 46,
"lpf_start": 164,
"lpf_stop": 186,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 121,
"crop_stop": 382,
"hpf_start": 138,
"hpf_stop": 123,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 740,
"pre_filter_stop": 768
}

55
tools/uvr5/lib/lib_v5/modelparams/4band_44100_mid.json Normal file
View File

@ -0,0 +1,55 @@
{
"bins": 768,
"unstable_bins": 7,
"mid_side": true,
"reduction_bins": 668,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 186,
"lpf_start": 37,
"lpf_stop": 73,
"res_type": "polyphase"
},
"2": {
"sr": 11025,
"hl": 128,
"n_fft": 512,
"crop_start": 4,
"crop_stop": 185,
"hpf_start": 36,
"hpf_stop": 18,
"lpf_start": 93,
"lpf_stop": 185,
"res_type": "polyphase"
},
"3": {
"sr": 22050,
"hl": 256,
"n_fft": 512,
"crop_start": 46,
"crop_stop": 186,
"hpf_start": 93,
"hpf_stop": 46,
"lpf_start": 164,
"lpf_stop": 186,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 121,
"crop_stop": 382,
"hpf_start": 138,
"hpf_stop": 123,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 740,
"pre_filter_stop": 768
}

55
tools/uvr5/lib/lib_v5/modelparams/4band_44100_msb.json Normal file
View File

@ -0,0 +1,55 @@
{
"mid_side_b": true,
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 668,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 186,
"lpf_start": 37,
"lpf_stop": 73,
"res_type": "polyphase"
},
"2": {
"sr": 11025,
"hl": 128,
"n_fft": 512,
"crop_start": 4,
"crop_stop": 185,
"hpf_start": 36,
"hpf_stop": 18,
"lpf_start": 93,
"lpf_stop": 185,
"res_type": "polyphase"
},
"3": {
"sr": 22050,
"hl": 256,
"n_fft": 512,
"crop_start": 46,
"crop_stop": 186,
"hpf_start": 93,
"hpf_stop": 46,
"lpf_start": 164,
"lpf_stop": 186,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 121,
"crop_stop": 382,
"hpf_start": 138,
"hpf_stop": 123,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 740,
"pre_filter_stop": 768
}

55
tools/uvr5/lib/lib_v5/modelparams/4band_44100_msb2.json Normal file
View File

@ -0,0 +1,55 @@
{
"mid_side_b": true,
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 668,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 186,
"lpf_start": 37,
"lpf_stop": 73,
"res_type": "polyphase"
},
"2": {
"sr": 11025,
"hl": 128,
"n_fft": 512,
"crop_start": 4,
"crop_stop": 185,
"hpf_start": 36,
"hpf_stop": 18,
"lpf_start": 93,
"lpf_stop": 185,
"res_type": "polyphase"
},
"3": {
"sr": 22050,
"hl": 256,
"n_fft": 512,
"crop_start": 46,
"crop_stop": 186,
"hpf_start": 93,
"hpf_stop": 46,
"lpf_start": 164,
"lpf_stop": 186,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 121,
"crop_stop": 382,
"hpf_start": 138,
"hpf_stop": 123,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 740,
"pre_filter_stop": 768
}

55
tools/uvr5/lib/lib_v5/modelparams/4band_44100_reverse.json Normal file
View File

@ -0,0 +1,55 @@
{
"reverse": true,
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 668,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 186,
"lpf_start": 37,
"lpf_stop": 73,
"res_type": "polyphase"
},
"2": {
"sr": 11025,
"hl": 128,
"n_fft": 512,
"crop_start": 4,
"crop_stop": 185,
"hpf_start": 36,
"hpf_stop": 18,
"lpf_start": 93,
"lpf_stop": 185,
"res_type": "polyphase"
},
"3": {
"sr": 22050,
"hl": 256,
"n_fft": 512,
"crop_start": 46,
"crop_stop": 186,
"hpf_start": 93,
"hpf_stop": 46,
"lpf_start": 164,
"lpf_stop": 186,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 121,
"crop_stop": 382,
"hpf_start": 138,
"hpf_stop": 123,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 740,
"pre_filter_stop": 768
}

55
tools/uvr5/lib/lib_v5/modelparams/4band_44100_sw.json Normal file
View File

@ -0,0 +1,55 @@
{
"stereo_w": true,
"bins": 768,
"unstable_bins": 7,
"reduction_bins": 668,
"band": {
"1": {
"sr": 11025,
"hl": 128,
"n_fft": 1024,
"crop_start": 0,
"crop_stop": 186,
"lpf_start": 37,
"lpf_stop": 73,
"res_type": "polyphase"
},
"2": {
"sr": 11025,
"hl": 128,
"n_fft": 512,
"crop_start": 4,
"crop_stop": 185,
"hpf_start": 36,
"hpf_stop": 18,
"lpf_start": 93,
"lpf_stop": 185,
"res_type": "polyphase"
},
"3": {
"sr": 22050,
"hl": 256,
"n_fft": 512,
"crop_start": 46,
"crop_stop": 186,
"hpf_start": 93,
"hpf_stop": 46,
"lpf_start": 164,
"lpf_stop": 186,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 512,
"n_fft": 768,
"crop_start": 121,
"crop_stop": 382,
"hpf_start": 138,
"hpf_stop": 123,
"res_type": "sinc_medium"
}
},
"sr": 44100,
"pre_filter_start": 740,
"pre_filter_stop": 768
}

54
tools/uvr5/lib/lib_v5/modelparams/4band_v2.json Normal file
View File

@ -0,0 +1,54 @@
{
"bins": 672,
"unstable_bins": 8,
"reduction_bins": 637,
"band": {
"1": {
"sr": 7350,
"hl": 80,
"n_fft": 640,
"crop_start": 0,
"crop_stop": 85,
"lpf_start": 25,
"lpf_stop": 53,
"res_type": "polyphase"
},
"2": {
"sr": 7350,
"hl": 80,
"n_fft": 320,
"crop_start": 4,
"crop_stop": 87,
"hpf_start": 25,
"hpf_stop": 12,
"lpf_start": 31,
"lpf_stop": 62,
"res_type": "polyphase"
},
"3": {
"sr": 14700,
"hl": 160,
"n_fft": 512,
"crop_start": 17,
"crop_stop": 216,
"hpf_start": 48,
"hpf_stop": 24,
"lpf_start": 139,
"lpf_stop": 210,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 480,
"n_fft": 960,
"crop_start": 78,
"crop_stop": 383,
"hpf_start": 130,
"hpf_stop": 86,
"res_type": "kaiser_fast"
}
},
"sr": 44100,
"pre_filter_start": 668,
"pre_filter_stop": 672
}

55
tools/uvr5/lib/lib_v5/modelparams/4band_v2_sn.json Normal file
View File

@ -0,0 +1,55 @@
{
"bins": 672,
"unstable_bins": 8,
"reduction_bins": 637,
"band": {
"1": {
"sr": 7350,
"hl": 80,
"n_fft": 640,
"crop_start": 0,
"crop_stop": 85,
"lpf_start": 25,
"lpf_stop": 53,
"res_type": "polyphase"
},
"2": {
"sr": 7350,
"hl": 80,
"n_fft": 320,
"crop_start": 4,
"crop_stop": 87,
"hpf_start": 25,
"hpf_stop": 12,
"lpf_start": 31,
"lpf_stop": 62,
"res_type": "polyphase"
},
"3": {
"sr": 14700,
"hl": 160,
"n_fft": 512,
"crop_start": 17,
"crop_stop": 216,
"hpf_start": 48,
"hpf_stop": 24,
"lpf_start": 139,
"lpf_stop": 210,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 480,
"n_fft": 960,
"crop_start": 78,
"crop_stop": 383,
"hpf_start": 130,
"hpf_stop": 86,
"convert_channels": "stereo_n",
"res_type": "kaiser_fast"
}
},
"sr": 44100,
"pre_filter_start": 668,
"pre_filter_stop": 672
}

54
tools/uvr5/lib/lib_v5/modelparams/4band_v3.json Normal file
View File

@ -0,0 +1,54 @@
{
"bins": 672,
"unstable_bins": 8,
"reduction_bins": 530,
"band": {
"1": {
"sr": 7350,
"hl": 80,
"n_fft": 640,
"crop_start": 0,
"crop_stop": 85,
"lpf_start": 25,
"lpf_stop": 53,
"res_type": "polyphase"
},
"2": {
"sr": 7350,
"hl": 80,
"n_fft": 320,
"crop_start": 4,
"crop_stop": 87,
"hpf_start": 25,
"hpf_stop": 12,
"lpf_start": 31,
"lpf_stop": 62,
"res_type": "polyphase"
},
"3": {
"sr": 14700,
"hl": 160,
"n_fft": 512,
"crop_start": 17,
"crop_stop": 216,
"hpf_start": 48,
"hpf_stop": 24,
"lpf_start": 139,
"lpf_stop": 210,
"res_type": "polyphase"
},
"4": {
"sr": 44100,
"hl": 480,
"n_fft": 960,
"crop_start": 78,
"crop_stop": 383,
"hpf_start": 130,
"hpf_stop": 86,
"res_type": "kaiser_fast"
}
},
"sr": 44100,
"pre_filter_start": 668,
"pre_filter_stop": 672
}

43
tools/uvr5/lib/lib_v5/modelparams/ensemble.json Normal file
View File

@ -0,0 +1,43 @@
{
"mid_side_b2": true,
"bins": 1280,
"unstable_bins": 7,
"reduction_bins": 565,
"band": {
"1": {
"sr": 11025,
"hl": 108,
"n_fft": 2048,
"crop_start": 0,
"crop_stop": 374,
"lpf_start": 92,
"lpf_stop": 186,
"res_type": "polyphase"
},
"2": {
"sr": 22050,
"hl": 216,
"n_fft": 1536,
"crop_start": 0,
"crop_stop": 424,
"hpf_start": 68,
"hpf_stop": 34,
"lpf_start": 348,
"lpf_stop": 418,
"res_type": "polyphase"
},
"3": {
"sr": 44100,
"hl": 432,
"n_fft": 1280,
"crop_start": 132,
"crop_stop": 614,
"hpf_start": 172,
"hpf_stop": 144,
"res_type": "polyphase"
}
},
"sr": 44100,
"pre_filter_start": 1280,
"pre_filter_stop": 1280
}

123
tools/uvr5/lib/lib_v5/nets.py Normal file
View File

@ -0,0 +1,123 @@
import layers
import torch
import torch.nn.functional as F
from torch import nn
from . import spec_utils
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 16)
self.stg1_high_band_net = BaseASPPNet(2, 16)
self.stg2_bridge = layers.Conv2DBNActiv(18, 8, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(8, 16)
self.stg3_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(16, 32)
self.out = nn.Conv2d(32, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(16, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(16, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

122
tools/uvr5/lib/lib_v5/nets_123812KB.py Normal file
View File

@ -0,0 +1,122 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_123821KB as layers
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 32)
self.stg1_high_band_net = BaseASPPNet(2, 32)
self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(16, 32)
self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(32, 64)
self.out = nn.Conv2d(64, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(32, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

122
tools/uvr5/lib/lib_v5/nets_123821KB.py Normal file
View File

@ -0,0 +1,122 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_123821KB as layers
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 32)
self.stg1_high_band_net = BaseASPPNet(2, 32)
self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(16, 32)
self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(32, 64)
self.out = nn.Conv2d(64, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(32, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

122
tools/uvr5/lib/lib_v5/nets_33966KB.py Normal file
View File

@ -0,0 +1,122 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_33966KB as layers
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16, 32)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 16)
self.stg1_high_band_net = BaseASPPNet(2, 16)
self.stg2_bridge = layers.Conv2DBNActiv(18, 8, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(8, 16)
self.stg3_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(16, 32)
self.out = nn.Conv2d(32, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(16, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(16, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

123
tools/uvr5/lib/lib_v5/nets_537227KB.py Normal file
View File

@ -0,0 +1,123 @@
import numpy as np
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_537238KB as layers
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 64)
self.stg1_high_band_net = BaseASPPNet(2, 64)
self.stg2_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(32, 64)
self.stg3_bridge = layers.Conv2DBNActiv(130, 64, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(64, 128)
self.out = nn.Conv2d(128, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(64, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(64, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

123
tools/uvr5/lib/lib_v5/nets_537238KB.py Normal file
View File

@ -0,0 +1,123 @@
import numpy as np
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_537238KB as layers
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 64)
self.stg1_high_band_net = BaseASPPNet(2, 64)
self.stg2_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(32, 64)
self.stg3_bridge = layers.Conv2DBNActiv(130, 64, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(64, 128)
self.out = nn.Conv2d(128, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(64, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(64, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

122
tools/uvr5/lib/lib_v5/nets_61968KB.py Normal file
View File

@ -0,0 +1,122 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_123821KB as layers
class BaseASPPNet(nn.Module):
def __init__(self, nin, ch, dilations=(4, 8, 16)):
super(BaseASPPNet, self).__init__()
self.enc1 = layers.Encoder(nin, ch, 3, 2, 1)
self.enc2 = layers.Encoder(ch, ch * 2, 3, 2, 1)
self.enc3 = layers.Encoder(ch * 2, ch * 4, 3, 2, 1)
self.enc4 = layers.Encoder(ch * 4, ch * 8, 3, 2, 1)
self.aspp = layers.ASPPModule(ch * 8, ch * 16, dilations)
self.dec4 = layers.Decoder(ch * (8 + 16), ch * 8, 3, 1, 1)
self.dec3 = layers.Decoder(ch * (4 + 8), ch * 4, 3, 1, 1)
self.dec2 = layers.Decoder(ch * (2 + 4), ch * 2, 3, 1, 1)
self.dec1 = layers.Decoder(ch * (1 + 2), ch, 3, 1, 1)
def __call__(self, x):
h, e1 = self.enc1(x)
h, e2 = self.enc2(h)
h, e3 = self.enc3(h)
h, e4 = self.enc4(h)
h = self.aspp(h)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = self.dec1(h, e1)
return h
class CascadedASPPNet(nn.Module):
def __init__(self, n_fft):
super(CascadedASPPNet, self).__init__()
self.stg1_low_band_net = BaseASPPNet(2, 32)
self.stg1_high_band_net = BaseASPPNet(2, 32)
self.stg2_bridge = layers.Conv2DBNActiv(34, 16, 1, 1, 0)
self.stg2_full_band_net = BaseASPPNet(16, 32)
self.stg3_bridge = layers.Conv2DBNActiv(66, 32, 1, 1, 0)
self.stg3_full_band_net = BaseASPPNet(32, 64)
self.out = nn.Conv2d(64, 2, 1, bias=False)
self.aux1_out = nn.Conv2d(32, 2, 1, bias=False)
self.aux2_out = nn.Conv2d(32, 2, 1, bias=False)
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.offset = 128
def forward(self, x, aggressiveness=None):
mix = x.detach()
x = x.clone()
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
aux1 = torch.cat(
[
self.stg1_low_band_net(x[:, :, :bandw]),
self.stg1_high_band_net(x[:, :, bandw:]),
],
dim=2,
)
h = torch.cat([x, aux1], dim=1)
aux2 = self.stg2_full_band_net(self.stg2_bridge(h))
h = torch.cat([x, aux1, aux2], dim=1)
h = self.stg3_full_band_net(self.stg3_bridge(h))
mask = torch.sigmoid(self.out(h))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux1 = torch.sigmoid(self.aux1_out(aux1))
aux1 = F.pad(
input=aux1,
pad=(0, 0, 0, self.output_bin - aux1.size()[2]),
mode="replicate",
)
aux2 = torch.sigmoid(self.aux2_out(aux2))
aux2 = F.pad(
input=aux2,
pad=(0, 0, 0, self.output_bin - aux2.size()[2]),
mode="replicate",
)
return mask * mix, aux1 * mix, aux2 * mix
else:
if aggressiveness:
mask[:, :, : aggressiveness["split_bin"]] = torch.pow(
mask[:, :, : aggressiveness["split_bin"]],
1 + aggressiveness["value"] / 3,
)
mask[:, :, aggressiveness["split_bin"] :] = torch.pow(
mask[:, :, aggressiveness["split_bin"] :],
1 + aggressiveness["value"],
)
return mask * mix
def predict(self, x_mag, aggressiveness=None):
h = self.forward(x_mag, aggressiveness)
if self.offset > 0:
h = h[:, :, :, self.offset : -self.offset]
assert h.size()[3] > 0
return h

133
tools/uvr5/lib/lib_v5/nets_new.py Normal file
View File

@ -0,0 +1,133 @@
import torch
import torch.nn.functional as F
from torch import nn
from . import layers_new
class BaseNet(nn.Module):
def __init__(
self, nin, nout, nin_lstm, nout_lstm, dilations=((4, 2), (8, 4), (12, 6))
):
super(BaseNet, self).__init__()
self.enc1 = layers_new.Conv2DBNActiv(nin, nout, 3, 1, 1)
self.enc2 = layers_new.Encoder(nout, nout * 2, 3, 2, 1)
self.enc3 = layers_new.Encoder(nout * 2, nout * 4, 3, 2, 1)
self.enc4 = layers_new.Encoder(nout * 4, nout * 6, 3, 2, 1)
self.enc5 = layers_new.Encoder(nout * 6, nout * 8, 3, 2, 1)
self.aspp = layers_new.ASPPModule(nout * 8, nout * 8, dilations, dropout=True)
self.dec4 = layers_new.Decoder(nout * (6 + 8), nout * 6, 3, 1, 1)
self.dec3 = layers_new.Decoder(nout * (4 + 6), nout * 4, 3, 1, 1)
self.dec2 = layers_new.Decoder(nout * (2 + 4), nout * 2, 3, 1, 1)
self.lstm_dec2 = layers_new.LSTMModule(nout * 2, nin_lstm, nout_lstm)
self.dec1 = layers_new.Decoder(nout * (1 + 2) + 1, nout * 1, 3, 1, 1)
def __call__(self, x):
e1 = self.enc1(x)
e2 = self.enc2(e1)
e3 = self.enc3(e2)
e4 = self.enc4(e3)
e5 = self.enc5(e4)
h = self.aspp(e5)
h = self.dec4(h, e4)
h = self.dec3(h, e3)
h = self.dec2(h, e2)
h = torch.cat([h, self.lstm_dec2(h)], dim=1)
h = self.dec1(h, e1)
return h
class CascadedNet(nn.Module):
def __init__(self, n_fft, nout=32, nout_lstm=128):
super(CascadedNet, self).__init__()
self.max_bin = n_fft // 2
self.output_bin = n_fft // 2 + 1
self.nin_lstm = self.max_bin // 2
self.offset = 64
self.stg1_low_band_net = nn.Sequential(
BaseNet(2, nout // 2, self.nin_lstm // 2, nout_lstm),
layers_new.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.stg2_low_band_net = nn.Sequential(
BaseNet(nout // 4 + 2, nout, self.nin_lstm // 2, nout_lstm),
layers_new.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.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.aux_out = nn.Conv2d(3 * nout // 4, 2, 1, bias=False)
def forward(self, x):
x = x[:, :, : self.max_bin]
bandw = x.size()[2] // 2
l1_in = x[:, :, :bandw]
h1_in = x[:, :, bandw:]
l1 = self.stg1_low_band_net(l1_in)
h1 = self.stg1_high_band_net(h1_in)
aux1 = torch.cat([l1, h1], dim=2)
l2_in = torch.cat([l1_in, l1], dim=1)
h2_in = torch.cat([h1_in, h1], dim=1)
l2 = self.stg2_low_band_net(l2_in)
h2 = self.stg2_high_band_net(h2_in)
aux2 = torch.cat([l2, h2], dim=2)
f3_in = torch.cat([x, aux1, aux2], dim=1)
f3 = self.stg3_full_band_net(f3_in)
mask = torch.sigmoid(self.out(f3))
mask = F.pad(
input=mask,
pad=(0, 0, 0, self.output_bin - mask.size()[2]),
mode="replicate",
)
if self.training:
aux = torch.cat([aux1, aux2], dim=1)
aux = torch.sigmoid(self.aux_out(aux))
aux = F.pad(
input=aux,
pad=(0, 0, 0, self.output_bin - aux.size()[2]),
mode="replicate",
)
return mask, aux
else:
return mask
def predict_mask(self, x):
mask = self.forward(x)
if self.offset > 0:
mask = mask[:, :, :, self.offset : -self.offset]
assert mask.size()[3] > 0
return mask
def predict(self, x, aggressiveness=None):
mask = self.forward(x)
pred_mag = x * mask
if self.offset > 0:
pred_mag = pred_mag[:, :, :, self.offset : -self.offset]
assert pred_mag.size()[3] > 0
return pred_mag

672
tools/uvr5/lib/lib_v5/spec_utils.py Normal file
View File

@ -0,0 +1,672 @@
import hashlib
import json
import math
import os
import librosa
import numpy as np
import soundfile as sf
from tqdm import tqdm
def crop_center(h1, h2):
h1_shape = h1.size()
h2_shape = h2.size()
if h1_shape[3] == h2_shape[3]:
return h1
elif h1_shape[3] < h2_shape[3]:
raise ValueError("h1_shape[3] must be greater than h2_shape[3]")
# s_freq = (h2_shape[2] - h1_shape[2]) // 2
# e_freq = s_freq + h1_shape[2]
s_time = (h1_shape[3] - h2_shape[3]) // 2
e_time = s_time + h2_shape[3]
h1 = h1[:, :, :, s_time:e_time]
return h1
def wave_to_spectrogram(
wave, hop_length, n_fft, mid_side=False, mid_side_b2=False, reverse=False
):
if reverse:
wave_left = np.flip(np.asfortranarray(wave[0]))
wave_right = np.flip(np.asfortranarray(wave[1]))
elif mid_side:
wave_left = np.asfortranarray(np.add(wave[0], wave[1]) / 2)
wave_right = np.asfortranarray(np.subtract(wave[0], wave[1]))
elif mid_side_b2:
wave_left = np.asfortranarray(np.add(wave[1], wave[0] * 0.5))
wave_right = np.asfortranarray(np.subtract(wave[0], wave[1] * 0.5))
else:
wave_left = np.asfortranarray(wave[0])
wave_right = np.asfortranarray(wave[1])
spec_left = librosa.stft(wave_left, n_fft, hop_length=hop_length)
spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length)
spec = np.asfortranarray([spec_left, spec_right])
return spec
def wave_to_spectrogram_mt(
wave, hop_length, n_fft, mid_side=False, mid_side_b2=False, reverse=False
):
import threading
if reverse:
wave_left = np.flip(np.asfortranarray(wave[0]))
wave_right = np.flip(np.asfortranarray(wave[1]))
elif mid_side:
wave_left = np.asfortranarray(np.add(wave[0], wave[1]) / 2)
wave_right = np.asfortranarray(np.subtract(wave[0], wave[1]))
elif mid_side_b2:
wave_left = np.asfortranarray(np.add(wave[1], wave[0] * 0.5))
wave_right = np.asfortranarray(np.subtract(wave[0], wave[1] * 0.5))
else:
wave_left = np.asfortranarray(wave[0])
wave_right = np.asfortranarray(wave[1])
def run_thread(**kwargs):
global spec_left
spec_left = librosa.stft(**kwargs)
thread = threading.Thread(
target=run_thread,
kwargs={"y": wave_left, "n_fft": n_fft, "hop_length": hop_length},
)
thread.start()
spec_right = librosa.stft(wave_right, n_fft, hop_length=hop_length)
thread.join()
spec = np.asfortranarray([spec_left, spec_right])
return spec
def combine_spectrograms(specs, mp):
l = min([specs[i].shape[2] for i in specs])
spec_c = np.zeros(shape=(2, mp.param["bins"] + 1, l), dtype=np.complex64)
offset = 0
bands_n = len(mp.param["band"])
for d in range(1, bands_n + 1):
h = mp.param["band"][d]["crop_stop"] - mp.param["band"][d]["crop_start"]
spec_c[:, offset : offset + h, :l] = specs[d][
:, mp.param["band"][d]["crop_start"] : mp.param["band"][d]["crop_stop"], :l
]
offset += h
if offset > mp.param["bins"]:
raise ValueError("Too much bins")
# lowpass fiter
if (
mp.param["pre_filter_start"] > 0
): # and mp.param['band'][bands_n]['res_type'] in ['scipy', 'polyphase']:
if bands_n == 1:
spec_c = fft_lp_filter(
spec_c, mp.param["pre_filter_start"], mp.param["pre_filter_stop"]
)
else:
gp = 1
for b in range(
mp.param["pre_filter_start"] + 1, mp.param["pre_filter_stop"]
):
g = math.pow(
10, -(b - mp.param["pre_filter_start"]) * (3.5 - gp) / 20.0
)
gp = g
spec_c[:, b, :] *= g
return np.asfortranarray(spec_c)
def spectrogram_to_image(spec, mode="magnitude"):
if mode == "magnitude":
if np.iscomplexobj(spec):
y = np.abs(spec)
else:
y = spec
y = np.log10(y**2 + 1e-8)
elif mode == "phase":
if np.iscomplexobj(spec):
y = np.angle(spec)
else:
y = spec
y -= y.min()
y *= 255 / y.max()
img = np.uint8(y)
if y.ndim == 3:
img = img.transpose(1, 2, 0)
img = np.concatenate([np.max(img, axis=2, keepdims=True), img], axis=2)
return img
def reduce_vocal_aggressively(X, y, softmask):
v = X - y
y_mag_tmp = np.abs(y)
v_mag_tmp = np.abs(v)
v_mask = v_mag_tmp > y_mag_tmp
y_mag = np.clip(y_mag_tmp - v_mag_tmp * v_mask * softmask, 0, np.inf)
return y_mag * np.exp(1.0j * np.angle(y))
def mask_silence(mag, ref, thres=0.2, min_range=64, fade_size=32):
if min_range < fade_size * 2:
raise ValueError("min_range must be >= fade_area * 2")
mag = mag.copy()
idx = np.where(ref.mean(axis=(0, 1)) < thres)[0]
starts = np.insert(idx[np.where(np.diff(idx) != 1)[0] + 1], 0, idx[0])
ends = np.append(idx[np.where(np.diff(idx) != 1)[0]], idx[-1])
uninformative = np.where(ends - starts > min_range)[0]
if len(uninformative) > 0:
starts = starts[uninformative]
ends = ends[uninformative]
old_e = None
for s, e in zip(starts, ends):
if old_e is not None and s - old_e < fade_size:
s = old_e - fade_size * 2
if s != 0:
weight = np.linspace(0, 1, fade_size)
mag[:, :, s : s + fade_size] += weight * ref[:, :, s : s + fade_size]
else:
s -= fade_size
if e != mag.shape[2]:
weight = np.linspace(1, 0, fade_size)
mag[:, :, e - fade_size : e] += weight * ref[:, :, e - fade_size : e]
else:
e += fade_size
mag[:, :, s + fade_size : e - fade_size] += ref[
:, :, s + fade_size : e - fade_size
]
old_e = e
return mag
def align_wave_head_and_tail(a, b):
l = min([a[0].size, b[0].size])
return a[:l, :l], b[:l, :l]
def cache_or_load(mix_path, inst_path, mp):
mix_basename = os.path.splitext(os.path.basename(mix_path))[0]
inst_basename = os.path.splitext(os.path.basename(inst_path))[0]
cache_dir = "mph{}".format(
hashlib.sha1(json.dumps(mp.param, sort_keys=True).encode("utf-8")).hexdigest()
)
mix_cache_dir = os.path.join("cache", cache_dir)
inst_cache_dir = os.path.join("cache", cache_dir)
os.makedirs(mix_cache_dir, exist_ok=True)
os.makedirs(inst_cache_dir, exist_ok=True)
mix_cache_path = os.path.join(mix_cache_dir, mix_basename + ".npy")
inst_cache_path = os.path.join(inst_cache_dir, inst_basename + ".npy")
if os.path.exists(mix_cache_path) and os.path.exists(inst_cache_path):
X_spec_m = np.load(mix_cache_path)
y_spec_m = np.load(inst_cache_path)
else:
X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {}
for d in range(len(mp.param["band"]), 0, -1):
bp = mp.param["band"][d]
if d == len(mp.param["band"]): # high-end band
X_wave[d], _ = librosa.load(
mix_path, bp["sr"], False, dtype=np.float32, res_type=bp["res_type"]
)
y_wave[d], _ = librosa.load(
inst_path,
bp["sr"],
False,
dtype=np.float32,
res_type=bp["res_type"],
)
else: # lower bands
X_wave[d] = librosa.resample(
X_wave[d + 1],
mp.param["band"][d + 1]["sr"],
bp["sr"],
res_type=bp["res_type"],
)
y_wave[d] = librosa.resample(
y_wave[d + 1],
mp.param["band"][d + 1]["sr"],
bp["sr"],
res_type=bp["res_type"],
)
X_wave[d], y_wave[d] = align_wave_head_and_tail(X_wave[d], y_wave[d])
X_spec_s[d] = wave_to_spectrogram(
X_wave[d],
bp["hl"],
bp["n_fft"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
)
y_spec_s[d] = wave_to_spectrogram(
y_wave[d],
bp["hl"],
bp["n_fft"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
)
del X_wave, y_wave
X_spec_m = combine_spectrograms(X_spec_s, mp)
y_spec_m = combine_spectrograms(y_spec_s, mp)
if X_spec_m.shape != y_spec_m.shape:
raise ValueError("The combined spectrograms are different: " + mix_path)
_, ext = os.path.splitext(mix_path)
np.save(mix_cache_path, X_spec_m)
np.save(inst_cache_path, y_spec_m)
return X_spec_m, y_spec_m
def spectrogram_to_wave(spec, hop_length, mid_side, mid_side_b2, reverse):
spec_left = np.asfortranarray(spec[0])
spec_right = np.asfortranarray(spec[1])
wave_left = librosa.istft(spec_left, hop_length=hop_length)
wave_right = librosa.istft(spec_right, hop_length=hop_length)
if reverse:
return np.asfortranarray([np.flip(wave_left), np.flip(wave_right)])
elif mid_side:
return np.asfortranarray(
[np.add(wave_left, wave_right / 2), np.subtract(wave_left, wave_right / 2)]
)
elif mid_side_b2:
return np.asfortranarray(
[
np.add(wave_right / 1.25, 0.4 * wave_left),
np.subtract(wave_left / 1.25, 0.4 * wave_right),
]
)
else:
return np.asfortranarray([wave_left, wave_right])
def spectrogram_to_wave_mt(spec, hop_length, mid_side, reverse, mid_side_b2):
import threading
spec_left = np.asfortranarray(spec[0])
spec_right = np.asfortranarray(spec[1])
def run_thread(**kwargs):
global wave_left
wave_left = librosa.istft(**kwargs)
thread = threading.Thread(
target=run_thread, kwargs={"stft_matrix": spec_left, "hop_length": hop_length}
)
thread.start()
wave_right = librosa.istft(spec_right, hop_length=hop_length)
thread.join()
if reverse:
return np.asfortranarray([np.flip(wave_left), np.flip(wave_right)])
elif mid_side:
return np.asfortranarray(
[np.add(wave_left, wave_right / 2), np.subtract(wave_left, wave_right / 2)]
)
elif mid_side_b2:
return np.asfortranarray(
[
np.add(wave_right / 1.25, 0.4 * wave_left),
np.subtract(wave_left / 1.25, 0.4 * wave_right),
]
)
else:
return np.asfortranarray([wave_left, wave_right])
def cmb_spectrogram_to_wave(spec_m, mp, extra_bins_h=None, extra_bins=None):
wave_band = {}
bands_n = len(mp.param["band"])
offset = 0
for d in range(1, bands_n + 1):
bp = mp.param["band"][d]
spec_s = np.ndarray(
shape=(2, bp["n_fft"] // 2 + 1, spec_m.shape[2]), dtype=complex
)
h = bp["crop_stop"] - bp["crop_start"]
spec_s[:, bp["crop_start"] : bp["crop_stop"], :] = spec_m[
:, offset : offset + h, :
]
offset += h
if d == bands_n: # higher
if extra_bins_h: # if --high_end_process bypass
max_bin = bp["n_fft"] // 2
spec_s[:, max_bin - extra_bins_h : max_bin, :] = extra_bins[
:, :extra_bins_h, :
]
if bp["hpf_start"] > 0:
spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1)
if bands_n == 1:
wave = spectrogram_to_wave(
spec_s,
bp["hl"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
)
else:
wave = np.add(
wave,
spectrogram_to_wave(
spec_s,
bp["hl"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
),
)
else:
sr = mp.param["band"][d + 1]["sr"]
if d == 1: # lower
spec_s = fft_lp_filter(spec_s, bp["lpf_start"], bp["lpf_stop"])
wave = librosa.resample(
spectrogram_to_wave(
spec_s,
bp["hl"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
),
bp["sr"],
sr,
res_type="sinc_fastest",
)
else: # mid
spec_s = fft_hp_filter(spec_s, bp["hpf_start"], bp["hpf_stop"] - 1)
spec_s = fft_lp_filter(spec_s, bp["lpf_start"], bp["lpf_stop"])
wave2 = np.add(
wave,
spectrogram_to_wave(
spec_s,
bp["hl"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
),
)
# wave = librosa.core.resample(wave2, bp['sr'], sr, res_type="sinc_fastest")
wave = librosa.core.resample(wave2, bp["sr"], sr, res_type="scipy")
return wave.T
def fft_lp_filter(spec, bin_start, bin_stop):
g = 1.0
for b in range(bin_start, bin_stop):
g -= 1 / (bin_stop - bin_start)
spec[:, b, :] = g * spec[:, b, :]
spec[:, bin_stop:, :] *= 0
return spec
def fft_hp_filter(spec, bin_start, bin_stop):
g = 1.0
for b in range(bin_start, bin_stop, -1):
g -= 1 / (bin_start - bin_stop)
spec[:, b, :] = g * spec[:, b, :]
spec[:, 0 : bin_stop + 1, :] *= 0
return spec
def mirroring(a, spec_m, input_high_end, mp):
if "mirroring" == a:
mirror = np.flip(
np.abs(
spec_m[
:,
mp.param["pre_filter_start"]
- 10
- input_high_end.shape[1] : mp.param["pre_filter_start"]
- 10,
:,
]
),
1,
)
mirror = mirror * np.exp(1.0j * np.angle(input_high_end))
return np.where(
np.abs(input_high_end) <= np.abs(mirror), input_high_end, mirror
)
if "mirroring2" == a:
mirror = np.flip(
np.abs(
spec_m[
:,
mp.param["pre_filter_start"]
- 10
- input_high_end.shape[1] : mp.param["pre_filter_start"]
- 10,
:,
]
),
1,
)
mi = np.multiply(mirror, input_high_end * 1.7)
return np.where(np.abs(input_high_end) <= np.abs(mi), input_high_end, mi)
def ensembling(a, specs):
for i in range(1, len(specs)):
if i == 1:
spec = specs[0]
ln = min([spec.shape[2], specs[i].shape[2]])
spec = spec[:, :, :ln]
specs[i] = specs[i][:, :, :ln]
if "min_mag" == a:
spec = np.where(np.abs(specs[i]) <= np.abs(spec), specs[i], spec)
if "max_mag" == a:
spec = np.where(np.abs(specs[i]) >= np.abs(spec), specs[i], spec)
return spec
def stft(wave, nfft, hl):
wave_left = np.asfortranarray(wave[0])
wave_right = np.asfortranarray(wave[1])
spec_left = librosa.stft(wave_left, nfft, hop_length=hl)
spec_right = librosa.stft(wave_right, nfft, hop_length=hl)
spec = np.asfortranarray([spec_left, spec_right])
return spec
def istft(spec, hl):
spec_left = np.asfortranarray(spec[0])
spec_right = np.asfortranarray(spec[1])
wave_left = librosa.istft(spec_left, hop_length=hl)
wave_right = librosa.istft(spec_right, hop_length=hl)
wave = np.asfortranarray([wave_left, wave_right])
if __name__ == "__main__":
import argparse
import sys
import time
import cv2
from model_param_init import ModelParameters
p = argparse.ArgumentParser()
p.add_argument(
"--algorithm",
"-a",
type=str,
choices=["invert", "invert_p", "min_mag", "max_mag", "deep", "align"],
default="min_mag",
)
p.add_argument(
"--model_params",
"-m",
type=str,
default=os.path.join("modelparams", "1band_sr44100_hl512.json"),
)
p.add_argument("--output_name", "-o", type=str, default="output")
p.add_argument("--vocals_only", "-v", action="store_true")
p.add_argument("input", nargs="+")
args = p.parse_args()
start_time = time.time()
if args.algorithm.startswith("invert") and len(args.input) != 2:
raise ValueError("There should be two input files.")
if not args.algorithm.startswith("invert") and len(args.input) < 2:
raise ValueError("There must be at least two input files.")
wave, specs = {}, {}
mp = ModelParameters(args.model_params)
for i in range(len(args.input)):
spec = {}
for d in range(len(mp.param["band"]), 0, -1):
bp = mp.param["band"][d]
if d == len(mp.param["band"]): # high-end band
wave[d], _ = librosa.load(
args.input[i],
bp["sr"],
False,
dtype=np.float32,
res_type=bp["res_type"],
)
if len(wave[d].shape) == 1: # mono to stereo
wave[d] = np.array([wave[d], wave[d]])
else: # lower bands
wave[d] = librosa.resample(
wave[d + 1],
mp.param["band"][d + 1]["sr"],
bp["sr"],
res_type=bp["res_type"],
)
spec[d] = wave_to_spectrogram(
wave[d],
bp["hl"],
bp["n_fft"],
mp.param["mid_side"],
mp.param["mid_side_b2"],
mp.param["reverse"],
)
specs[i] = combine_spectrograms(spec, mp)
del wave
if args.algorithm == "deep":
d_spec = np.where(np.abs(specs[0]) <= np.abs(spec[1]), specs[0], spec[1])
v_spec = d_spec - specs[1]
sf.write(
os.path.join("{}.wav".format(args.output_name)),
cmb_spectrogram_to_wave(v_spec, mp),
mp.param["sr"],
)
if args.algorithm.startswith("invert"):
ln = min([specs[0].shape[2], specs[1].shape[2]])
specs[0] = specs[0][:, :, :ln]
specs[1] = specs[1][:, :, :ln]
if "invert_p" == args.algorithm:
X_mag = np.abs(specs[0])
y_mag = np.abs(specs[1])
max_mag = np.where(X_mag >= y_mag, X_mag, y_mag)
v_spec = specs[1] - max_mag * np.exp(1.0j * np.angle(specs[0]))
else:
specs[1] = reduce_vocal_aggressively(specs[0], specs[1], 0.2)
v_spec = specs[0] - specs[1]
if not args.vocals_only:
X_mag = np.abs(specs[0])
y_mag = np.abs(specs[1])
v_mag = np.abs(v_spec)
X_image = spectrogram_to_image(X_mag)
y_image = spectrogram_to_image(y_mag)
v_image = spectrogram_to_image(v_mag)
cv2.imwrite("{}_X.png".format(args.output_name), X_image)
cv2.imwrite("{}_y.png".format(args.output_name), y_image)
cv2.imwrite("{}_v.png".format(args.output_name), v_image)
sf.write(
"{}_X.wav".format(args.output_name),
cmb_spectrogram_to_wave(specs[0], mp),
mp.param["sr"],
)
sf.write(
"{}_y.wav".format(args.output_name),
cmb_spectrogram_to_wave(specs[1], mp),
mp.param["sr"],
)
sf.write(
"{}_v.wav".format(args.output_name),
cmb_spectrogram_to_wave(v_spec, mp),
mp.param["sr"],
)
else:
if not args.algorithm == "deep":
sf.write(
os.path.join("ensembled", "{}.wav".format(args.output_name)),
cmb_spectrogram_to_wave(ensembling(args.algorithm, specs), mp),
mp.param["sr"],
)
if args.algorithm == "align":
trackalignment = [
{
"file1": '"{}"'.format(args.input[0]),
"file2": '"{}"'.format(args.input[1]),
}
]
for i, e in tqdm(enumerate(trackalignment), desc="Performing Alignment..."):
os.system(f"python lib/align_tracks.py {e['file1']} {e['file2']}")
# print('Total time: {0:.{1}f}s'.format(time.time() - start_time, 1))

263
tools/uvr5/lib/name_params.json Normal file
View File

@ -0,0 +1,263 @@
{
"equivalent" : [
{
"model_hash_name" : [
{
"hash_name": "47939caf0cfe52a0e81442b85b971dfd",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100"
},
{
"hash_name": "4e4ecb9764c50a8c414fee6e10395bbe",
"model_params": "lib/lib_v5/modelparams/4band_v2.json",
"param_name": "4band_v2"
},
{
"hash_name": "ca106edd563e034bde0bdec4bb7a4b36",
"model_params": "lib/lib_v5/modelparams/4band_v2.json",
"param_name": "4band_v2"
},
{
"hash_name": "e60a1e84803ce4efc0a6551206cc4b71",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100"
},
{
"hash_name": "a82f14e75892e55e994376edbf0c8435",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100"
},
{
"hash_name": "6dd9eaa6f0420af9f1d403aaafa4cc06",
"model_params": "lib/lib_v5/modelparams/4band_v2_sn.json",
"param_name": "4band_v2_sn"
},
{
"hash_name": "08611fb99bd59eaa79ad27c58d137727",
"model_params": "lib/lib_v5/modelparams/4band_v2_sn.json",
"param_name": "4band_v2_sn"
},
{
"hash_name": "5c7bbca45a187e81abbbd351606164e5",
"model_params": "lib/lib_v5/modelparams/3band_44100_msb2.json",
"param_name": "3band_44100_msb2"
},
{
"hash_name": "d6b2cb685a058a091e5e7098192d3233",
"model_params": "lib/lib_v5/modelparams/3band_44100_msb2.json",
"param_name": "3band_44100_msb2"
},
{
"hash_name": "c1b9f38170a7c90e96f027992eb7c62b",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100"
},
{
"hash_name": "c3448ec923fa0edf3d03a19e633faa53",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100"
},
{
"hash_name": "68aa2c8093d0080704b200d140f59e54",
"model_params": "lib/lib_v5/modelparams/3band_44100.json",
"param_name": "3band_44100"
},
{
"hash_name": "fdc83be5b798e4bd29fe00fe6600e147",
"model_params": "lib/lib_v5/modelparams/3band_44100_mid.json",
"param_name": "3band_44100_mid.json"
},
{
"hash_name": "2ce34bc92fd57f55db16b7a4def3d745",
"model_params": "lib/lib_v5/modelparams/3band_44100_mid.json",
"param_name": "3band_44100_mid.json"
},
{
"hash_name": "52fdca89576f06cf4340b74a4730ee5f",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100.json"
},
{
"hash_name": "41191165b05d38fc77f072fa9e8e8a30",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100.json"
},
{
"hash_name": "89e83b511ad474592689e562d5b1f80e",
"model_params": "lib/lib_v5/modelparams/2band_32000.json",
"param_name": "2band_32000.json"
},
{
"hash_name": "0b954da81d453b716b114d6d7c95177f",
"model_params": "lib/lib_v5/modelparams/2band_32000.json",
"param_name": "2band_32000.json"
}
],
"v4 Models": [
{
"hash_name": "6a00461c51c2920fd68937d4609ed6c8",
"model_params": "lib/lib_v5/modelparams/1band_sr16000_hl512.json",
"param_name": "1band_sr16000_hl512"
},
{
"hash_name": "0ab504864d20f1bd378fe9c81ef37140",
"model_params": "lib/lib_v5/modelparams/1band_sr32000_hl512.json",
"param_name": "1band_sr32000_hl512"
},
{
"hash_name": "7dd21065bf91c10f7fccb57d7d83b07f",
"model_params": "lib/lib_v5/modelparams/1band_sr32000_hl512.json",
"param_name": "1band_sr32000_hl512"
},
{
"hash_name": "80ab74d65e515caa3622728d2de07d23",
"model_params": "lib/lib_v5/modelparams/1band_sr32000_hl512.json",
"param_name": "1band_sr32000_hl512"
},
{
"hash_name": "edc115e7fc523245062200c00caa847f",
"model_params": "lib/lib_v5/modelparams/1band_sr33075_hl384.json",
"param_name": "1band_sr33075_hl384"
},
{
"hash_name": "28063e9f6ab5b341c5f6d3c67f2045b7",
"model_params": "lib/lib_v5/modelparams/1band_sr33075_hl384.json",
"param_name": "1band_sr33075_hl384"
},
{
"hash_name": "b58090534c52cbc3e9b5104bad666ef2",
"model_params": "lib/lib_v5/modelparams/1band_sr44100_hl512.json",
"param_name": "1band_sr44100_hl512"
},
{
"hash_name": "0cdab9947f1b0928705f518f3c78ea8f",
"model_params": "lib/lib_v5/modelparams/1band_sr44100_hl512.json",
"param_name": "1band_sr44100_hl512"
},
{
"hash_name": "ae702fed0238afb5346db8356fe25f13",
"model_params": "lib/lib_v5/modelparams/1band_sr44100_hl1024.json",
"param_name": "1band_sr44100_hl1024"
}
]
}
],
"User Models" : [
{
"1 Band": [
{
"hash_name": "1band_sr16000_hl512",
"model_params": "lib/lib_v5/modelparams/1band_sr16000_hl512.json",
"param_name": "1band_sr16000_hl512"
},
{
"hash_name": "1band_sr32000_hl512",
"model_params": "lib/lib_v5/modelparams/1band_sr32000_hl512.json",
"param_name": "1band_sr16000_hl512"
},
{
"hash_name": "1band_sr33075_hl384",
"model_params": "lib/lib_v5/modelparams/1band_sr33075_hl384.json",
"param_name": "1band_sr33075_hl384"
},
{
"hash_name": "1band_sr44100_hl256",
"model_params": "lib/lib_v5/modelparams/1band_sr44100_hl256.json",
"param_name": "1band_sr44100_hl256"
},
{
"hash_name": "1band_sr44100_hl512",
"model_params": "lib/lib_v5/modelparams/1band_sr44100_hl512.json",
"param_name": "1band_sr44100_hl512"
},
{
"hash_name": "1band_sr44100_hl1024",
"model_params": "lib/lib_v5/modelparams/1band_sr44100_hl1024.json",
"param_name": "1band_sr44100_hl1024"
}
],
"2 Band": [
{
"hash_name": "2band_44100_lofi",
"model_params": "lib/lib_v5/modelparams/2band_44100_lofi.json",
"param_name": "2band_44100_lofi"
},
{
"hash_name": "2band_32000",
"model_params": "lib/lib_v5/modelparams/2band_32000.json",
"param_name": "2band_32000"
},
{
"hash_name": "2band_48000",
"model_params": "lib/lib_v5/modelparams/2band_48000.json",
"param_name": "2band_48000"
}
],
"3 Band": [
{
"hash_name": "3band_44100",
"model_params": "lib/lib_v5/modelparams/3band_44100.json",
"param_name": "3band_44100"
},
{
"hash_name": "3band_44100_mid",
"model_params": "lib/lib_v5/modelparams/3band_44100_mid.json",
"param_name": "3band_44100_mid"
},
{
"hash_name": "3band_44100_msb2",
"model_params": "lib/lib_v5/modelparams/3band_44100_msb2.json",
"param_name": "3band_44100_msb2"
}
],
"4 Band": [
{
"hash_name": "4band_44100",
"model_params": "lib/lib_v5/modelparams/4band_44100.json",
"param_name": "4band_44100"
},
{
"hash_name": "4band_44100_mid",
"model_params": "lib/lib_v5/modelparams/4band_44100_mid.json",
"param_name": "4band_44100_mid"
},
{
"hash_name": "4band_44100_msb",
"model_params": "lib/lib_v5/modelparams/4band_44100_msb.json",
"param_name": "4band_44100_msb"
},
{
"hash_name": "4band_44100_msb2",
"model_params": "lib/lib_v5/modelparams/4band_44100_msb2.json",
"param_name": "4band_44100_msb2"
},
{
"hash_name": "4band_44100_reverse",
"model_params": "lib/lib_v5/modelparams/4band_44100_reverse.json",
"param_name": "4band_44100_reverse"
},
{
"hash_name": "4band_44100_sw",
"model_params": "lib/lib_v5/modelparams/4band_44100_sw.json",
"param_name": "4band_44100_sw"
},
{
"hash_name": "4band_v2",
"model_params": "lib/lib_v5/modelparams/4band_v2.json",
"param_name": "4band_v2"
},
{
"hash_name": "4band_v2_sn",
"model_params": "lib/lib_v5/modelparams/4band_v2_sn.json",
"param_name": "4band_v2_sn"
},
{
"hash_name": "tmodelparam",
"model_params": "lib/lib_v5/modelparams/tmodelparam.json",
"param_name": "User Model Param Set"
}
]
}
]
}

121
tools/uvr5/lib/utils.py Normal file
View File

@ -0,0 +1,121 @@
import json
import numpy as np
import torch
from tqdm import tqdm
def load_data(file_name: str = "./lib/name_params.json") -> dict:
with open(file_name, "r") as f:
data = json.load(f)
return data
def make_padding(width, cropsize, offset):
left = offset
roi_size = cropsize - left * 2
if roi_size == 0:
roi_size = cropsize
right = roi_size - (width % roi_size) + left
return left, right, roi_size
def inference(X_spec, device, model, aggressiveness, data):
"""
data ： dic configs
"""
def _execute(
X_mag_pad, roi_size, n_window, device, model, aggressiveness, is_half=True
):
model.eval()
with torch.no_grad():
preds = []
iterations = [n_window]
total_iterations = sum(iterations)
for i in tqdm(range(n_window)):
start = i * roi_size
X_mag_window = X_mag_pad[
None, :, :, start : start + data["window_size"]
]
X_mag_window = torch.from_numpy(X_mag_window)
if is_half:
X_mag_window = X_mag_window.half()
X_mag_window = X_mag_window.to(device)
pred = model.predict(X_mag_window, aggressiveness)
pred = pred.detach().cpu().numpy()
preds.append(pred[0])
pred = np.concatenate(preds, axis=2)
return pred
def preprocess(X_spec):
X_mag = np.abs(X_spec)
X_phase = np.angle(X_spec)
return X_mag, X_phase
X_mag, X_phase = preprocess(X_spec)
coef = X_mag.max()
X_mag_pre = X_mag / coef
n_frame = X_mag_pre.shape[2]
pad_l, pad_r, roi_size = make_padding(n_frame, data["window_size"], model.offset)
n_window = int(np.ceil(n_frame / roi_size))
X_mag_pad = np.pad(X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode="constant")
if list(model.state_dict().values())[0].dtype == torch.float16:
is_half = True
else:
is_half = False
pred = _execute(
X_mag_pad, roi_size, n_window, device, model, aggressiveness, is_half
)
pred = pred[:, :, :n_frame]
if data["tta"]:
pad_l += roi_size // 2
pad_r += roi_size // 2
n_window += 1
X_mag_pad = np.pad(X_mag_pre, ((0, 0), (0, 0), (pad_l, pad_r)), mode="constant")
pred_tta = _execute(
X_mag_pad, roi_size, n_window, device, model, aggressiveness, is_half
)
pred_tta = pred_tta[:, :, roi_size // 2 :]
pred_tta = pred_tta[:, :, :n_frame]
return (pred + pred_tta) * 0.5 * coef, X_mag, np.exp(1.0j * X_phase)
else:
return pred * coef, X_mag, np.exp(1.0j * X_phase)
def _get_name_params(model_path, model_hash):
data = load_data()
flag = False
ModelName = model_path
for type in list(data):
for model in list(data[type][0]):
for i in range(len(data[type][0][model])):
if str(data[type][0][model][i]["hash_name"]) == model_hash:
flag = True
elif str(data[type][0][model][i]["hash_name"]) in ModelName:
flag = True
if flag:
model_params_auto = data[type][0][model][i]["model_params"]
param_name_auto = data[type][0][model][i]["param_name"]
if type == "equivalent":
return param_name_auto, model_params_auto
else:
flag = False
return param_name_auto, model_params_auto

256
tools/uvr5/mdxnet.py Normal file
View File

@ -0,0 +1,256 @@
import os
import logging
logger = logging.getLogger(__name__)
import librosa
import numpy as np
import soundfile as sf
import torch
from tqdm import tqdm
cpu = torch.device("cpu")
class ConvTDFNetTrim:
def __init__(
self, device, model_name, target_name, L, dim_f, dim_t, n_fft, hop=1024
):
super(ConvTDFNetTrim, self).__init__()
self.dim_f = dim_f
self.dim_t = 2**dim_t
self.n_fft = n_fft
self.hop = hop
self.n_bins = self.n_fft // 2 + 1
self.chunk_size = hop * (self.dim_t - 1)
self.window = torch.hann_window(window_length=self.n_fft, periodic=True).to(
device
)
self.target_name = target_name
self.blender = "blender" in model_name
self.dim_c = 4
out_c = self.dim_c * 4 if target_name == "*" else self.dim_c
self.freq_pad = torch.zeros(
[1, out_c, self.n_bins - self.dim_f, self.dim_t]
).to(device)
self.n = L // 2
def stft(self, x):
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.view_as_real(x)
x = x.permute([0, 3, 1, 2])
x = x.reshape([-1, 2, 2, self.n_bins, self.dim_t]).reshape(
[-1, self.dim_c, self.n_bins, self.dim_t]
)
return x[:, :, : self.dim_f]
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
)
x = torch.cat([x, freq_pad], -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.permute([0, 2, 3, 1])
x = x.contiguous()
x = torch.view_as_complex(x)
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])
def get_models(device, dim_f, dim_t, n_fft):
return ConvTDFNetTrim(
device=device,
model_name="Conv-TDF",
target_name="vocals",
L=11,
dim_f=dim_f,
dim_t=dim_t,
n_fft=n_fft,
)
class Predictor:
def __init__(self, args):
import onnxruntime as ort
logger.info(ort.get_available_providers())
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 = ort.InferenceSession(
os.path.join(args.onnx, self.model_.target_name + ".onnx"),
providers=[
"CUDAExecutionProvider",
"DmlExecutionProvider",
"CPUExecutionProvider",
],
)
logger.info("ONNX load done")
def demix(self, mix):
samples = mix.shape[-1]
margin = self.args.margin
chunk_size = self.args.chunks * 44100
assert not margin == 0, "margin cannot be zero!"
if margin > chunk_size:
margin = chunk_size
segmented_mix = {}
if self.args.chunks == 0 or samples < chunk_size:
chunk_size = samples
counter = -1
for skip in range(0, samples, chunk_size):
counter += 1
s_margin = 0 if counter == 0 else margin
end = min(skip + chunk_size + margin, samples)
start = skip - s_margin
segmented_mix[skip] = mix[:, start:end].copy()
if end == samples:
break
sources = self.demix_base(segmented_mix, margin_size=margin)
"""
mix:(2,big_sample)
segmented_mix:offset->(2,small_sample)
sources:(1,2,big_sample)
"""
return sources
def demix_base(self, mixes, margin_size):
chunked_sources = []
progress_bar = tqdm(total=len(mixes))
progress_bar.set_description("Processing")
for mix in mixes:
cmix = mixes[mix]
sources = []
n_sample = cmix.shape[1]
model = self.model_
trim = model.n_fft // 2
gen_size = model.chunk_size - 2 * trim
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_waves = []
i = 0
while i < n_sample + pad:
waves = np.array(mix_p[:, i : i + model.chunk_size])
mix_waves.append(waves)
i += gen_size
mix_waves = torch.tensor(mix_waves, dtype=torch.float32).to(cpu)
with torch.no_grad():
_ort = self.model
spek = model.stft(mix_waves)
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
)
tar_waves = model.istft(torch.tensor(spec_pred))
else:
tar_waves = model.istft(
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
end = None if mix == list(mixes.keys())[::-1][0] else -margin_size
if margin_size == 0:
end = None
sources.append(tar_signal[:, start:end])
progress_bar.update(1)
chunked_sources.append(sources)
_sources = np.concatenate(chunked_sources, axis=-1)
# del self.model
progress_bar.close()
return _sources
def prediction(self, m, vocal_root, others_root, format):
os.makedirs(vocal_root, exist_ok=True)
os.makedirs(others_root, exist_ok=True)
basename = os.path.basename(m)
mix, rate = librosa.load(m, mono=False, sr=44100)
if mix.ndim == 1:
mix = np.asfortranarray([mix, mix])
mix = mix.T
sources = self.demix(mix.T)
opt = sources[0].T
if format in ["wav", "flac"]:
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)
else:
path_vocal = "%s/%s_main_vocal.wav" % (vocal_root, basename)
path_other = "%s/%s_others.wav" % (others_root, basename)
sf.write(path_vocal, mix - opt, rate)
sf.write(path_other, opt, rate)
opt_path_vocal = path_vocal[:-4] + ".%s" % format
opt_path_other = path_other[:-4] + ".%s" % format
if os.path.exists(path_vocal):
os.system(
"ffmpeg -i %s -vn %s -q:a 2 -y" % (path_vocal, opt_path_vocal)
)
if os.path.exists(opt_path_vocal):
try:
os.remove(path_vocal)
except:
pass
if os.path.exists(path_other):
os.system(
"ffmpeg -i %s -vn %s -q:a 2 -y" % (path_other, opt_path_other)
)
if os.path.exists(opt_path_other):
try:
os.remove(path_other)
except:
pass
class MDXNetDereverb:
def __init__(self, chunks, device):
self.onnx = "%s/uvr5_weights/onnx_dereverb_By_FoxJoy"%os.path.dirname(os.path.abspath(__file__))
self.shifts = 10 # 'Predict with randomised equivariant stabilisation'
self.mixing = "min_mag" # ['default','min_mag','max_mag']
self.chunks = chunks
self.margin = 44100
self.dim_t = 9
self.dim_f = 3072
self.n_fft = 6144
self.denoise = True
self.pred = Predictor(self)
self.device = device
def _path_audio_(self, input, vocal_root, others_root, format, is_hp3=False):
self.pred.prediction(input, vocal_root, others_root, format)

367
tools/uvr5/vr.py Normal file
View File

@ -0,0 +1,367 @@
import os,sys
parent_directory = os.path.dirname(os.path.abspath(__file__))
import logging,pdb
logger = logging.getLogger(__name__)
import librosa
import numpy as np
import soundfile as sf
import torch
from lib.lib_v5 import nets_61968KB as Nets
from lib.lib_v5 import spec_utils
from lib.lib_v5.model_param_init import ModelParameters
from lib.lib_v5.nets_new import CascadedNet
from lib.utils import inference
class AudioPre:
def __init__(self, agg, model_path, device, is_half, tta=False):
self.model_path = model_path
self.device = device
self.data = {
# Processing Options
"postprocess": False,
"tta": tta,
# Constants
"window_size": 512,
"agg": agg,
"high_end_process": "mirroring",
}
mp = ModelParameters("%s/lib/lib_v5/modelparams/4band_v2.json"%parent_directory)
model = Nets.CascadedASPPNet(mp.param["bins"] * 2)
cpk = torch.load(model_path, map_location="cpu")
model.load_state_dict(cpk)
model.eval()
if is_half:
model = model.half().to(device)
else:
model = model.to(device)
self.mp = mp
self.model = model
def _path_audio_(
self, music_file, ins_root=None, vocal_root=None, format="flac", is_hp3=False
):
if ins_root is None and vocal_root is None:
return "No save root."
name = os.path.basename(music_file)
if ins_root is not None:
os.makedirs(ins_root, exist_ok=True)
if vocal_root is not None:
os.makedirs(vocal_root, exist_ok=True)
X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {}
bands_n = len(self.mp.param["band"])
# print(bands_n)
for d in range(bands_n, 0, -1):
bp = self.mp.param["band"][d]
if d == bands_n: # high-end band
(
X_wave[d],
_,
) = librosa.core.load( # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑
music_file,
bp["sr"],
False,
dtype=np.float32,
res_type=bp["res_type"],
)
if X_wave[d].ndim == 1:
X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]])
else: # lower bands
X_wave[d] = librosa.core.resample(
X_wave[d + 1],
self.mp.param["band"][d + 1]["sr"],
bp["sr"],
res_type=bp["res_type"],
)
# Stft of wave source
X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(
X_wave[d],
bp["hl"],
bp["n_fft"],
self.mp.param["mid_side"],
self.mp.param["mid_side_b2"],
self.mp.param["reverse"],
)
# pdb.set_trace()
if d == bands_n and self.data["high_end_process"] != "none":
input_high_end_h = (bp["n_fft"] // 2 - bp["crop_stop"]) + (
self.mp.param["pre_filter_stop"] - self.mp.param["pre_filter_start"]
)
input_high_end = X_spec_s[d][
:, bp["n_fft"] // 2 - input_high_end_h : bp["n_fft"] // 2, :
]
X_spec_m = spec_utils.combine_spectrograms(X_spec_s, self.mp)
aggresive_set = float(self.data["agg"] / 100)
aggressiveness = {
"value": aggresive_set,
"split_bin": self.mp.param["band"][1]["crop_stop"],
}
with torch.no_grad():
pred, X_mag, X_phase = inference(
X_spec_m, self.device, self.model, aggressiveness, self.data
)
# Postprocess
if self.data["postprocess"]:
pred_inv = np.clip(X_mag - pred, 0, np.inf)
pred = spec_utils.mask_silence(pred, pred_inv)
y_spec_m = pred * X_phase
v_spec_m = X_spec_m - y_spec_m
if ins_root is not None:
if self.data["high_end_process"].startswith("mirroring"):
input_high_end_ = spec_utils.mirroring(
self.data["high_end_process"], y_spec_m, input_high_end, self.mp
)
wav_instrument = spec_utils.cmb_spectrogram_to_wave(
y_spec_m, self.mp, input_high_end_h, input_high_end_
)
else:
wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, self.mp)
logger.info("%s instruments done" % name)
if is_hp3 == True:
head = "vocal_"
else:
head = "instrument_"
if format in ["wav", "flac"]:
sf.write(
os.path.join(
ins_root,
head + "{}_{}.{}".format(name, self.data["agg"], format),
),
(np.array(wav_instrument) * 32768).astype("int16"),
self.mp.param["sr"],
) #
else:
path = os.path.join(
ins_root, head + "{}_{}.wav".format(name, self.data["agg"])
)
sf.write(
path,
(np.array(wav_instrument) * 32768).astype("int16"),
self.mp.param["sr"],
)
if os.path.exists(path):
opt_format_path = path[:-4] + ".%s" % format
os.system("ffmpeg -i %s -vn %s -q:a 2 -y" % (path, opt_format_path))
if os.path.exists(opt_format_path):
try:
os.remove(path)
except:
pass
if vocal_root is not None:
if is_hp3 == True:
head = "instrument_"
else:
head = "vocal_"
if self.data["high_end_process"].startswith("mirroring"):
input_high_end_ = spec_utils.mirroring(
self.data["high_end_process"], v_spec_m, input_high_end, self.mp
)
wav_vocals = spec_utils.cmb_spectrogram_to_wave(
v_spec_m, self.mp, input_high_end_h, input_high_end_
)
else:
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, self.mp)
logger.info("%s vocals done" % name)
if format in ["wav", "flac"]:
sf.write(
os.path.join(
vocal_root,
head + "{}_{}.{}".format(name, self.data["agg"], format),
),
(np.array(wav_vocals) * 32768).astype("int16"),
self.mp.param["sr"],
)
else:
path = os.path.join(
vocal_root, head + "{}_{}.wav".format(name, self.data["agg"])
)
sf.write(
path,
(np.array(wav_vocals) * 32768).astype("int16"),
self.mp.param["sr"],
)
if os.path.exists(path):
opt_format_path = path[:-4] + ".%s" % format
os.system("ffmpeg -i %s -vn %s -q:a 2 -y" % (path, opt_format_path))
if os.path.exists(opt_format_path):
try:
os.remove(path)
except:
pass
class AudioPreDeEcho:
def __init__(self, agg, model_path, device, is_half, tta=False):
self.model_path = model_path
self.device = device
self.data = {
# Processing Options
"postprocess": False,
"tta": tta,
# Constants
"window_size": 512,
"agg": agg,
"high_end_process": "mirroring",
}
mp = ModelParameters("%s/lib/lib_v5/modelparams/4band_v3.json"%parent_directory)
nout = 64 if "DeReverb" in model_path else 48
model = CascadedNet(mp.param["bins"] * 2, nout)
cpk = torch.load(model_path, map_location="cpu")
model.load_state_dict(cpk)
model.eval()
if is_half:
model = model.half().to(device)
else:
model = model.to(device)
self.mp = mp
self.model = model
def _path_audio_(
self, music_file, vocal_root=None, ins_root=None, format="flac", is_hp3=False
): # 3个VR模型vocal和ins是反的
if ins_root is None and vocal_root is None:
return "No save root."
name = os.path.basename(music_file)
if ins_root is not None:
os.makedirs(ins_root, exist_ok=True)
if vocal_root is not None:
os.makedirs(vocal_root, exist_ok=True)
X_wave, y_wave, X_spec_s, y_spec_s = {}, {}, {}, {}
bands_n = len(self.mp.param["band"])
# print(bands_n)
for d in range(bands_n, 0, -1):
bp = self.mp.param["band"][d]
if d == bands_n: # high-end band
(
X_wave[d],
_,
) = librosa.core.load( # 理论上librosa读取可能对某些音频有bug，应该上ffmpeg读取，但是太麻烦了弃坑
music_file,
bp["sr"],
False,
dtype=np.float32,
res_type=bp["res_type"],
)
if X_wave[d].ndim == 1:
X_wave[d] = np.asfortranarray([X_wave[d], X_wave[d]])
else: # lower bands
X_wave[d] = librosa.core.resample(
X_wave[d + 1],
self.mp.param["band"][d + 1]["sr"],
bp["sr"],
res_type=bp["res_type"],
)
# Stft of wave source
X_spec_s[d] = spec_utils.wave_to_spectrogram_mt(
X_wave[d],
bp["hl"],
bp["n_fft"],
self.mp.param["mid_side"],
self.mp.param["mid_side_b2"],
self.mp.param["reverse"],
)
# pdb.set_trace()
if d == bands_n and self.data["high_end_process"] != "none":
input_high_end_h = (bp["n_fft"] // 2 - bp["crop_stop"]) + (
self.mp.param["pre_filter_stop"] - self.mp.param["pre_filter_start"]
)
input_high_end = X_spec_s[d][
:, bp["n_fft"] // 2 - input_high_end_h : bp["n_fft"] // 2, :
]
X_spec_m = spec_utils.combine_spectrograms(X_spec_s, self.mp)
aggresive_set = float(self.data["agg"] / 100)
aggressiveness = {
"value": aggresive_set,
"split_bin": self.mp.param["band"][1]["crop_stop"],
}
with torch.no_grad():
pred, X_mag, X_phase = inference(
X_spec_m, self.device, self.model, aggressiveness, self.data
)
# Postprocess
if self.data["postprocess"]:
pred_inv = np.clip(X_mag - pred, 0, np.inf)
pred = spec_utils.mask_silence(pred, pred_inv)
y_spec_m = pred * X_phase
v_spec_m = X_spec_m - y_spec_m
if ins_root is not None:
if self.data["high_end_process"].startswith("mirroring"):
input_high_end_ = spec_utils.mirroring(
self.data["high_end_process"], y_spec_m, input_high_end, self.mp
)
wav_instrument = spec_utils.cmb_spectrogram_to_wave(
y_spec_m, self.mp, input_high_end_h, input_high_end_
)
else:
wav_instrument = spec_utils.cmb_spectrogram_to_wave(y_spec_m, self.mp)
logger.info("%s instruments done" % name)
if format in ["wav", "flac"]:
sf.write(
os.path.join(
ins_root,
"vocal_{}_{}.{}".format(name, self.data["agg"], format),
),
(np.array(wav_instrument) * 32768).astype("int16"),
self.mp.param["sr"],
) #
else:
path = os.path.join(
ins_root, "vocal_{}_{}.wav".format(name, self.data["agg"])
)
sf.write(
path,
(np.array(wav_instrument) * 32768).astype("int16"),
self.mp.param["sr"],
)
if os.path.exists(path):
opt_format_path = path[:-4] + ".%s" % format
os.system("ffmpeg -i %s -vn %s -q:a 2 -y" % (path, opt_format_path))
if os.path.exists(opt_format_path):
try:
os.remove(path)
except:
pass
if vocal_root is not None:
if self.data["high_end_process"].startswith("mirroring"):
input_high_end_ = spec_utils.mirroring(
self.data["high_end_process"], v_spec_m, input_high_end, self.mp
)
wav_vocals = spec_utils.cmb_spectrogram_to_wave(
v_spec_m, self.mp, input_high_end_h, input_high_end_
)
else:
wav_vocals = spec_utils.cmb_spectrogram_to_wave(v_spec_m, self.mp)
logger.info("%s vocals done" % name)
if format in ["wav", "flac"]:
sf.write(
os.path.join(
vocal_root,
"instrument_{}_{}.{}".format(name, self.data["agg"], format),
),
(np.array(wav_vocals) * 32768).astype("int16"),
self.mp.param["sr"],
)
else:
path = os.path.join(
vocal_root, "instrument_{}_{}.wav".format(name, self.data["agg"])
)
sf.write(
path,
(np.array(wav_vocals) * 32768).astype("int16"),
self.mp.param["sr"],
)
if os.path.exists(path):
opt_format_path = path[:-4] + ".%s" % format
os.system("ffmpeg -i %s -vn %s -q:a 2 -y" % (path, opt_format_path))
if os.path.exists(opt_format_path):
try:
os.remove(path)
except:
pass

184
tools/uvr5/webui.py Normal file
View File

@ -0,0 +1,184 @@
import os
import traceback,gradio as gr
import logging
from i18n.i18n import I18nAuto
i18n = I18nAuto()
logger = logging.getLogger(__name__)
import ffmpeg
import torch
import sys
from mdxnet import MDXNetDereverb
from vr import AudioPre, AudioPreDeEcho
weight_uvr5_root = "tools/uvr5/uvr5_weights"
uvr5_names = []
for name in os.listdir(weight_uvr5_root):
if name.endswith(".pth") or "onnx" in name:
uvr5_names.append(name.replace(".pth", ""))
device=sys.argv[1]
is_half=sys.argv[2]
def uvr(model_name, inp_root, save_root_vocal, paths, save_root_ins, agg, format0):
infos = []
try:
inp_root = inp_root.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
save_root_vocal = (
save_root_vocal.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
)
save_root_ins = (
save_root_ins.strip(" ").strip('"').strip("\n").strip('"').strip(" ")
)
if model_name == "onnx_dereverb_By_FoxJoy":
pre_fun = MDXNetDereverb(15, device)
else:
func = AudioPre if "DeEcho" not in model_name else AudioPreDeEcho
pre_fun = func(
agg=int(agg),
model_path=os.path.join(
weight_uvr5_root, model_name + ".pth"
),
device=device,
is_half=is_half,
)
is_hp3 = "HP3" in model_name
if inp_root != "":
paths = [os.path.join(inp_root, name) for name in os.listdir(inp_root)]
else:
paths = [path.name for path in paths]
for path in paths:
inp_path = os.path.join(inp_root, path)
need_reformat = 1
done = 0
try:
info = ffmpeg.probe(inp_path, cmd="ffprobe")
if (
info["streams"][0]["channels"] == 2
and info["streams"][0]["sample_rate"] == "44100"
):
need_reformat = 0
pre_fun._path_audio_(
inp_path, save_root_ins, save_root_vocal, format0, is_hp3=is_hp3
)
done = 1
except:
need_reformat = 1
traceback.print_exc()
if need_reformat == 1:
tmp_path = "%s/%s.reformatted.wav" % (
os.path.join(os.environ["TEMP"]),
os.path.basename(inp_path),
)
os.system(
"ffmpeg -i %s -vn -acodec pcm_s16le -ac 2 -ar 44100 %s -y"
% (inp_path, tmp_path)
)
inp_path = tmp_path
try:
if done == 0:
pre_fun._path_audio_(
inp_path, save_root_ins, save_root_vocal, format0
)
infos.append("%s->Success" % (os.path.basename(inp_path)))
yield "\n".join(infos)
except:
try:
if done == 0:
pre_fun._path_audio_(
inp_path, save_root_ins, save_root_vocal, format0
)
infos.append("%s->Success" % (os.path.basename(inp_path)))
yield "\n".join(infos)
except:
infos.append(
"%s->%s" % (os.path.basename(inp_path), traceback.format_exc())
)
yield "\n".join(infos)
except:
infos.append(traceback.format_exc())
yield "\n".join(infos)
finally:
try:
if model_name == "onnx_dereverb_By_FoxJoy":
del pre_fun.pred.model
del pre_fun.pred.model_
else:
del pre_fun.model
del pre_fun
except:
traceback.print_exc()
if torch.cuda.is_available():
torch.cuda.empty_cache()
logger.info("Executed torch.cuda.empty_cache()")
yield "\n".join(infos)
with gr.Blocks(title="RVC WebUI") as app:
gr.Markdown(
value=
"本软件以MIT协议开源, 作者不对软件具备任何控制力, 使用软件者、传播软件导出的声音者自负全责. <br>如不认可该条款, 则不能使用或引用软件包内任何代码和文件. 详见根目录<b>LICENSE</b>."
)
with gr.Tabs():
with gr.TabItem(i18n("伴奏人声分离&去混响&去回声")):
with gr.Group():
gr.Markdown(
value=i18n(
"人声伴奏分离批量处理， 使用UVR5模型。 <br>合格的文件夹路径格式举例： E:\\codes\\py39\\vits_vc_gpu\\白鹭霜华测试样例(去文件管理器地址栏拷就行了)。 <br>模型分为三类： <br>1、保留人声：不带和声的音频选这个，对主人声保留比HP5更好。内置HP2和HP3两个模型，HP3可能轻微漏伴奏但对主人声保留比HP2稍微好一丁点； <br>2、仅保留主人声：带和声的音频选这个，对主人声可能有削弱。内置HP5一个模型； <br> 3、去混响、去延迟模型（by FoxJoy）：<br>  (1)MDX-Net(onnx_dereverb):对于双通道混响是最好的选择，不能去除单通道混响；<br>&emsp;(234)DeEcho:去除延迟效果。Aggressive比Normal去除得更彻底，DeReverb额外去除混响，可去除单声道混响，但是对高频重的板式混响去不干净。<br>去混响/去延迟，附：<br>1、DeEcho-DeReverb模型的耗时是另外2个DeEcho模型的接近2倍；<br>2、MDX-Net-Dereverb模型挺慢的；<br>3、个人推荐的最干净的配置是先MDX-Net再DeEcho-Aggressive。"
)
)
with gr.Row():
with gr.Column():
dir_wav_input = gr.Textbox(
label=i18n("输入待处理音频文件夹路径"),
placeholder="C:\\Users\\Desktop\\todo-songs",
)
wav_inputs = gr.File(
file_count="multiple", label=i18n("也可批量输入音频文件, 二选一, 优先读文件夹")
)
with gr.Column():
model_choose = gr.Dropdown(label=i18n("模型"), choices=uvr5_names)
agg = gr.Slider(
minimum=0,
maximum=20,
step=1,
label="人声提取激进程度",
value=10,
interactive=True,
visible=False, # 先不开放调整
)
opt_vocal_root = gr.Textbox(
label=i18n("指定输出主人声文件夹"), value="output/uvr5_opt"
)
opt_ins_root = gr.Textbox(
label=i18n("指定输出非主人声文件夹"), value="output/uvr5_opt"
)
format0 = gr.Radio(
label=i18n("导出文件格式"),
choices=["wav", "flac", "mp3", "m4a"],
value="flac",
interactive=True,
)
but2 = gr.Button(i18n("转换"), variant="primary")
vc_output4 = gr.Textbox(label=i18n("输出信息"))
but2.click(
uvr,
[
model_choose,
dir_wav_input,
opt_vocal_root,
wav_inputs,
opt_ins_root,
agg,
format0,
],
[vc_output4],
api_name="uvr_convert",
)
app.queue(concurrency_count=511, max_size=1022).launch(
server_name="0.0.0.0",
inbrowser=True,
server_port=9873,
quiet=True,
)