2023-10-06 11:14:33 +02:00
|
|
|
|
from io import BytesIO
|
|
|
|
|
import os
|
|
|
|
|
from typing import List, Optional, Tuple
|
2023-08-19 13:00:56 +02:00
|
|
|
|
import numpy as np
|
2023-08-28 09:08:31 +02:00
|
|
|
|
import torch
|
2023-09-14 02:34:30 +02:00
|
|
|
|
|
2023-10-06 11:14:33 +02:00
|
|
|
|
from infer.lib import jit
|
|
|
|
|
|
2023-09-09 06:00:29 +02:00
|
|
|
|
try:
|
2023-09-14 02:34:30 +02:00
|
|
|
|
# Fix "Torch not compiled with CUDA enabled"
|
|
|
|
|
import intel_extension_for_pytorch as ipex # pylint: disable=import-error, unused-import
|
|
|
|
|
|
2023-09-09 06:00:29 +02:00
|
|
|
|
if torch.xpu.is_available():
|
|
|
|
|
from infer.modules.ipex import ipex_init
|
2023-09-14 02:34:30 +02:00
|
|
|
|
|
2023-09-09 06:00:29 +02:00
|
|
|
|
ipex_init()
|
2023-10-06 11:14:33 +02:00
|
|
|
|
except Exception: # pylint: disable=broad-exception-caught
|
2023-09-09 06:00:29 +02:00
|
|
|
|
pass
|
2023-08-28 09:08:31 +02:00
|
|
|
|
import torch.nn as nn
|
2023-08-19 13:00:56 +02:00
|
|
|
|
import torch.nn.functional as F
|
2023-08-28 09:08:31 +02:00
|
|
|
|
from librosa.util import normalize, pad_center, tiny
|
2023-08-19 13:00:56 +02:00
|
|
|
|
from scipy.signal import get_window
|
2023-08-19 13:01:49 +02:00
|
|
|
|
|
2023-09-01 09:18:08 +02:00
|
|
|
|
import logging
|
|
|
|
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
2023-08-19 13:01:49 +02:00
|
|
|
|
|
2023-08-19 13:00:56 +02:00
|
|
|
|
class STFT(torch.nn.Module):
|
2023-08-19 13:01:49 +02:00
|
|
|
|
def __init__(
|
|
|
|
|
self, filter_length=1024, hop_length=512, win_length=None, window="hann"
|
|
|
|
|
):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
"""
|
|
|
|
|
This module implements an STFT using 1D convolution and 1D transpose convolutions.
|
|
|
|
|
This is a bit tricky so there are some cases that probably won't work as working
|
|
|
|
|
out the same sizes before and after in all overlap add setups is tough. Right now,
|
|
|
|
|
this code should work with hop lengths that are half the filter length (50% overlap
|
|
|
|
|
between frames).
|
|
|
|
|
|
|
|
|
|
Keyword Arguments:
|
|
|
|
|
filter_length {int} -- Length of filters used (default: {1024})
|
|
|
|
|
hop_length {int} -- Hop length of STFT (restrict to 50% overlap between frames) (default: {512})
|
|
|
|
|
win_length {[type]} -- Length of the window function applied to each frame (if not specified, it
|
|
|
|
|
equals the filter length). (default: {None})
|
|
|
|
|
window {str} -- Type of window to use (options are bartlett, hann, hamming, blackman, blackmanharris)
|
|
|
|
|
(default: {'hann'})
|
|
|
|
|
"""
|
|
|
|
|
super(STFT, self).__init__()
|
|
|
|
|
self.filter_length = filter_length
|
|
|
|
|
self.hop_length = hop_length
|
|
|
|
|
self.win_length = win_length if win_length else filter_length
|
|
|
|
|
self.window = window
|
|
|
|
|
self.forward_transform = None
|
|
|
|
|
self.pad_amount = int(self.filter_length / 2)
|
|
|
|
|
fourier_basis = np.fft.fft(np.eye(self.filter_length))
|
|
|
|
|
|
|
|
|
|
cutoff = int((self.filter_length / 2 + 1))
|
2023-08-19 13:01:49 +02:00
|
|
|
|
fourier_basis = np.vstack(
|
|
|
|
|
[np.real(fourier_basis[:cutoff, :]), np.imag(fourier_basis[:cutoff, :])]
|
|
|
|
|
)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
forward_basis = torch.FloatTensor(fourier_basis)
|
|
|
|
|
inverse_basis = torch.FloatTensor(np.linalg.pinv(fourier_basis))
|
2023-08-19 13:00:56 +02:00
|
|
|
|
|
2023-08-19 13:01:49 +02:00
|
|
|
|
assert filter_length >= self.win_length
|
2023-08-19 13:00:56 +02:00
|
|
|
|
# get window and zero center pad it to filter_length
|
|
|
|
|
fft_window = get_window(window, self.win_length, fftbins=True)
|
|
|
|
|
fft_window = pad_center(fft_window, size=filter_length)
|
|
|
|
|
fft_window = torch.from_numpy(fft_window).float()
|
|
|
|
|
|
|
|
|
|
# window the bases
|
|
|
|
|
forward_basis *= fft_window
|
2023-10-06 11:14:33 +02:00
|
|
|
|
inverse_basis = (inverse_basis.T * fft_window).T
|
2023-08-19 13:00:56 +02:00
|
|
|
|
|
2023-08-19 13:01:49 +02:00
|
|
|
|
self.register_buffer("forward_basis", forward_basis.float())
|
|
|
|
|
self.register_buffer("inverse_basis", inverse_basis.float())
|
2023-10-06 11:14:33 +02:00
|
|
|
|
self.register_buffer("fft_window", fft_window.float())
|
2023-08-19 13:00:56 +02:00
|
|
|
|
|
2023-10-06 11:14:33 +02:00
|
|
|
|
def transform(self, input_data, return_phase=False):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
"""Take input data (audio) to STFT domain.
|
|
|
|
|
|
|
|
|
|
Arguments:
|
|
|
|
|
input_data {tensor} -- Tensor of floats, with shape (num_batch, num_samples)
|
|
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
|
magnitude {tensor} -- Magnitude of STFT with shape (num_batch,
|
|
|
|
|
num_frequencies, num_frames)
|
|
|
|
|
phase {tensor} -- Phase of STFT with shape (num_batch,
|
|
|
|
|
num_frequencies, num_frames)
|
|
|
|
|
"""
|
2023-08-19 13:01:49 +02:00
|
|
|
|
input_data = F.pad(
|
2023-10-06 11:14:33 +02:00
|
|
|
|
input_data,
|
|
|
|
|
(self.pad_amount, self.pad_amount),
|
2023-08-19 13:01:49 +02:00
|
|
|
|
mode="reflect",
|
|
|
|
|
)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
forward_transform = input_data.unfold(
|
|
|
|
|
1, self.filter_length, self.hop_length
|
|
|
|
|
).permute(0, 2, 1)
|
|
|
|
|
forward_transform = torch.matmul(self.forward_basis, forward_transform)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
cutoff = int((self.filter_length / 2) + 1)
|
|
|
|
|
real_part = forward_transform[:, :cutoff, :]
|
|
|
|
|
imag_part = forward_transform[:, cutoff:, :]
|
2023-08-19 13:01:49 +02:00
|
|
|
|
magnitude = torch.sqrt(real_part**2 + imag_part**2)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
if return_phase:
|
|
|
|
|
phase = torch.atan2(imag_part.data, real_part.data)
|
|
|
|
|
return magnitude, phase
|
|
|
|
|
else:
|
|
|
|
|
return magnitude
|
2023-08-19 13:00:56 +02:00
|
|
|
|
|
|
|
|
|
def inverse(self, magnitude, phase):
|
|
|
|
|
"""Call the inverse STFT (iSTFT), given magnitude and phase tensors produced
|
|
|
|
|
by the ```transform``` function.
|
|
|
|
|
|
|
|
|
|
Arguments:
|
|
|
|
|
magnitude {tensor} -- Magnitude of STFT with shape (num_batch,
|
|
|
|
|
num_frequencies, num_frames)
|
|
|
|
|
phase {tensor} -- Phase of STFT with shape (num_batch,
|
|
|
|
|
num_frequencies, num_frames)
|
|
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
|
inverse_transform {tensor} -- Reconstructed audio given magnitude and phase. Of
|
|
|
|
|
shape (num_batch, num_samples)
|
|
|
|
|
"""
|
2023-10-06 11:14:33 +02:00
|
|
|
|
cat = torch.cat(
|
2023-08-19 13:01:49 +02:00
|
|
|
|
[magnitude * torch.cos(phase), magnitude * torch.sin(phase)], dim=1
|
|
|
|
|
)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
fold = torch.nn.Fold(
|
|
|
|
|
output_size=(1, (cat.size(-1) - 1) * self.hop_length + self.filter_length),
|
|
|
|
|
kernel_size=(1, self.filter_length),
|
|
|
|
|
stride=(1, self.hop_length),
|
2023-08-19 13:01:49 +02:00
|
|
|
|
)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
inverse_transform = torch.matmul(self.inverse_basis, cat)
|
|
|
|
|
inverse_transform = fold(inverse_transform)[
|
|
|
|
|
:, 0, 0, self.pad_amount : -self.pad_amount
|
|
|
|
|
]
|
|
|
|
|
window_square_sum = (
|
|
|
|
|
self.fft_window.pow(2).repeat(cat.size(-1), 1).T.unsqueeze(0)
|
|
|
|
|
)
|
|
|
|
|
window_square_sum = fold(window_square_sum)[
|
|
|
|
|
:, 0, 0, self.pad_amount : -self.pad_amount
|
|
|
|
|
]
|
|
|
|
|
inverse_transform /= window_square_sum
|
2023-08-19 13:00:56 +02:00
|
|
|
|
return inverse_transform
|
|
|
|
|
|
|
|
|
|
def forward(self, input_data):
|
|
|
|
|
"""Take input data (audio) to STFT domain and then back to audio.
|
|
|
|
|
|
|
|
|
|
Arguments:
|
|
|
|
|
input_data {tensor} -- Tensor of floats, with shape (num_batch, num_samples)
|
|
|
|
|
|
|
|
|
|
Returns:
|
|
|
|
|
reconstruction {tensor} -- Reconstructed audio given magnitude and phase. Of
|
|
|
|
|
shape (num_batch, num_samples)
|
|
|
|
|
"""
|
2023-10-06 11:14:33 +02:00
|
|
|
|
self.magnitude, self.phase = self.transform(input_data, return_phase=True)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
reconstruction = self.inverse(self.magnitude, self.phase)
|
|
|
|
|
return reconstruction
|
2023-08-19 13:01:49 +02:00
|
|
|
|
|
|
|
|
|
|
2023-08-19 13:00:56 +02:00
|
|
|
|
from time import time as ttime
|
2023-08-19 13:01:49 +02:00
|
|
|
|
|
|
|
|
|
|
2023-08-19 13:00:56 +02:00
|
|
|
|
class BiGRU(nn.Module):
|
|
|
|
|
def __init__(self, input_features, hidden_features, num_layers):
|
|
|
|
|
super(BiGRU, self).__init__()
|
|
|
|
|
self.gru = nn.GRU(
|
|
|
|
|
input_features,
|
|
|
|
|
hidden_features,
|
|
|
|
|
num_layers=num_layers,
|
|
|
|
|
batch_first=True,
|
|
|
|
|
bidirectional=True,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def forward(self, x):
|
|
|
|
|
return self.gru(x)[0]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class ConvBlockRes(nn.Module):
|
|
|
|
|
def __init__(self, in_channels, out_channels, momentum=0.01):
|
|
|
|
|
super(ConvBlockRes, self).__init__()
|
|
|
|
|
self.conv = nn.Sequential(
|
|
|
|
|
nn.Conv2d(
|
|
|
|
|
in_channels=in_channels,
|
|
|
|
|
out_channels=out_channels,
|
|
|
|
|
kernel_size=(3, 3),
|
|
|
|
|
stride=(1, 1),
|
|
|
|
|
padding=(1, 1),
|
|
|
|
|
bias=False,
|
|
|
|
|
),
|
|
|
|
|
nn.BatchNorm2d(out_channels, momentum=momentum),
|
|
|
|
|
nn.ReLU(),
|
|
|
|
|
nn.Conv2d(
|
|
|
|
|
in_channels=out_channels,
|
|
|
|
|
out_channels=out_channels,
|
|
|
|
|
kernel_size=(3, 3),
|
|
|
|
|
stride=(1, 1),
|
|
|
|
|
padding=(1, 1),
|
|
|
|
|
bias=False,
|
|
|
|
|
),
|
|
|
|
|
nn.BatchNorm2d(out_channels, momentum=momentum),
|
|
|
|
|
nn.ReLU(),
|
|
|
|
|
)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
# self.shortcut:Optional[nn.Module] = None
|
2023-08-19 13:00:56 +02:00
|
|
|
|
if in_channels != out_channels:
|
|
|
|
|
self.shortcut = nn.Conv2d(in_channels, out_channels, (1, 1))
|
|
|
|
|
|
2023-10-06 11:14:33 +02:00
|
|
|
|
def forward(self, x: torch.Tensor):
|
|
|
|
|
if not hasattr(self, "shortcut"):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
return self.conv(x) + x
|
2023-10-06 11:14:33 +02:00
|
|
|
|
else:
|
|
|
|
|
return self.conv(x) + self.shortcut(x)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class Encoder(nn.Module):
|
|
|
|
|
def __init__(
|
|
|
|
|
self,
|
|
|
|
|
in_channels,
|
|
|
|
|
in_size,
|
|
|
|
|
n_encoders,
|
|
|
|
|
kernel_size,
|
|
|
|
|
n_blocks,
|
|
|
|
|
out_channels=16,
|
|
|
|
|
momentum=0.01,
|
|
|
|
|
):
|
|
|
|
|
super(Encoder, self).__init__()
|
|
|
|
|
self.n_encoders = n_encoders
|
|
|
|
|
self.bn = nn.BatchNorm2d(in_channels, momentum=momentum)
|
|
|
|
|
self.layers = nn.ModuleList()
|
|
|
|
|
self.latent_channels = []
|
|
|
|
|
for i in range(self.n_encoders):
|
|
|
|
|
self.layers.append(
|
|
|
|
|
ResEncoderBlock(
|
|
|
|
|
in_channels, out_channels, kernel_size, n_blocks, momentum=momentum
|
|
|
|
|
)
|
|
|
|
|
)
|
|
|
|
|
self.latent_channels.append([out_channels, in_size])
|
|
|
|
|
in_channels = out_channels
|
|
|
|
|
out_channels *= 2
|
|
|
|
|
in_size //= 2
|
|
|
|
|
self.out_size = in_size
|
|
|
|
|
self.out_channel = out_channels
|
|
|
|
|
|
2023-10-06 11:14:33 +02:00
|
|
|
|
def forward(self, x: torch.Tensor):
|
|
|
|
|
concat_tensors: List[torch.Tensor] = []
|
2023-08-19 13:00:56 +02:00
|
|
|
|
x = self.bn(x)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
for i, layer in enumerate(self.layers):
|
|
|
|
|
t, x = layer(x)
|
|
|
|
|
concat_tensors.append(t)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
return x, concat_tensors
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class ResEncoderBlock(nn.Module):
|
|
|
|
|
def __init__(
|
|
|
|
|
self, in_channels, out_channels, kernel_size, n_blocks=1, momentum=0.01
|
|
|
|
|
):
|
|
|
|
|
super(ResEncoderBlock, self).__init__()
|
|
|
|
|
self.n_blocks = n_blocks
|
|
|
|
|
self.conv = nn.ModuleList()
|
|
|
|
|
self.conv.append(ConvBlockRes(in_channels, out_channels, momentum))
|
|
|
|
|
for i in range(n_blocks - 1):
|
|
|
|
|
self.conv.append(ConvBlockRes(out_channels, out_channels, momentum))
|
|
|
|
|
self.kernel_size = kernel_size
|
|
|
|
|
if self.kernel_size is not None:
|
|
|
|
|
self.pool = nn.AvgPool2d(kernel_size=kernel_size)
|
|
|
|
|
|
|
|
|
|
def forward(self, x):
|
2023-10-06 11:14:33 +02:00
|
|
|
|
for i, conv in enumerate(self.conv):
|
|
|
|
|
x = conv(x)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
if self.kernel_size is not None:
|
|
|
|
|
return x, self.pool(x)
|
|
|
|
|
else:
|
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class Intermediate(nn.Module): #
|
|
|
|
|
def __init__(self, in_channels, out_channels, n_inters, n_blocks, momentum=0.01):
|
|
|
|
|
super(Intermediate, self).__init__()
|
|
|
|
|
self.n_inters = n_inters
|
|
|
|
|
self.layers = nn.ModuleList()
|
|
|
|
|
self.layers.append(
|
|
|
|
|
ResEncoderBlock(in_channels, out_channels, None, n_blocks, momentum)
|
|
|
|
|
)
|
|
|
|
|
for i in range(self.n_inters - 1):
|
|
|
|
|
self.layers.append(
|
|
|
|
|
ResEncoderBlock(out_channels, out_channels, None, n_blocks, momentum)
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def forward(self, x):
|
2023-10-06 11:14:33 +02:00
|
|
|
|
for i, layer in enumerate(self.layers):
|
|
|
|
|
x = layer(x)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class ResDecoderBlock(nn.Module):
|
|
|
|
|
def __init__(self, in_channels, out_channels, stride, n_blocks=1, momentum=0.01):
|
|
|
|
|
super(ResDecoderBlock, self).__init__()
|
|
|
|
|
out_padding = (0, 1) if stride == (1, 2) else (1, 1)
|
|
|
|
|
self.n_blocks = n_blocks
|
|
|
|
|
self.conv1 = nn.Sequential(
|
|
|
|
|
nn.ConvTranspose2d(
|
|
|
|
|
in_channels=in_channels,
|
|
|
|
|
out_channels=out_channels,
|
|
|
|
|
kernel_size=(3, 3),
|
|
|
|
|
stride=stride,
|
|
|
|
|
padding=(1, 1),
|
|
|
|
|
output_padding=out_padding,
|
|
|
|
|
bias=False,
|
|
|
|
|
),
|
|
|
|
|
nn.BatchNorm2d(out_channels, momentum=momentum),
|
|
|
|
|
nn.ReLU(),
|
|
|
|
|
)
|
|
|
|
|
self.conv2 = nn.ModuleList()
|
|
|
|
|
self.conv2.append(ConvBlockRes(out_channels * 2, out_channels, momentum))
|
|
|
|
|
for i in range(n_blocks - 1):
|
|
|
|
|
self.conv2.append(ConvBlockRes(out_channels, out_channels, momentum))
|
|
|
|
|
|
|
|
|
|
def forward(self, x, concat_tensor):
|
|
|
|
|
x = self.conv1(x)
|
|
|
|
|
x = torch.cat((x, concat_tensor), dim=1)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
for i, conv2 in enumerate(self.conv2):
|
|
|
|
|
x = conv2(x)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class Decoder(nn.Module):
|
|
|
|
|
def __init__(self, in_channels, n_decoders, stride, n_blocks, momentum=0.01):
|
|
|
|
|
super(Decoder, self).__init__()
|
|
|
|
|
self.layers = nn.ModuleList()
|
|
|
|
|
self.n_decoders = n_decoders
|
|
|
|
|
for i in range(self.n_decoders):
|
|
|
|
|
out_channels = in_channels // 2
|
|
|
|
|
self.layers.append(
|
|
|
|
|
ResDecoderBlock(in_channels, out_channels, stride, n_blocks, momentum)
|
|
|
|
|
)
|
|
|
|
|
in_channels = out_channels
|
|
|
|
|
|
2023-10-06 11:14:33 +02:00
|
|
|
|
def forward(self, x: torch.Tensor, concat_tensors: List[torch.Tensor]):
|
|
|
|
|
for i, layer in enumerate(self.layers):
|
|
|
|
|
x = layer(x, concat_tensors[-1 - i])
|
2023-08-19 13:00:56 +02:00
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class DeepUnet(nn.Module):
|
|
|
|
|
def __init__(
|
|
|
|
|
self,
|
|
|
|
|
kernel_size,
|
|
|
|
|
n_blocks,
|
|
|
|
|
en_de_layers=5,
|
|
|
|
|
inter_layers=4,
|
|
|
|
|
in_channels=1,
|
|
|
|
|
en_out_channels=16,
|
|
|
|
|
):
|
|
|
|
|
super(DeepUnet, self).__init__()
|
|
|
|
|
self.encoder = Encoder(
|
|
|
|
|
in_channels, 128, en_de_layers, kernel_size, n_blocks, en_out_channels
|
|
|
|
|
)
|
|
|
|
|
self.intermediate = Intermediate(
|
|
|
|
|
self.encoder.out_channel // 2,
|
|
|
|
|
self.encoder.out_channel,
|
|
|
|
|
inter_layers,
|
|
|
|
|
n_blocks,
|
|
|
|
|
)
|
|
|
|
|
self.decoder = Decoder(
|
|
|
|
|
self.encoder.out_channel, en_de_layers, kernel_size, n_blocks
|
|
|
|
|
)
|
|
|
|
|
|
2023-10-06 11:14:33 +02:00
|
|
|
|
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
2023-08-19 13:00:56 +02:00
|
|
|
|
x, concat_tensors = self.encoder(x)
|
|
|
|
|
x = self.intermediate(x)
|
|
|
|
|
x = self.decoder(x, concat_tensors)
|
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class E2E(nn.Module):
|
|
|
|
|
def __init__(
|
|
|
|
|
self,
|
|
|
|
|
n_blocks,
|
|
|
|
|
n_gru,
|
|
|
|
|
kernel_size,
|
|
|
|
|
en_de_layers=5,
|
|
|
|
|
inter_layers=4,
|
|
|
|
|
in_channels=1,
|
|
|
|
|
en_out_channels=16,
|
|
|
|
|
):
|
|
|
|
|
super(E2E, self).__init__()
|
|
|
|
|
self.unet = DeepUnet(
|
|
|
|
|
kernel_size,
|
|
|
|
|
n_blocks,
|
|
|
|
|
en_de_layers,
|
|
|
|
|
inter_layers,
|
|
|
|
|
in_channels,
|
|
|
|
|
en_out_channels,
|
|
|
|
|
)
|
|
|
|
|
self.cnn = nn.Conv2d(en_out_channels, 3, (3, 3), padding=(1, 1))
|
|
|
|
|
if n_gru:
|
|
|
|
|
self.fc = nn.Sequential(
|
|
|
|
|
BiGRU(3 * 128, 256, n_gru),
|
|
|
|
|
nn.Linear(512, 360),
|
|
|
|
|
nn.Dropout(0.25),
|
|
|
|
|
nn.Sigmoid(),
|
|
|
|
|
)
|
|
|
|
|
else:
|
|
|
|
|
self.fc = nn.Sequential(
|
|
|
|
|
nn.Linear(3 * nn.N_MELS, nn.N_CLASS), nn.Dropout(0.25), nn.Sigmoid()
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
def forward(self, mel):
|
|
|
|
|
# print(mel.shape)
|
|
|
|
|
mel = mel.transpose(-1, -2).unsqueeze(1)
|
|
|
|
|
x = self.cnn(self.unet(mel)).transpose(1, 2).flatten(-2)
|
|
|
|
|
x = self.fc(x)
|
|
|
|
|
# print(x.shape)
|
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
from librosa.filters import mel
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class MelSpectrogram(torch.nn.Module):
|
|
|
|
|
def __init__(
|
|
|
|
|
self,
|
|
|
|
|
is_half,
|
|
|
|
|
n_mel_channels,
|
|
|
|
|
sampling_rate,
|
|
|
|
|
win_length,
|
|
|
|
|
hop_length,
|
|
|
|
|
n_fft=None,
|
|
|
|
|
mel_fmin=0,
|
|
|
|
|
mel_fmax=None,
|
|
|
|
|
clamp=1e-5,
|
|
|
|
|
):
|
|
|
|
|
super().__init__()
|
|
|
|
|
n_fft = win_length if n_fft is None else n_fft
|
|
|
|
|
self.hann_window = {}
|
|
|
|
|
mel_basis = mel(
|
|
|
|
|
sr=sampling_rate,
|
|
|
|
|
n_fft=n_fft,
|
|
|
|
|
n_mels=n_mel_channels,
|
|
|
|
|
fmin=mel_fmin,
|
|
|
|
|
fmax=mel_fmax,
|
|
|
|
|
htk=True,
|
|
|
|
|
)
|
|
|
|
|
mel_basis = torch.from_numpy(mel_basis).float()
|
|
|
|
|
self.register_buffer("mel_basis", mel_basis)
|
|
|
|
|
self.n_fft = win_length if n_fft is None else n_fft
|
|
|
|
|
self.hop_length = hop_length
|
|
|
|
|
self.win_length = win_length
|
|
|
|
|
self.sampling_rate = sampling_rate
|
|
|
|
|
self.n_mel_channels = n_mel_channels
|
|
|
|
|
self.clamp = clamp
|
|
|
|
|
self.is_half = is_half
|
|
|
|
|
|
|
|
|
|
def forward(self, audio, keyshift=0, speed=1, center=True):
|
|
|
|
|
factor = 2 ** (keyshift / 12)
|
|
|
|
|
n_fft_new = int(np.round(self.n_fft * factor))
|
|
|
|
|
win_length_new = int(np.round(self.win_length * factor))
|
|
|
|
|
hop_length_new = int(np.round(self.hop_length * speed))
|
|
|
|
|
keyshift_key = str(keyshift) + "_" + str(audio.device)
|
|
|
|
|
if keyshift_key not in self.hann_window:
|
|
|
|
|
self.hann_window[keyshift_key] = torch.hann_window(win_length_new).to(
|
|
|
|
|
audio.device
|
|
|
|
|
)
|
2023-10-06 11:14:33 +02:00
|
|
|
|
if "privateuseone" in str(audio.device):
|
|
|
|
|
if not hasattr(self, "stft"):
|
|
|
|
|
self.stft = STFT(
|
|
|
|
|
filter_length=n_fft_new,
|
|
|
|
|
hop_length=hop_length_new,
|
|
|
|
|
win_length=win_length_new,
|
|
|
|
|
window="hann",
|
|
|
|
|
).to(audio.device)
|
|
|
|
|
magnitude = self.stft.transform(audio)
|
|
|
|
|
else:
|
|
|
|
|
fft = torch.stft(
|
|
|
|
|
audio,
|
|
|
|
|
n_fft=n_fft_new,
|
2023-08-19 13:00:56 +02:00
|
|
|
|
hop_length=hop_length_new,
|
|
|
|
|
win_length=win_length_new,
|
2023-10-06 11:14:33 +02:00
|
|
|
|
window=self.hann_window[keyshift_key],
|
|
|
|
|
center=center,
|
|
|
|
|
return_complex=True,
|
|
|
|
|
)
|
|
|
|
|
magnitude = torch.sqrt(fft.real.pow(2) + fft.imag.pow(2))
|
2023-08-19 13:00:56 +02:00
|
|
|
|
if keyshift != 0:
|
|
|
|
|
size = self.n_fft // 2 + 1
|
|
|
|
|
resize = magnitude.size(1)
|
|
|
|
|
if resize < size:
|
|
|
|
|
magnitude = F.pad(magnitude, (0, 0, 0, size - resize))
|
|
|
|
|
magnitude = magnitude[:, :size, :] * self.win_length / win_length_new
|
|
|
|
|
mel_output = torch.matmul(self.mel_basis, magnitude)
|
|
|
|
|
if self.is_half == True:
|
|
|
|
|
mel_output = mel_output.half()
|
|
|
|
|
log_mel_spec = torch.log(torch.clamp(mel_output, min=self.clamp))
|
|
|
|
|
return log_mel_spec
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class RMVPE:
|
2023-10-06 11:14:33 +02:00
|
|
|
|
def __init__(self, model_path: str, is_half, device=None, use_jit=False):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
self.resample_kernel = {}
|
|
|
|
|
self.resample_kernel = {}
|
|
|
|
|
self.is_half = is_half
|
|
|
|
|
if device is None:
|
2023-10-06 11:14:33 +02:00
|
|
|
|
device = "cuda:0" if torch.cuda.is_available() else "cpu"
|
2023-08-19 13:00:56 +02:00
|
|
|
|
self.device = device
|
|
|
|
|
self.mel_extractor = MelSpectrogram(
|
|
|
|
|
is_half, 128, 16000, 1024, 160, None, 30, 8000
|
|
|
|
|
).to(device)
|
2023-08-19 13:01:49 +02:00
|
|
|
|
if "privateuseone" in str(device):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
import onnxruntime as ort
|
2023-08-19 13:01:49 +02:00
|
|
|
|
|
|
|
|
|
ort_session = ort.InferenceSession(
|
2023-08-30 13:03:21 +02:00
|
|
|
|
"%s/rmvpe.onnx" % os.environ["rmvpe_root"],
|
|
|
|
|
providers=["DmlExecutionProvider"],
|
2023-08-19 13:01:49 +02:00
|
|
|
|
)
|
|
|
|
|
self.model = ort_session
|
2023-08-19 13:00:56 +02:00
|
|
|
|
else:
|
2023-10-06 11:14:33 +02:00
|
|
|
|
if str(self.device) == "cuda":
|
|
|
|
|
self.device = torch.device("cuda:0")
|
|
|
|
|
|
|
|
|
|
def get_jit_model():
|
|
|
|
|
jit_model_path = model_path.rstrip(".pth")
|
|
|
|
|
jit_model_path += ".half.jit" if is_half else ".jit"
|
|
|
|
|
reload = False
|
|
|
|
|
if os.path.exists(jit_model_path):
|
|
|
|
|
ckpt = jit.load(jit_model_path)
|
|
|
|
|
model_device = ckpt["device"]
|
|
|
|
|
if model_device != str(self.device):
|
|
|
|
|
reload = True
|
|
|
|
|
else:
|
|
|
|
|
reload = True
|
|
|
|
|
|
|
|
|
|
if reload:
|
|
|
|
|
ckpt = jit.rmvpe_jit_export(
|
|
|
|
|
model_path=model_path,
|
|
|
|
|
mode="script",
|
|
|
|
|
inputs_path=None,
|
|
|
|
|
save_path=jit_model_path,
|
|
|
|
|
device=device,
|
|
|
|
|
is_half=is_half,
|
|
|
|
|
)
|
|
|
|
|
model = torch.jit.load(BytesIO(ckpt["model"]), map_location=device)
|
|
|
|
|
return model
|
|
|
|
|
|
|
|
|
|
def get_default_model():
|
|
|
|
|
model = E2E(4, 1, (2, 2))
|
|
|
|
|
ckpt = torch.load(model_path, map_location="cpu")
|
|
|
|
|
model.load_state_dict(ckpt)
|
|
|
|
|
model.eval()
|
|
|
|
|
if is_half:
|
|
|
|
|
model = model.half()
|
|
|
|
|
else:
|
|
|
|
|
model = model.float()
|
|
|
|
|
return model
|
|
|
|
|
|
|
|
|
|
if use_jit:
|
|
|
|
|
if is_half and "cpu" in str(self.device):
|
|
|
|
|
logger.warning(
|
|
|
|
|
"Use default rmvpe model. \
|
|
|
|
|
Jit is not supported on the CPU for half floating point"
|
|
|
|
|
)
|
|
|
|
|
self.model = get_default_model()
|
|
|
|
|
else:
|
|
|
|
|
self.model = get_jit_model()
|
|
|
|
|
else:
|
|
|
|
|
self.model = get_default_model()
|
|
|
|
|
|
2023-08-19 13:00:56 +02:00
|
|
|
|
self.model = self.model.to(device)
|
|
|
|
|
cents_mapping = 20 * np.arange(360) + 1997.3794084376191
|
|
|
|
|
self.cents_mapping = np.pad(cents_mapping, (4, 4)) # 368
|
|
|
|
|
|
|
|
|
|
def mel2hidden(self, mel):
|
|
|
|
|
with torch.no_grad():
|
|
|
|
|
n_frames = mel.shape[-1]
|
2023-10-06 11:14:33 +02:00
|
|
|
|
n_pad = 32 * ((n_frames - 1) // 32 + 1) - n_frames
|
|
|
|
|
if n_pad > 0:
|
|
|
|
|
mel = F.pad(mel, (0, n_pad), mode="constant")
|
2023-08-19 13:01:49 +02:00
|
|
|
|
if "privateuseone" in str(self.device):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
onnx_input_name = self.model.get_inputs()[0].name
|
|
|
|
|
onnx_outputs_names = self.model.get_outputs()[0].name
|
2023-08-19 13:01:49 +02:00
|
|
|
|
hidden = self.model.run(
|
|
|
|
|
[onnx_outputs_names],
|
|
|
|
|
input_feed={onnx_input_name: mel.cpu().numpy()},
|
|
|
|
|
)[0]
|
2023-08-19 13:00:56 +02:00
|
|
|
|
else:
|
2023-10-06 11:14:33 +02:00
|
|
|
|
mel = mel.half() if self.is_half else mel.float()
|
2023-08-19 13:00:56 +02:00
|
|
|
|
hidden = self.model(mel)
|
|
|
|
|
return hidden[:, :n_frames]
|
|
|
|
|
|
|
|
|
|
def decode(self, hidden, thred=0.03):
|
|
|
|
|
cents_pred = self.to_local_average_cents(hidden, thred=thred)
|
|
|
|
|
f0 = 10 * (2 ** (cents_pred / 1200))
|
|
|
|
|
f0[f0 == 10] = 0
|
|
|
|
|
# f0 = np.array([10 * (2 ** (cent_pred / 1200)) if cent_pred else 0 for cent_pred in cents_pred])
|
|
|
|
|
return f0
|
|
|
|
|
|
|
|
|
|
def infer_from_audio(self, audio, thred=0.03):
|
|
|
|
|
# torch.cuda.synchronize()
|
2023-08-19 13:01:49 +02:00
|
|
|
|
t0 = ttime()
|
|
|
|
|
mel = self.mel_extractor(
|
|
|
|
|
torch.from_numpy(audio).float().to(self.device).unsqueeze(0), center=True
|
|
|
|
|
)
|
2023-08-19 13:00:56 +02:00
|
|
|
|
# print(123123123,mel.device.type)
|
|
|
|
|
# torch.cuda.synchronize()
|
2023-08-19 13:01:49 +02:00
|
|
|
|
t1 = ttime()
|
2023-08-19 13:00:56 +02:00
|
|
|
|
hidden = self.mel2hidden(mel)
|
|
|
|
|
# torch.cuda.synchronize()
|
2023-08-19 13:01:49 +02:00
|
|
|
|
t2 = ttime()
|
2023-08-19 13:00:56 +02:00
|
|
|
|
# print(234234,hidden.device.type)
|
2023-08-19 13:01:49 +02:00
|
|
|
|
if "privateuseone" not in str(self.device):
|
2023-08-19 13:00:56 +02:00
|
|
|
|
hidden = hidden.squeeze(0).cpu().numpy()
|
|
|
|
|
else:
|
2023-08-19 13:01:49 +02:00
|
|
|
|
hidden = hidden[0]
|
2023-08-19 13:00:56 +02:00
|
|
|
|
if self.is_half == True:
|
|
|
|
|
hidden = hidden.astype("float32")
|
|
|
|
|
|
|
|
|
|
f0 = self.decode(hidden, thred=thred)
|
|
|
|
|
# torch.cuda.synchronize()
|
2023-08-19 13:01:49 +02:00
|
|
|
|
t3 = ttime()
|
2023-08-19 13:00:56 +02:00
|
|
|
|
# print("hmvpe:%s\t%s\t%s\t%s"%(t1-t0,t2-t1,t3-t2,t3-t0))
|
|
|
|
|
return f0
|
|
|
|
|
|
|
|
|
|
def to_local_average_cents(self, salience, thred=0.05):
|
|
|
|
|
# t0 = ttime()
|
|
|
|
|
center = np.argmax(salience, axis=1) # 帧长#index
|
|
|
|
|
salience = np.pad(salience, ((0, 0), (4, 4))) # 帧长,368
|
|
|
|
|
# t1 = ttime()
|
|
|
|
|
center += 4
|
|
|
|
|
todo_salience = []
|
|
|
|
|
todo_cents_mapping = []
|
|
|
|
|
starts = center - 4
|
|
|
|
|
ends = center + 5
|
|
|
|
|
for idx in range(salience.shape[0]):
|
|
|
|
|
todo_salience.append(salience[:, starts[idx] : ends[idx]][idx])
|
|
|
|
|
todo_cents_mapping.append(self.cents_mapping[starts[idx] : ends[idx]])
|
|
|
|
|
# t2 = ttime()
|
|
|
|
|
todo_salience = np.array(todo_salience) # 帧长,9
|
|
|
|
|
todo_cents_mapping = np.array(todo_cents_mapping) # 帧长,9
|
|
|
|
|
product_sum = np.sum(todo_salience * todo_cents_mapping, 1)
|
|
|
|
|
weight_sum = np.sum(todo_salience, 1) # 帧长
|
|
|
|
|
devided = product_sum / weight_sum # 帧长
|
|
|
|
|
# t3 = ttime()
|
|
|
|
|
maxx = np.max(salience, axis=1) # 帧长
|
|
|
|
|
devided[maxx <= thred] = 0
|
|
|
|
|
# t4 = ttime()
|
|
|
|
|
# print("decode:%s\t%s\t%s\t%s" % (t1 - t0, t2 - t1, t3 - t2, t4 - t3))
|
|
|
|
|
return devided
|
|
|
|
|
|
|
|
|
|
|
2023-08-19 13:01:49 +02:00
|
|
|
|
if __name__ == "__main__":
|
2023-08-28 09:08:31 +02:00
|
|
|
|
import librosa
|
|
|
|
|
import soundfile as sf
|
2023-08-19 13:01:49 +02:00
|
|
|
|
|
2023-08-19 13:00:56 +02:00
|
|
|
|
audio, sampling_rate = sf.read(r"C:\Users\liujing04\Desktop\Z\冬之花clip1.wav")
|
|
|
|
|
if len(audio.shape) > 1:
|
|
|
|
|
audio = librosa.to_mono(audio.transpose(1, 0))
|
|
|
|
|
audio_bak = audio.copy()
|
|
|
|
|
if sampling_rate != 16000:
|
|
|
|
|
audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
|
|
|
|
|
model_path = r"D:\BaiduNetdiskDownload\RVC-beta-v2-0727AMD_realtime\rmvpe.pt"
|
|
|
|
|
thred = 0.03 # 0.01
|
2023-08-19 13:01:49 +02:00
|
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
|
|
rmvpe = RMVPE(model_path, is_half=False, device=device)
|
|
|
|
|
t0 = ttime()
|
2023-08-19 13:00:56 +02:00
|
|
|
|
f0 = rmvpe.infer_from_audio(audio, thred=thred)
|
|
|
|
|
# f0 = rmvpe.infer_from_audio(audio, thred=thred)
|
|
|
|
|
# f0 = rmvpe.infer_from_audio(audio, thred=thred)
|
|
|
|
|
# f0 = rmvpe.infer_from_audio(audio, thred=thred)
|
|
|
|
|
# f0 = rmvpe.infer_from_audio(audio, thred=thred)
|
2023-08-19 13:01:49 +02:00
|
|
|
|
t1 = ttime()
|
2023-09-02 06:09:19 +02:00
|
|
|
|
logger.info("%s %.2f", f0.shape, t1 - t0)
|