143 lines
4.2 KiB
Python
143 lines
4.2 KiB
Python
from torch import nn
|
|
import torch.nn.functional as F
|
|
import math
|
|
|
|
from enhancer.utils.io import Audio as audio
|
|
|
|
class DeLSTM(nn.Module):
|
|
def __init__(
|
|
self,
|
|
input_size:int,
|
|
hidden_size:int,
|
|
num_layers:int,
|
|
bidirectional:bool=True
|
|
|
|
):
|
|
super().__init__()
|
|
self.lstm = nn.LSTM(input_size, hidden_size, num_layers, bidirectional=bidirectional)
|
|
dim = 2 if bidirectional else 1
|
|
self.linear = nn.Linear(dim*hidden_size,hidden_size)
|
|
|
|
def forward(self,x):
|
|
|
|
output,(h,c) = self.lstm(x)
|
|
output = self.linear(output)
|
|
|
|
return output,(h,c)
|
|
|
|
class Demucs(nn.Module):
|
|
|
|
def __init__(
|
|
self,
|
|
c_in:int=1,
|
|
c_out:int=1,
|
|
hidden:int=48,
|
|
kernel_size:int=8,
|
|
stride:int=4,
|
|
growth_factor:int=2,
|
|
depth:int = 5,
|
|
glu:bool = True,
|
|
bidirectional:bool=True,
|
|
resample:int=4,
|
|
sampling_rate = 16000
|
|
|
|
):
|
|
super().__init__()
|
|
self.c_in = c_in
|
|
self.c_out = c_out
|
|
self.hidden = hidden
|
|
self.growth_factor = growth_factor
|
|
self.stride = stride
|
|
self.kernel_size = kernel_size
|
|
self.depth = depth
|
|
self.bidirectional = bidirectional
|
|
self.activation = nn.GLU(1) if glu else nn.ReLU()
|
|
self.resample = resample
|
|
self.sampling_rate = sampling_rate
|
|
multi_factor = 2 if glu else 1
|
|
|
|
self.encoder = nn.ModuleList()
|
|
self.decoder = nn.ModuleList()
|
|
|
|
for layer in range(self.depth):
|
|
|
|
encoder_layer = [nn.Conv1d(c_in,hidden,kernel_size,stride),
|
|
nn.ReLU(),
|
|
nn.Conv1d(hidden, hidden*multi_factor,kernel_size,1),
|
|
self.activation]
|
|
encoder_layer = nn.Sequential(*encoder_layer)
|
|
self.encoder.append(encoder_layer)
|
|
|
|
decoder_layer = [nn.Conv1d(hidden,hidden*multi_factor,kernel_size,1),
|
|
self.activation,
|
|
nn.ConvTranspose1d(hidden,c_out,kernel_size,stride)
|
|
]
|
|
if layer>0:
|
|
decoder_layer.append(nn.ReLU())
|
|
decoder_layer = nn.Sequential(*decoder_layer)
|
|
self.decoder.insert(0,decoder_layer)
|
|
|
|
c_out = hidden
|
|
c_in = hidden
|
|
hidden = self.growth_factor * hidden
|
|
|
|
|
|
self.de_lstm = DeLSTM(input_size=c_in,hidden_size=c_in,num_layers=2,bidirectional=self.bidirectional)
|
|
|
|
def forward(self,mixed_signal):
|
|
|
|
if mixed_signal.dim() == 2:
|
|
mixed_signal = mixed_signal.unsqueeze(1)
|
|
|
|
length = mixed_signal.shape[-1]
|
|
x = F.pad(mixed_signal, (0,self.get_padding_length(length) - length))
|
|
if self.resample>1:
|
|
x = audio.pt_resample_audio(audio=x, sr=self.sampling_rate,
|
|
target_sr=int(self.sampling_rate * self.resample))
|
|
|
|
encoder_outputs = []
|
|
for encoder in self.encoder:
|
|
x = encoder(x)
|
|
print(x.shape)
|
|
encoder_outputs.append(x)
|
|
x = x.permute(0,2,1)
|
|
x,_ = self.de_lstm(x)
|
|
|
|
x = x.permute(0,2,1)
|
|
for decoder in self.decoder:
|
|
skip_connection = encoder_outputs.pop(-1)
|
|
x += skip_connection[..., :x.shape[-1]]
|
|
x = decoder(x)
|
|
|
|
if self.resample > 1:
|
|
x = audio.pt_resample_audio(x,int(self.sampling_rate * self.resample),
|
|
self.sampling_rate)
|
|
|
|
return x
|
|
|
|
def get_padding_length(self,input_length):
|
|
|
|
input_length = math.ceil(input_length * self.resample)
|
|
|
|
|
|
for layer in range(self.depth): # encoder operation
|
|
input_length = math.ceil((input_length - self.kernel_size)/self.stride)+1
|
|
input_length = max(1,input_length)
|
|
for layer in range(self.depth): # decoder operaration
|
|
input_length = (input_length-1) * self.stride + self.kernel_size
|
|
input_length = math.ceil(input_length/self.resample)
|
|
|
|
return int(input_length)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|