You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
211 lines
8.2 KiB
211 lines
8.2 KiB
# Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
"""
|
|
Implementation of model from:
|
|
Kum et al. - "Joint Detection and Classification of Singing Voice Melody Using
|
|
Convolutional Recurrent Neural Networks" (2019)
|
|
Link: https://www.semanticscholar.org/paper/Joint-Detection-and-Classification-of-Singing-Voice-Kum-Nam/60a2ad4c7db43bace75805054603747fcd062c0d
|
|
"""
|
|
import paddle
|
|
from paddle import nn
|
|
|
|
|
|
class JDCNet(nn.Layer):
|
|
"""
|
|
Joint Detection and Classification Network model for singing voice melody.
|
|
"""
|
|
|
|
def __init__(self,
|
|
num_class: int=722,
|
|
seq_len: int=31,
|
|
leaky_relu_slope: float=0.01):
|
|
super().__init__()
|
|
self.seq_len = seq_len
|
|
self.num_class = num_class
|
|
# input: (B, num_class, T, n_mels)
|
|
self.conv_block = nn.Sequential(
|
|
# output: (B, out_channels, T, n_mels)
|
|
nn.Conv2D(
|
|
in_channels=1,
|
|
out_channels=64,
|
|
kernel_size=3,
|
|
padding=1,
|
|
bias_attr=False),
|
|
nn.BatchNorm2D(num_features=64),
|
|
nn.LeakyReLU(leaky_relu_slope),
|
|
# out: (B, out_channels, T, n_mels)
|
|
nn.Conv2D(64, 64, 3, padding=1, bias_attr=False), )
|
|
# output: (B, out_channels, T, n_mels // 2)
|
|
self.res_block1 = ResBlock(in_channels=64, out_channels=128)
|
|
# output: (B, out_channels, T, n_mels // 4)
|
|
self.res_block2 = ResBlock(in_channels=128, out_channels=192)
|
|
# output: (B, out_channels, T, n_mels // 8)
|
|
self.res_block3 = ResBlock(in_channels=192, out_channels=256)
|
|
# pool block
|
|
self.pool_block = nn.Sequential(
|
|
nn.BatchNorm2D(num_features=256),
|
|
nn.LeakyReLU(leaky_relu_slope),
|
|
# (B, num_features, T, 2)
|
|
nn.MaxPool2D(kernel_size=(1, 4)),
|
|
nn.Dropout(p=0.5), )
|
|
# input: (B, T, input_size), resized from (B, input_size // 2, T, 2)
|
|
# output: (B, T, input_size)
|
|
self.bilstm_classifier = nn.LSTM(
|
|
input_size=512,
|
|
hidden_size=256,
|
|
time_major=False,
|
|
direction='bidirectional')
|
|
# input: (B * T, in_features)
|
|
# output: (B * T, num_class)
|
|
self.classifier = nn.Linear(
|
|
in_features=512, out_features=self.num_class)
|
|
|
|
# initialize weights
|
|
self.apply(self.init_weights)
|
|
|
|
def get_feature_GAN(self, x: paddle.Tensor):
|
|
"""Calculate feature_GAN.
|
|
Args:
|
|
x(Tensor(float32)):
|
|
Shape (B, num_class, n_mels, T).
|
|
Returns:
|
|
Tensor:
|
|
Shape (B, num_features, n_mels // 8, T).
|
|
"""
|
|
x = x.astype(paddle.float32)
|
|
x = x.transpose([0, 1, 3, 2] if len(x.shape) == 4 else [0, 2, 1])
|
|
convblock_out = self.conv_block(x)
|
|
resblock1_out = self.res_block1(convblock_out)
|
|
resblock2_out = self.res_block2(resblock1_out)
|
|
resblock3_out = self.res_block3(resblock2_out)
|
|
poolblock_out = self.pool_block[0](resblock3_out)
|
|
poolblock_out = self.pool_block[1](poolblock_out)
|
|
GAN_feature = poolblock_out.transpose([0, 1, 3, 2] if len(
|
|
poolblock_out.shape) == 4 else [0, 2, 1])
|
|
return GAN_feature
|
|
|
|
def forward(self, x: paddle.Tensor):
|
|
"""Calculate forward propagation.
|
|
Args:
|
|
x(Tensor(float32)):
|
|
Shape (B, num_class, n_mels, seq_len).
|
|
Returns:
|
|
Tensor:
|
|
classifier output consists of predicted pitch classes per frame.
|
|
Shape: (B, seq_len, num_class).
|
|
Tensor:
|
|
GAN_feature. Shape: (B, num_features, n_mels // 8, seq_len)
|
|
Tensor:
|
|
poolblock_out. Shape (B, seq_len, 512)
|
|
"""
|
|
###############################
|
|
# forward pass for classifier #
|
|
###############################
|
|
# (B, num_class, n_mels, T) -> (B, num_class, T, n_mels)
|
|
x = x.transpose([0, 1, 3, 2] if len(x.shape) == 4 else
|
|
[0, 2, 1]).astype(paddle.float32)
|
|
|
|
convblock_out = self.conv_block(x)
|
|
resblock1_out = self.res_block1(convblock_out)
|
|
resblock2_out = self.res_block2(resblock1_out)
|
|
resblock3_out = self.res_block3(resblock2_out)
|
|
poolblock_out = self.pool_block[0](resblock3_out)
|
|
poolblock_out = self.pool_block[1](poolblock_out)
|
|
GAN_feature = poolblock_out.transpose([0, 1, 3, 2] if len(
|
|
poolblock_out.shape) == 4 else [0, 2, 1])
|
|
poolblock_out = self.pool_block[2](poolblock_out)
|
|
# (B, 256, seq_len, 2) => (B, seq_len, 256, 2) => (B, seq_len, 512)
|
|
classifier_out = poolblock_out.transpose([0, 2, 1, 3]).reshape(
|
|
(-1, self.seq_len, 512))
|
|
self.bilstm_classifier.flatten_parameters()
|
|
# ignore the hidden states
|
|
classifier_out, _ = self.bilstm_classifier(classifier_out)
|
|
# (B * seq_len, 512)
|
|
classifier_out = classifier_out.reshape((-1, 512))
|
|
classifier_out = self.classifier(classifier_out)
|
|
# (B, seq_len, num_class)
|
|
classifier_out = classifier_out.reshape(
|
|
(-1, self.seq_len, self.num_class))
|
|
return paddle.abs(classifier_out.squeeze()), GAN_feature, poolblock_out
|
|
|
|
@staticmethod
|
|
def init_weights(m):
|
|
if isinstance(m, nn.Linear):
|
|
nn.initializer.KaimingUniform()(m.weight)
|
|
if m.bias is not None:
|
|
nn.initializer.Constant(0)(m.bias)
|
|
elif isinstance(m, nn.Conv2D):
|
|
nn.initializer.XavierNormal()(m.weight)
|
|
elif isinstance(m, nn.LSTM) or isinstance(m, nn.LSTMCell):
|
|
for p in m.parameters():
|
|
if len(p.shape) >= 2:
|
|
nn.initializer.Orthogonal()(p)
|
|
else:
|
|
nn.initializer.Normal()(p)
|
|
|
|
|
|
class ResBlock(nn.Layer):
|
|
def __init__(self,
|
|
in_channels: int,
|
|
out_channels: int,
|
|
leaky_relu_slope: float=0.01):
|
|
super().__init__()
|
|
self.downsample = in_channels != out_channels
|
|
# BN / LReLU / MaxPool layer before the conv layer - see Figure 1b in the paper
|
|
self.pre_conv = nn.Sequential(
|
|
nn.BatchNorm2D(num_features=in_channels),
|
|
nn.LeakyReLU(leaky_relu_slope),
|
|
# apply downsampling on the y axis only
|
|
nn.MaxPool2D(kernel_size=(1, 2)), )
|
|
|
|
# conv layers
|
|
self.conv = nn.Sequential(
|
|
nn.Conv2D(
|
|
in_channels=in_channels,
|
|
out_channels=out_channels,
|
|
kernel_size=3,
|
|
padding=1,
|
|
bias_attr=False),
|
|
nn.BatchNorm2D(out_channels),
|
|
nn.LeakyReLU(leaky_relu_slope),
|
|
nn.Conv2D(
|
|
in_channels=out_channels,
|
|
out_channels=out_channels,
|
|
kernel_size=3,
|
|
padding=1,
|
|
bias_attr=False), )
|
|
# 1 x 1 convolution layer to match the feature dimensions
|
|
self.conv1by1 = None
|
|
if self.downsample:
|
|
self.conv1by1 = nn.Conv2D(
|
|
in_channels=in_channels,
|
|
out_channels=out_channels,
|
|
kernel_size=1,
|
|
bias_attr=False)
|
|
|
|
def forward(self, x: paddle.Tensor):
|
|
"""Calculate forward propagation.
|
|
Args:
|
|
x(Tensor(float32)): Shape (B, in_channels, T, n_mels).
|
|
Returns:
|
|
Tensor:
|
|
The residual output, Shape (B, out_channels, T, n_mels // 2).
|
|
"""
|
|
x = self.pre_conv(x)
|
|
if self.downsample:
|
|
x = self.conv(x) + self.conv1by1(x)
|
|
else:
|
|
x = self.conv(x) + x
|
|
return x
|