|
|
import torchvision.models as models
|
|
|
import torch
|
|
|
import torch.nn as nn
|
|
|
import torch.nn.functional as F
|
|
|
|
|
|
class BasicBlock1D(nn.Module):
|
|
|
expansion = 1
|
|
|
|
|
|
def __init__(self, in_channels, out_channels, stride=1, downsample=None):
|
|
|
super(BasicBlock1D, self).__init__()
|
|
|
self.conv1 = nn.Conv1d(in_channels, out_channels, kernel_size=3,
|
|
|
stride=stride, padding=1, bias=False)
|
|
|
self.bn1 = nn.BatchNorm1d(out_channels)
|
|
|
self.relu = nn.ReLU(inplace=True)
|
|
|
self.conv2 = nn.Conv1d(out_channels, out_channels, kernel_size=3,
|
|
|
stride=1, padding=1, bias=False)
|
|
|
self.bn2 = nn.BatchNorm1d(out_channels)
|
|
|
self.downsample = downsample
|
|
|
self.stride = stride
|
|
|
|
|
|
def forward(self, x):
|
|
|
identity = x
|
|
|
|
|
|
out = self.conv1(x)
|
|
|
out = self.bn1(out)
|
|
|
out = self.relu(out)
|
|
|
|
|
|
out = self.conv2(out)
|
|
|
out = self.bn2(out)
|
|
|
|
|
|
if self.downsample is not None:
|
|
|
identity = self.downsample(x)
|
|
|
|
|
|
out += identity
|
|
|
out = self.relu(out)
|
|
|
|
|
|
return out
|
|
|
|
|
|
|
|
|
class Bottleneck1D(nn.Module):
|
|
|
expansion = 4
|
|
|
|
|
|
def __init__(self, in_channels, out_channels, stride=1, downsample=None):
|
|
|
super(Bottleneck1D, self).__init__()
|
|
|
self.conv1 = nn.Conv1d(in_channels, out_channels, kernel_size=1, bias=False)
|
|
|
self.bn1 = nn.BatchNorm1d(out_channels)
|
|
|
self.conv2 = nn.Conv1d(out_channels, out_channels, kernel_size=3, stride=stride,
|
|
|
padding=1, bias=False)
|
|
|
self.bn2 = nn.BatchNorm1d(out_channels)
|
|
|
self.conv3 = nn.Conv1d(out_channels, out_channels * self.expansion, kernel_size=1, bias=False)
|
|
|
self.bn3 = nn.BatchNorm1d(out_channels * self.expansion)
|
|
|
self.relu = nn.ReLU(inplace=True)
|
|
|
self.downsample = downsample
|
|
|
self.stride = stride
|
|
|
|
|
|
def forward(self, x):
|
|
|
identity = x
|
|
|
|
|
|
out = self.conv1(x)
|
|
|
out = self.bn1(out)
|
|
|
out = self.relu(out)
|
|
|
|
|
|
out = self.conv2(out)
|
|
|
out = self.bn2(out)
|
|
|
out = self.relu(out)
|
|
|
|
|
|
out = self.conv3(out)
|
|
|
out = self.bn3(out)
|
|
|
|
|
|
if self.downsample is not None:
|
|
|
identity = self.downsample(x)
|
|
|
|
|
|
out += identity
|
|
|
out = self.relu(out)
|
|
|
|
|
|
return out
|
|
|
|
|
|
|
|
|
class AudioResNet(nn.Module):
|
|
|
def __init__(self, block, layers, num_classes=7):
|
|
|
"""
|
|
|
构建用于音频分类的1D ResNet
|
|
|
|
|
|
参数:
|
|
|
block: 使用的残差块类型(BasicBlock1D or Bottleneck1D)
|
|
|
layers: 每个层的块数量的列表
|
|
|
num_classes: 分类的类别数量，默认为7种情感
|
|
|
"""
|
|
|
super(AudioResNet, self).__init__()
|
|
|
self.in_channels = 64
|
|
|
|
|
|
# 初始卷积层，缩减序列长度
|
|
|
self.conv1 = nn.Conv1d(1, 64, kernel_size=7, stride=2, padding=3, bias=False)
|
|
|
self.bn1 = nn.BatchNorm1d(64)
|
|
|
self.relu = nn.ReLU(inplace=True)
|
|
|
self.maxpool = nn.MaxPool1d(kernel_size=3, stride=2, padding=1)
|
|
|
|
|
|
# 残差块堆叠
|
|
|
self.layer1 = self._make_layer(block, 64, layers[0])
|
|
|
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
|
|
|
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
|
|
|
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
|
|
|
|
|
|
# 全局平均池化和分类器
|
|
|
self.avgpool = nn.AdaptiveAvgPool1d(1)
|
|
|
self.fc = nn.Linear(512 * block.expansion, num_classes)
|
|
|
|
|
|
# 权重初始化
|
|
|
for m in self.modules():
|
|
|
if isinstance(m, nn.Conv1d):
|
|
|
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
|
|
|
elif isinstance(m, nn.BatchNorm1d):
|
|
|
nn.init.constant_(m.weight, 1)
|
|
|
nn.init.constant_(m.bias, 0)
|
|
|
|
|
|
def _make_layer(self, block, out_channels, blocks, stride=1):
|
|
|
downsample = None
|
|
|
if stride != 1 or self.in_channels != out_channels * block.expansion:
|
|
|
downsample = nn.Sequential(
|
|
|
nn.Conv1d(self.in_channels, out_channels * block.expansion,
|
|
|
kernel_size=1, stride=stride, bias=False),
|
|
|
nn.BatchNorm1d(out_channels * block.expansion),
|
|
|
)
|
|
|
|
|
|
layers = []
|
|
|
layers.append(block(self.in_channels, out_channels, stride, downsample))
|
|
|
self.in_channels = out_channels * block.expansion
|
|
|
for _ in range(1, blocks):
|
|
|
layers.append(block(self.in_channels, out_channels))
|
|
|
|
|
|
return nn.Sequential(*layers)
|
|
|
|
|
|
def forward(self, x):
|
|
|
# 输入 x 形状: [batch_size, 1, 24000]
|
|
|
|
|
|
x = self.conv1(x) # [batch_size, 64, 12000]
|
|
|
x = self.bn1(x)
|
|
|
x = self.relu(x)
|
|
|
x = self.maxpool(x) # [batch_size, 64, 6000]
|
|
|
|
|
|
x = self.layer1(x) # [batch_size, 64*expansion, 6000]
|
|
|
x = self.layer2(x) # [batch_size, 128*expansion, 3000]
|
|
|
x = self.layer3(x) # [batch_size, 256*expansion, 1500]
|
|
|
x = self.layer4(x) # [batch_size, 512*expansion, 750]
|
|
|
|
|
|
x = self.avgpool(x) # [batch_size, 512*expansion, 1]
|
|
|
x = torch.flatten(x, 1) # [batch_size, 512*expansion]
|
|
|
x = self.fc(x) # [batch_size, num_classes]
|
|
|
|
|
|
return x
|
|
|
|
|
|
|
|
|
# 定义不同规模的ResNet模型
|
|
|
def waveform_resnet18(num_classes=7):
|
|
|
"""
|
|
|
构建类似ResNet18的音频分类模型
|
|
|
"""
|
|
|
return AudioResNet(BasicBlock1D, [2, 2, 2, 2], num_classes)
|
|
|
|
|
|
def waveform_resnet34(num_classes=7):
|
|
|
"""
|
|
|
构建类似ResNet34的音频分类模型
|
|
|
"""
|
|
|
return AudioResNet(BasicBlock1D, [3, 4, 6, 3], num_classes)
|
|
|
|
|
|
def waveform_resnet50(num_classes=7):
|
|
|
"""
|
|
|
构建类似ResNet50的音频分类模型
|
|
|
"""
|
|
|
return AudioResNet(Bottleneck1D, [3, 4, 6, 3], num_classes)
|
|
|
|
|
|
def waveform_resnet101(num_classes=7):
|
|
|
"""
|
|
|
构建类似ResNet101的音频分类模型
|
|
|
"""
|
|
|
return AudioResNet(Bottleneck1D, [3, 4, 23, 3], num_classes)
|
|
|
|
|
|
|
|
|
class SpectrogramResNet(nn.Module):
|
|
|
"""
|
|
|
使用预训练的ResNet模型对音频频谱图进行情感分类
|
|
|
"""
|
|
|
def __init__(self, model_name='resnet18', num_classes=6, pretrained=True):
|
|
|
"""
|
|
|
初始化频谱图ResNet分类模型
|
|
|
|
|
|
参数:
|
|
|
model_name: 使用的ResNet版本 ('resnet18', 'resnet34', 'resnet50', 'resnet101')
|
|
|
num_classes: 情感类别数量
|
|
|
pretrained: 是否使用预训练权重
|
|
|
"""
|
|
|
super(SpectrogramResNet, self).__init__()
|
|
|
|
|
|
# 选择预训练的ResNet模型
|
|
|
if model_name == 'resnet18':
|
|
|
base_model = models.resnet18(pretrained=pretrained)
|
|
|
elif model_name == 'resnet34':
|
|
|
base_model = models.resnet34(pretrained=pretrained)
|
|
|
elif model_name == 'resnet50':
|
|
|
base_model = models.resnet50(pretrained=pretrained)
|
|
|
elif model_name == 'resnet101':
|
|
|
base_model = models.resnet101(pretrained=pretrained)
|
|
|
else:
|
|
|
raise ValueError(f"不支持的模型名称: {model_name}")
|
|
|
|
|
|
# 修改第一个卷积层以接受单通道输入
|
|
|
self.conv1 = nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3, bias=False)
|
|
|
|
|
|
# 使用预训练权重初始化第一层（如果可用）
|
|
|
if pretrained:
|
|
|
# 将预训练的三通道权重平均为单通道权重
|
|
|
with torch.no_grad():
|
|
|
self.conv1.weight.data = base_model.conv1.weight.data.mean(dim=1, keepdim=True)
|
|
|
|
|
|
# 使用其余的预训练层
|
|
|
self.bn1 = base_model.bn1
|
|
|
self.relu = base_model.relu
|
|
|
self.maxpool = base_model.maxpool
|
|
|
self.layer1 = base_model.layer1
|
|
|
self.layer2 = base_model.layer2
|
|
|
self.layer3 = base_model.layer3
|
|
|
self.layer4 = base_model.layer4
|
|
|
self.avgpool = base_model.avgpool
|
|
|
|
|
|
# 修改全连接层以匹配目标类别数
|
|
|
in_features = base_model.fc.in_features
|
|
|
self.fc = nn.Linear(in_features, num_classes)
|
|
|
|
|
|
def forward(self, x):
|
|
|
"""
|
|
|
前向传播
|
|
|
|
|
|
参数:
|
|
|
x: 形状为 [batch_size, 1, 128, 128] 的频谱图
|
|
|
|
|
|
返回:
|
|
|
形状为 [batch_size, num_classes] 的类别预测
|
|
|
"""
|
|
|
x = self.conv1(x)
|
|
|
x = self.bn1(x)
|
|
|
x = self.relu(x)
|
|
|
x = self.maxpool(x)
|
|
|
|
|
|
x = self.layer1(x)
|
|
|
x = self.layer2(x)
|
|
|
x = self.layer3(x)
|
|
|
x = self.layer4(x)
|
|
|
|
|
|
x = self.avgpool(x)
|
|
|
x = torch.flatten(x, 1)
|
|
|
x = self.fc(x)
|
|
|
|
|
|
return x
|
|
|
|
|
|
# 便捷函数用于创建不同版本的模型
|
|
|
def spectrogram_resnet18(num_classes=6, pretrained=True):
|
|
|
return SpectrogramResNet('resnet18', num_classes, pretrained)
|
|
|
|
|
|
def spectrogram_resnet34(num_classes=6, pretrained=True):
|
|
|
return SpectrogramResNet('resnet34', num_classes, pretrained)
|
|
|
|
|
|
def spectrogram_resnet50(num_classes=6, pretrained=True):
|
|
|
return SpectrogramResNet('resnet50', num_classes, pretrained)
|
|
|
|
|
|
def spectrogram_resnet101(num_classes=6, pretrained=True):
|
|
|
return SpectrogramResNet('resnet101', num_classes, pretrained)
|
|
|
|
