File size: 1,817 Bytes
2be48c4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
import torch
import torch.nn as nn
import torch.nn.functional as F

import random
import numpy as np
from scipy.fftpack import fft
import wave


class Model(nn.Module):
    def __init__(self, input_dim=1, hidden_dim = 256, tone_class=5, syllable_class=1000):

        super().__init__()
        self.input_dim = input_dim
        self.tone_class = tone_class
        self.syllable_class = syllable_class
        # hidden_size = 128*hidden_dim//16
        conv_layers = []
        in_channels = input_dim  # Input channels for the first layer
        channel_list = [16,16,'p2',32,32,'p2',64,64,'p1',64]
#         channel_list = [32,'p','p',128]
#         channel_list = [32,32,64,64,128]

        for out_channels in channel_list:
            if out_channels=='p2':
                conv_layers.append(nn.MaxPool2d(kernel_size=2))
                continue
            elif out_channels=='p1':
                conv_layers.append(nn.MaxPool2d(kernel_size=1))
                continue
            conv_layers.append(nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1))
            conv_layers.append(nn.BatchNorm2d(out_channels))
            conv_layers.append(nn.ReLU(inplace=True))
            conv_layers.append(nn.Dropout(0.1))
            in_channels = out_channels

        self.conv = nn.Sequential(*conv_layers)
            
        
            
        self.output = nn.Sequential(
            nn.Linear(4096, 128),
            nn.ReLU(),
            nn.Dropout(0.1),
 
            nn.Linear(128,syllable_class)
        )
        
    def forward(self, x):

        x = self.conv(x) #[batch_size,channel,length(input_length//4),hidden_dim]
        x = x.permute((0,2,1,3))#[batch_size,length,channel,hidden_dim]
        x = x.reshape(x.shape[0],x.shape[1],-1)
        
        return self.output(x)