dnn_models.py

'''
This file contains the implementation of SincNet, by Mirco Ravanelli and Yoshua Bengio
Circular padding has been added before each convolution.
Source: https://github.com/mravanelli/SincNet
'''

import numpy as np
import torch
import torch.nn.functional as F
import torch.nn as nn
from torch.autograd import Variable
import math
from torch_same_pad import get_pad


def flip(x, dim):
    xsize = x.size()
    dim = x.dim() + dim if dim < 0 else dim
    x = x.contiguous()
    x = x.view(-1, *xsize[dim:])
    x = x.view(x.size(0), x.size(1), -1)[:, getattr(torch.arange(x.size(1)-1,
                                                                 -1, -1), ('cpu', 'cuda')[x.is_cuda])().long(), :]
    return x.view(xsize)


def sinc(band, t_right):
    y_right = torch.sin(2*math.pi*band*t_right)/(2*math.pi*band*t_right)
    y_left = flip(y_right, 0)

    y = torch.cat([y_left, Variable(torch.ones(1)).cuda(), y_right])

    return y


class SincConv_fast(nn.Module):
    """Sinc-based convolution
    Parameters
    ----------
    in_channels : `int`
        Number of input channels. Must be 1.
    out_channels : `int`
        Number of filters.
    kernel_size : `int`
        Filter length.
    sample_rate : `int`, optional
        Sample rate. Defaults to 16000.
    Usage
    -----
    See `torch.nn.Conv1d`
    Reference
    ---------
    Mirco Ravanelli, Yoshua Bengio,
    "Speaker Recognition from raw waveform with SincNet".
    https://arxiv.org/abs/1808.00158
    """

    @staticmethod
    def to_mel(hz):
        return 2595 * np.log10(1 + hz / 700)

    @staticmethod
    def to_hz(mel):
        return 700 * (10 ** (mel / 2595) - 1)

    def __init__(self, out_channels, kernel_size, sample_rate=16000, in_channels=1,
                 stride=1, padding=0, dilation=1, bias=False, groups=1, min_low_hz=50, min_band_hz=50):

        super(SincConv_fast, self).__init__()

        if in_channels != 1:
            #msg = (f'SincConv only support one input channel '
            #       f'(here, in_channels = {in_channels:d}).')
            msg = "SincConv only support one input channel (here, in_channels = {%i})" % (
                in_channels)
            raise ValueError(msg)

        self.out_channels = out_channels
        self.kernel_size = kernel_size

        # Forcing the filters to be odd (i.e, perfectly symmetrics)
        if kernel_size % 2 == 0:
            self.kernel_size = self.kernel_size+1

        self.stride = stride
        self.padding = padding
        self.dilation = dilation

        if bias:
            raise ValueError('SincConv does not support bias.')
        if groups > 1:
            raise ValueError('SincConv does not support groups.')

        self.sample_rate = sample_rate
        self.min_low_hz = min_low_hz
        self.min_band_hz = min_band_hz

        # initialize filterbanks such that they are equally spaced in Mel scale
        low_hz = 30
        high_hz = self.sample_rate / 2 - (self.min_low_hz + self.min_band_hz)

        mel = np.linspace(self.to_mel(low_hz),
                          self.to_mel(high_hz),
                          self.out_channels + 1)
        hz = self.to_hz(mel)

        # filter lower frequency (out_channels, 1)
        self.low_hz_ = nn.Parameter(torch.Tensor(hz[:-1]).view(-1, 1))

        # filter frequency band (out_channels, 1)
        self.band_hz_ = nn.Parameter(torch.Tensor(np.diff(hz)).view(-1, 1))

        # Hamming window
        #self.window_ = torch.hamming_window(self.kernel_size)
        # computing only half of the window
        n_lin = torch.linspace(0, (self.kernel_size/2)-1,
                               steps=int((self.kernel_size/2)))
        self.window_ = 0.54-0.46*torch.cos(2*math.pi*n_lin/self.kernel_size)

        # (1, kernel_size/2)
        n = (self.kernel_size - 1) / 2.0
        # Due to symmetry, I only need half of the time axes
        self.n_ = 2*math.pi*torch.arange(-n, 0).view(1, -1) / self.sample_rate

    def forward(self, waveforms):
        """
        Parameters
        ----------
        waveforms : `torch.Tensor` (batch_size, 1, n_samples)
            Batch of waveforms.
        Returns
        -------
        features : `torch.Tensor` (batch_size, out_channels, n_samples_out)
            Batch of sinc filters activations.
        """

        self.n_ = self.n_.to(waveforms.device)

        self.window_ = self.window_.to(waveforms.device)

        low = self.min_low_hz + torch.abs(self.low_hz_)

        high = torch.clamp(low + self.min_band_hz
                           + torch.abs(self.band_hz_), self.min_low_hz, self.sample_rate/2)
        band = (high-low)[:, 0]

        f_times_t_low = torch.matmul(low, self.n_)
        f_times_t_high = torch.matmul(high, self.n_)

        # Equivalent of Eq.4 of the reference paper (SPEAKER RECOGNITION FROM RAW WAVEFORM WITH SINCNET). I just have expanded the sinc and simplified the terms. This way I avoid several useless computations.
        band_pass_left = ((torch.sin(f_times_t_high)
                          - torch.sin(f_times_t_low))/(self.n_/2))*self.window_
        band_pass_center = 2*band.view(-1, 1)
        band_pass_right = torch.flip(band_pass_left, dims=[1])

        band_pass = torch.cat(
            [band_pass_left, band_pass_center, band_pass_right], dim=1)

        band_pass = band_pass / (2*band[:, None])

        self.filters = (band_pass).view(
            self.out_channels, 1, self.kernel_size)

        return F.conv1d(waveforms, self.filters, stride=self.stride,
                        padding=self.padding, dilation=self.dilation,
                        bias=None, groups=1)


class sinc_conv(nn.Module):

    def __init__(self, N_filt, Filt_dim, fs):
        super(sinc_conv, self).__init__()

        # Mel Initialization of the filterbanks
        low_freq_mel = 80
        high_freq_mel = (2595 * np.log10(1 + (fs / 2) / 700)
                         )  # Convert Hz to Mel
        # Equally spaced in Mel scale
        mel_points = np.linspace(low_freq_mel, high_freq_mel, N_filt)
        f_cos = (700 * (10**(mel_points / 2595) - 1))  # Convert Mel to Hz
        b1 = np.roll(f_cos, 1)
        b2 = np.roll(f_cos, -1)
        b1[0] = 30
        b2[-1] = (fs/2)-100

        self.freq_scale = fs*1.0
        self.filt_b1 = nn.Parameter(torch.from_numpy(b1/self.freq_scale))
        self.filt_band = nn.Parameter(
            torch.from_numpy((b2-b1)/self.freq_scale))

        self.N_filt = N_filt
        self.Filt_dim = Filt_dim
        self.fs = fs

    def forward(self, x):

        filters = Variable(torch.zeros((self.N_filt, self.Filt_dim))).cuda()
        N = self.Filt_dim
        t_right = Variable(torch.linspace(
            1, (N-1)/2, steps=int((N-1)/2))/self.fs).cuda()

        min_freq = 50.0
        min_band = 50.0

        filt_beg_freq = torch.abs(self.filt_b1)+min_freq/self.freq_scale
        filt_end_freq = filt_beg_freq + \
            (torch.abs(self.filt_band)+min_band/self.freq_scale)

        n = torch.linspace(0, N, steps=N)

        # Filter window (hamming)
        window = 0.54-0.46*torch.cos(2*math.pi*n/N)
        window = Variable(window.float().cuda())

        for i in range(self.N_filt):

            low_pass1 = 2 * \
                filt_beg_freq[i].float()*sinc(filt_beg_freq[i].float()
                                              * self.freq_scale, t_right)
            low_pass2 = 2 * \
                filt_end_freq[i].float()*sinc(filt_end_freq[i].float()
                                              * self.freq_scale, t_right)
            band_pass = (low_pass2-low_pass1)

            band_pass = band_pass/torch.max(band_pass)

            filters[i, :] = band_pass.cuda()*window

        out = F.conv1d(x, filters.view(self.N_filt, 1, self.Filt_dim))

        return out


def act_fun(act_type):

 if act_type == "relu":
    return nn.ReLU()

 if act_type == "tanh":
    return nn.Tanh()

 if act_type == "sigmoid":
    return nn.Sigmoid()

 if act_type == "leaky_relu":
    return nn.LeakyReLU(0.2)

 if act_type == "elu":
    return nn.ELU()

 if act_type == "softmax":
    return nn.LogSoftmax(dim=1)

 if act_type == "linear":
    return nn.LeakyReLU(1)  # initializzed like this, but not used in forward!


class LayerNorm(nn.Module):

    def __init__(self, features, eps=1e-6):
        super(LayerNorm, self).__init__()
        self.gamma = nn.Parameter(torch.ones(features))
        self.beta = nn.Parameter(torch.zeros(features))
        self.eps = eps

    def forward(self, x):
        mean = x.mean(-1, keepdim=True)
        std = x.std(-1, keepdim=True)
        return self.gamma * (x - mean) / (std + self.eps) + self.beta


class MLP(nn.Module):
    def __init__(self, options):
        super(MLP, self).__init__()

        self.input_dim = int(options['input_dim'])
        self.fc_lay = options['fc_lay']
        self.fc_drop = options['fc_drop']
        self.fc_use_batchnorm = options['fc_use_batchnorm']
        self.fc_use_laynorm = options['fc_use_laynorm']
        self.fc_use_laynorm_inp = options['fc_use_laynorm_inp']
        self.fc_use_batchnorm_inp = options['fc_use_batchnorm_inp']
        self.fc_act = options['fc_act']

        self.wx = nn.ModuleList([])
        self.bn = nn.ModuleList([])
        self.ln = nn.ModuleList([])
        self.act = nn.ModuleList([])
        self.drop = nn.ModuleList([])

        # input layer normalization
        if self.fc_use_laynorm_inp:
           self.ln0 = LayerNorm(self.input_dim)

        # input batch normalization
        if self.fc_use_batchnorm_inp:
           self.bn0 = nn.BatchNorm1d([self.input_dim], momentum=0.05)

        self.N_fc_lay = len(self.fc_lay)

        current_input = self.input_dim

        # Initialization of hidden layers

        for i in range(self.N_fc_lay):

         # dropout
         self.drop.append(nn.Dropout(p=self.fc_drop[i]))

         # activation
         self.act.append(act_fun(self.fc_act[i]))

         add_bias = True

         # layer norm initialization
         self.ln.append(LayerNorm(self.fc_lay[i]))
         self.bn.append(nn.BatchNorm1d(self.fc_lay[i], momentum=0.05))

         if self.fc_use_laynorm[i] or self.fc_use_batchnorm[i]:
             add_bias = False

         # Linear operations
         self.wx.append(
             nn.Linear(current_input, self.fc_lay[i], bias=add_bias))

         # weight initialization
         self.wx[i].weight = torch.nn.Parameter(torch.Tensor(self.fc_lay[i], current_input).uniform_(
             -np.sqrt(0.01/(current_input+self.fc_lay[i])), np.sqrt(0.01/(current_input+self.fc_lay[i]))))
         self.wx[i].bias = torch.nn.Parameter(torch.zeros(self.fc_lay[i]))

         current_input = self.fc_lay[i]

    def forward(self, x):

      # Applying Layer/Batch Norm
      if bool(self.fc_use_laynorm_inp):
        x = self.ln0((x))

      if bool(self.fc_use_batchnorm_inp):
        x = self.bn0((x))

      for i in range(self.N_fc_lay):

        if self.fc_act[i] != 'linear':

          if self.fc_use_laynorm[i]:
           x = self.drop[i](self.act[i](self.ln[i](self.wx[i](x))))

          if self.fc_use_batchnorm[i]:
           x = self.drop[i](self.act[i](self.bn[i](self.wx[i](x))))

          if self.fc_use_batchnorm[i] == False and self.fc_use_laynorm[i] == False:
           x = self.drop[i](self.act[i](self.wx[i](x)))

        else:
          if self.fc_use_laynorm[i]:
           x = self.drop[i](self.ln[i](self.wx[i](x)))

          if self.fc_use_batchnorm[i]:
           x = self.drop[i](self.bn[i](self.wx[i](x)))

          if self.fc_use_batchnorm[i] == False and self.fc_use_laynorm[i] == False:
           x = self.drop[i](self.wx[i](x))

      return x


class SincNet(nn.Module):

    def __init__(self, options):
       super(SincNet, self).__init__()

       self.cnn_N_filt = options['cnn_N_filt']
       self.cnn_len_filt = options['cnn_len_filt']
       self.cnn_max_pool_len = options['cnn_max_pool_len']

       self.cnn_act = options['cnn_act']
       self.cnn_drop = options['cnn_drop']

       self.cnn_use_laynorm = options['cnn_use_laynorm']
       self.cnn_use_batchnorm = options['cnn_use_batchnorm']
       self.cnn_use_laynorm_inp = options['cnn_use_laynorm_inp']
       self.cnn_use_batchnorm_inp = options['cnn_use_batchnorm_inp']

       self.input_dim = int(options['input_dim'])

       self.fs = options['fs']

       self.N_cnn_lay = len(options['cnn_N_filt'])
       self.conv = nn.ModuleList([])
       self.bn = nn.ModuleList([])
       self.ln = nn.ModuleList([])
       self.act = nn.ModuleList([])
       self.drop = nn.ModuleList([])
       self.use_sinc = options['use_sinc']

       if self.cnn_use_laynorm_inp:
           self.ln0 = LayerNorm(self.input_dim)

       if self.cnn_use_batchnorm_inp:
           self.bn0 = nn.BatchNorm1d([self.input_dim], momentum=0.05)

       current_input = self.input_dim

       for i in range(self.N_cnn_lay):

         N_filt = int(self.cnn_N_filt[i])
         len_filt = int(self.cnn_len_filt[i])

         # dropout
         self.drop.append(nn.Dropout(p=self.cnn_drop[i]))

         # activation
         self.act.append(act_fun(self.cnn_act[i]))

         # layer norm initialization
         #self.ln.append(LayerNorm([N_filt,int((current_input-self.cnn_len_filt[i]+1)/self.cnn_max_pool_len[i])]))

         self.bn.append(nn.BatchNorm1d(N_filt, momentum=0.05))

         if i == 0:
          if self.use_sinc:
            self.conv.append(SincConv_fast(
                self.cnn_N_filt[0], self.cnn_len_filt[0], self.fs))
          else:
            self.conv.append(
                nn.Conv1d(1, self.cnn_N_filt[i], self.cnn_len_filt[i]))

         else:
          self.conv.append(
              nn.Conv1d(self.cnn_N_filt[i-1], self.cnn_N_filt[i], self.cnn_len_filt[i]))

         current_input = int(
             (current_input-self.cnn_len_filt[i]+1)/self.cnn_max_pool_len[i])

       self.out_dim = current_input*N_filt

    def forward(self, x):
       batch = x.shape[0]
       seq_len = x.shape[-1]

       if bool(self.cnn_use_laynorm_inp):
        x = self.ln0((x))

       if bool(self.cnn_use_batchnorm_inp):
        x = self.bn0((x))

       x = x.view(batch, 1, seq_len)

       for i in range(self.N_cnn_lay):

         s = x.shape[2]
         padding = get_pad(
             size=s, kernel_size=self.cnn_len_filt[i], stride=1, dilation=1)
         x = F.pad(x, pad=padding, mode='circular')

         if self.cnn_use_laynorm[i]:
          if i == 0:
           x = self.drop[i](self.act[i](self.ln[i](F.max_pool1d(
               torch.abs(self.conv[i](x)), self.cnn_max_pool_len[i]))))
          else:
           x = self.drop[i](self.act[i](self.ln[i](
               F.max_pool1d(self.conv[i](x), self.cnn_max_pool_len[i]))))

         if self.cnn_use_batchnorm[i]:
          x = self.drop[i](self.act[i](self.bn[i](
              F.max_pool1d(self.conv[i](x), self.cnn_max_pool_len[i]))))

         if self.cnn_use_batchnorm[i] == False and self.cnn_use_laynorm[i] == False:
          x = self.drop[i](self.act[i](F.max_pool1d(
              self.conv[i](x), self.cnn_max_pool_len[i])))

       #x = x.view(batch,-1)

       return x