(self, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=0,
p_dropout=0, share_cond_layers=False)
| 20 | |
| 21 | class WN(torch.nn.Module): |
| 22 | def __init__(self, hidden_channels, kernel_size, dilation_rate, n_layers, gin_channels=0, |
| 23 | p_dropout=0, share_cond_layers=False): |
| 24 | super(WN, self).__init__() |
| 25 | assert (kernel_size % 2 == 1) |
| 26 | assert (hidden_channels % 2 == 0) |
| 27 | self.hidden_channels = hidden_channels |
| 28 | self.kernel_size = kernel_size |
| 29 | self.dilation_rate = dilation_rate |
| 30 | self.n_layers = n_layers |
| 31 | self.gin_channels = gin_channels |
| 32 | self.p_dropout = p_dropout |
| 33 | self.share_cond_layers = share_cond_layers |
| 34 | |
| 35 | self.in_layers = torch.nn.ModuleList() |
| 36 | self.res_skip_layers = torch.nn.ModuleList() |
| 37 | |
| 38 | self.drop = nn.Dropout(p_dropout) |
| 39 | |
| 40 | self.use_adapters = hparams.get("use_adapters", False) |
| 41 | if self.use_adapters: |
| 42 | self.adapter_layers = torch.nn.ModuleList() |
| 43 | |
| 44 | if gin_channels != 0 and not share_cond_layers: |
| 45 | cond_layer = torch.nn.Conv1d(gin_channels, 2 * hidden_channels * n_layers, 1) |
| 46 | self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name='weight') |
| 47 | |
| 48 | for i in range(n_layers): |
| 49 | dilation = dilation_rate ** i |
| 50 | padding = int((kernel_size * dilation - dilation) / 2) |
| 51 | in_layer = torch.nn.Conv1d(hidden_channels, 2 * hidden_channels, kernel_size, |
| 52 | dilation=dilation, padding=padding) |
| 53 | in_layer = torch.nn.utils.weight_norm(in_layer, name='weight') |
| 54 | self.in_layers.append(in_layer) |
| 55 | |
| 56 | # last one is not necessary |
| 57 | if i < n_layers - 1: |
| 58 | res_skip_channels = 2 * hidden_channels |
| 59 | else: |
| 60 | res_skip_channels = hidden_channels |
| 61 | |
| 62 | res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1) |
| 63 | res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name='weight') |
| 64 | self.res_skip_layers.append(res_skip_layer) |
| 65 | |
| 66 | if self.use_adapters: |
| 67 | adapter_layer = MlpAdapter(in_out_dim=res_skip_channels, hid_dim=res_skip_channels//4) |
| 68 | self.adapter_layers.append(adapter_layer) |
| 69 | |
| 70 | def forward(self, x, x_mask=None, g=None, **kwargs): |
| 71 | output = torch.zeros_like(x) |
no test coverage detected