small refactoring

layers/common_layers.py

@@ -84,15 +84,17 @@ class Prenet(nn.Module):
 
 
 class LocationLayer(nn.Module):
-    def __init__(self, attention_n_filters, attention_kernel_size,
-                 attention_dim):
+    def __init__(self,
+                 attention_dim,
+                 attention_n_filters=32,
+                 attention_kernel_size=31):
         super(LocationLayer, self).__init__()
         self.location_conv = nn.Conv1d(
             in_channels=2,
             out_channels=attention_n_filters,
-            kernel_size=31,
+            kernel_size=attention_kernel_size,
             stride=1,
-            padding=(31 - 1) // 2,
+            padding=(attention_kernel_size - 1) // 2,
             bias=False)
         self.location_dense = Linear(
             attention_n_filters, attention_dim, bias=False, init_gain='tanh')
@@ -120,8 +122,10 @@ class Attention(nn.Module):
                 attention_rnn_dim + embedding_dim, 1, bias=True)
         if location_attention:
             self.location_layer = LocationLayer(
-                attention_location_n_filters, attention_location_kernel_size,
-                attention_dim)
+                attention_dim,
+                attention_location_n_filters,
+                attention_location_kernel_size,
+            )
         self._mask_value = -float("inf")
         self.windowing = windowing
         self.win_idx = None
@@ -203,14 +207,18 @@ class Attention(nn.Module):
         # compute transition potentials
         alpha = (((1 - self.u) * self.alpha.clone().to(inputs.device) +
                   self.u * prev_alpha) + 1e-8) * alignment
-        # force incremental alignment - TODO: make configurable
+        # force incremental alignment
         if not self.training and self.forward_attn_mask:
             _, n = prev_alpha.max(1)
             val, n2 = alpha.max(1)
             for b in range(alignment.shape[0]):
                 alpha[b, n[b] + 3:] = 0
-                alpha[b, :(n[b] - 1)] = 0  # ignore all previous states to prevent repetition.
-                alpha[b, (n[b] - 2)] = 0.01 * val[b]  # smoothing factor for the prev step
+                alpha[b, :(
+                    n[b] - 1
+                )] = 0  # ignore all previous states to prevent repetition.
+                alpha[b,
+                      (n[b] - 2
+                       )] = 0.01 * val[b]  # smoothing factor for the prev step
         # compute attention weights
         self.alpha = alpha / alpha.sum(dim=1).unsqueeze(1)
         # compute context
@@ -240,7 +248,8 @@ class Attention(nn.Module):
             alignment = torch.softmax(attention, dim=-1)
         elif self.norm == "sigmoid":
             alignment = torch.sigmoid(attention) / torch.sigmoid(
-                attention).sum(dim=1).unsqueeze(1)
+                attention).sum(
+                    dim=1, keepdim=True)
         else:
             raise RuntimeError("Unknown value for attention norm type")
         if self.location_attention:

layers/tacotron.py

@@ -268,6 +268,7 @@ class Decoder(nn.Module):
         memory_dim (int): memory vector (prev. time-step output) sample size.
         r (int): number of outputs per time step.
         memory_size (int): size of the past window. if <= 0 memory_size = r
+        TODO: arguments
     """
 
     def __init__(self, in_features, memory_dim, r, memory_size, attn_windowing,