Attention convolution padding correction for TF "SAME"

layers/attention.py

@@ -42,7 +42,7 @@ class LocationSensitiveAttention(nn.Module):
         super(LocationSensitiveAttention, self).__init__()
         self.kernel_size = kernel_size
         self.filters = filters
-        padding = int((kernel_size - 1) / 2)
+        padding = [(kernel_size - 1) // 2, (kernel_size - 1) // 2]
         self.loc_conv = nn.Conv1d(
             2,
             filters,

layers/tacotron.py

@@ -136,7 +136,7 @@ class CBHG(nn.Module):
                 padding=[(k - 1) // 2, k // 2],
                 activation=self.relu) for k in range(1, K + 1)
         ])
-        # max pooling of conv bank, padding with nn.functional
+        # max pooling of conv bank, with padding
         # TODO: try average pooling OR larger kernel size
         self.max_pool1d = nn.Sequential(
             nn.ConstantPad1d([0, 1], value=0),