Pass mask instead of length to model

layers/attention.py

@@ -105,13 +105,7 @@ class AttentionRNNCell(nn.Module):
                 'b' (Bahdanau) or 'ls' (Location Sensitive)."
                                .format(align_model))
 
-    def forward(self,
-                memory,
-                context,
-                rnn_state,
-                annots,
-                atten,
-                annot_lens=None):
+    def forward(self, memory, context, rnn_state, annots, atten, mask):
         """
         Shapes:
             - memory: (batch, 1, dim) or (batch, dim)
@@ -119,7 +113,7 @@ class AttentionRNNCell(nn.Module):
             - rnn_state: (batch, out_dim)
             - annots: (batch, max_time, annot_dim)
             - atten: (batch, 2, max_time)
-            - annot_lens: (batch,)
+            - mask: (batch,)
         """
         # Concat input query and previous context context
         rnn_input = torch.cat((memory, context), -1)
@@ -133,8 +127,7 @@ class AttentionRNNCell(nn.Module):
             alignment = self.alignment_model(annots, rnn_output)
         else:
             alignment = self.alignment_model(annots, rnn_output, atten)
-        if annot_lens is not None:
-            mask = sequence_mask(annot_lens)
+        if mask is not None:
             mask = mask.view(memory.size(0), -1)
             alignment.masked_fill_(1 - mask, -float("inf"))
         # Normalize context weight

layers/tacotron.py

@@ -189,7 +189,7 @@ class CBHG(nn.Module):
             x = highway(x)
         # (B, T_in, hid_features*2)
         # TODO: replace GRU with convolution as in Deep Voice 3
-        # self.gru.flatten_parameters()
+        self.gru.flatten_parameters()
         outputs, _ = self.gru(x)
         return outputs
 
@@ -268,7 +268,7 @@ class Decoder(nn.Module):
         self.proj_to_mel = nn.Linear(256, memory_dim * r)
         self.stopnet = StopNet(r, memory_dim)
 
-    def forward(self, inputs, memory=None, input_lens=None):
+    def forward(self, inputs, memory=None, mask=None):
         """
         Decoder forward step.
 
@@ -280,7 +280,7 @@ class Decoder(nn.Module):
             memory (None): Decoder memory (autoregression. If None (at eval-time),
               decoder outputs are used as decoder inputs. If None, it uses the last
               output as the input.
-            input_lens (None): Time length of each input in batch.
+            mask (None): Attention mask for sequence padding.
 
         Shapes:
             - inputs: batch x time x encoder_out_dim
@@ -332,7 +332,7 @@ class Decoder(nn.Module):
                 (attention.unsqueeze(1), attention_cum.unsqueeze(1)), dim=1)
             attention_rnn_hidden, current_context_vec, attention = self.attention_rnn(
                 processed_memory, current_context_vec, attention_rnn_hidden,
-                inputs, attention_cat, input_lens)
+                inputs, attention_cat, mask)
             attention_cum += attention
             # Concat RNN output and attention context vector
             decoder_input = self.project_to_decoder_in(

models/tacotron.py

@@ -25,14 +25,14 @@ class Tacotron(nn.Module):
         self.postnet = PostCBHG(mel_dim)
         self.last_linear = nn.Linear(256, linear_dim)
 
-    def forward(self, characters, mel_specs=None, text_lens=None):
+    def forward(self, characters, mel_specs=None, mask=None):
         B = characters.size(0)
         inputs = self.embedding(characters)
         # batch x time x dim
         encoder_outputs = self.encoder(inputs)
         # batch x time x dim*r
         mel_outputs, alignments, stop_tokens = self.decoder(
-            encoder_outputs, mel_specs, text_lens)
+            encoder_outputs, mel_specs, mask)
         # Reshape
         # batch x time x dim
         mel_outputs = mel_outputs.view(B, -1, self.mel_dim)