fixed dropout and prenet bug

analyze.py

@@ -0,0 +1 @@
+# visualisation tools for mimic2 

hparams.py

@@ -34,7 +34,7 @@ hparams = tf.contrib.training.HParams(
     adam_beta2=0.999,
     initial_learning_rate=0.0015,
     learning_rate_decay_halflife=100000,
-    use_cmudict=True,   # Use CMUDict during training to learn pronunciation of ARPAbet phonemes
+    use_cmudict=False,   # Use CMUDict during training to learn pronunciation of ARPAbet phonemes
 
     # Eval:
     max_iters=200,

models/modules.py

@@ -8,7 +8,7 @@ def prenet(inputs, is_training, layer_sizes=[256, 128], scope=None):
   with tf.variable_scope(scope or 'prenet'):
     for i, size in enumerate(layer_sizes):
       dense = tf.layers.dense(x, units=size, activation=tf.nn.relu, name='dense_%d' % (i+1))
-      x = tf.layers.dropout(dense, rate=drop_rate, training=True, name='dropout_%d' % (i+1))
+      x = tf.layers.dropout(dense, rate=drop_rate, training=is_training, name='dropout_%d' % (i+1))
   return x
 
 

models/tacotron.py

@@ -49,11 +49,13 @@ class Tacotron():
 
       # Attention
       attention_cell = AttentionWrapper(
-        DecoderPrenetWrapper(GRUCell(256), is_training),
+        GRUCell(256),
         LocationSensitiveAttention(256, encoder_outputs),
         alignment_history=True,
         output_attention=False)                                                  # [N, T_in, 256]
 
+      attention_cell = DecoderPrenetWrapper(attention_cell, is_training)
+
       # Concatenate attention context vector and RNN cell output into a 512D vector.
       concat_cell = ConcatOutputAndAttentionWrapper(attention_cell)              # [N, T_in, 512]