Make embedding dimension configurable

hparams.py

@@ -20,6 +20,7 @@ hparams = tf.contrib.training.HParams(
   # Model:
   # TODO: add more configurable hparams
   outputs_per_step=5,
+  embedding_dim=256,
 
   # Training:
   batch_size=32,

models/tacotron.py

@@ -39,7 +39,7 @@ class Tacotron():
 
       # Embeddings
       embedding_table = tf.get_variable(
-        'embedding', [len(symbols), 256], dtype=tf.float32,
+        'embedding', [len(symbols), hp.embedding_dim], dtype=tf.float32,
         initializer=tf.truncated_normal_initializer(stddev=0.5))
       embedded_inputs = tf.nn.embedding_lookup(embedding_table, inputs)           # [N, T_in, 256]