shithub: opus

Info  •  Files  •  Log  •  Branches

Move back to tanh for frame rate network

Swish has lower loss, but doesn't seem to improve quality

--- a/dnn/training_tf2/lpcnet.py

+++ b/dnn/training_tf2/lpcnet.py

@@ -241,14 +241,14 @@

     dec_state2 = Input(shape=(rnn_units2,))

     padding = 'valid' if training else 'same'

-    fconv1 = Conv1D(cond_size, 3, padding=padding, activation='swish', name='feature_conv1')

-    fconv2 = Conv1D(cond_size, 3, padding=padding, activation='swish', name='feature_conv2')

+    fconv1 = Conv1D(cond_size, 3, padding=padding, activation='tanh', name='feature_conv1')

+    fconv2 = Conv1D(cond_size, 3, padding=padding, activation='tanh', name='feature_conv2')

     pembed = Embedding(256, 64, name='embed_pitch')

     cat_feat = Concatenate()([feat, Reshape((-1, 64))(pembed(pitch))])

     cfeat = fconv2(fconv1(cat_feat))

-    fdense1 = Dense(cond_size, activation='swish', name='feature_dense1')

+    fdense1 = Dense(cond_size, activation='tanh', name='feature_dense1')

     fdense2 = Dense(cond_size, activation='tanh', name='feature_dense2')

     if flag_e2e and quantize:

-- 

⑨