shithub: opus

Info  •  Files  •  Log  •  Branches

Add convolution

--- a/dnn/lpcnet.py

+++ b/dnn/lpcnet.py

@@ -23,13 +23,22 @@

     conv1 = Conv1D(16, 7, padding='causal')

     pconv1 = Conv1D(16, 5, padding='same')

     pconv2 = Conv1D(16, 5, padding='same')

+    fconv1 = Conv1D(128, 3, padding='same')

+    fconv2 = Conv1D(32, 3, padding='same')

-    cpcm = conv1(pcm)

-    cpitch = pconv2(pconv1(pitch))

+    if True:

+        cpcm = conv1(pcm)

+        cpitch = pconv2(pconv1(pitch))

+    else:

+        cpcm = pcm

+        cpitch = pitch

+

+    cfeat = fconv2(fconv1(feat))

+

     rep = Lambda(lambda x: K.repeat_elements(x, 160, 1))

     rnn = CuDNNGRU(rnn_units, return_sequences=True)

-    rnn_in = Concatenate()([cpcm, cpitch, rep(feat)])

+    rnn_in = Concatenate()([cpcm, cpitch, rep(cfeat)])

     md = MDense(pcm_levels, activation='softmax')

     ulaw_prob = md(rnn(rnn_in))

--- a/dnn/train_lpcnet.py

+++ b/dnn/train_lpcnet.py

@@ -62,7 +62,7 @@

 # f.create_dataset('data', data=in_data[:50000, :, :])

 # f.create_dataset('feat', data=features[:50000, :, :])

-checkpoint = ModelCheckpoint('lpcnet1e_{epoch:02d}.h5')

+checkpoint = ModelCheckpoint('lpcnet1g_{epoch:02d}.h5')

 #model.load_weights('wavernn1c_01.h5')

 model.compile(optimizer=Adam(0.002, amsgrad=True, decay=2e-4), loss='sparse_categorical_crossentropy', metrics=['sparse_categorical_accuracy'])

-- 

⑨