shithub: opus

Info  •  Files  •  Log  •  Branches

add convolution

--- a/dnn/lpcnet.py

+++ b/dnn/lpcnet.py

@@ -16,15 +16,22 @@

 def new_wavernn_model():

-    pcm = Input(shape=(None, 2))

+    pcm = Input(shape=(None, 1))

+    pitch = Input(shape=(None, 1))

     feat = Input(shape=(None, nb_used_features))

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

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

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

+

+    cpcm = conv1(pcm)

+    cpitch = pconv2(pconv1(pitch))

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

     rnn = CuDNNGRU(rnn_units, return_sequences=True)

-    rnn_in = Concatenate()([pcm, rep(feat)])

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

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

     ulaw_prob = md(rnn(rnn_in))

-    model = Model([pcm, feat], ulaw_prob)

+    model = Model([pcm, pitch, feat], ulaw_prob)

     return model

--- a/dnn/train_lpcnet.py

+++ b/dnn/train_lpcnet.py

@@ -56,7 +56,7 @@

 features = features[:, :, :nb_used_features]

-in_data = np.concatenate([in_data, in_pitch], axis=-1)

+#in_data = np.concatenate([in_data, in_pitch], axis=-1)

 #with h5py.File('in_data.h5', 'w') as f:

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

@@ -66,4 +66,4 @@

 #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'])

-model.fit([in_data, features], out_data, batch_size=batch_size, epochs=30, validation_split=0.2, callbacks=[checkpoint])

+model.fit([in_data, in_pitch, features], out_data, batch_size=batch_size, epochs=30, validation_split=0.2, callbacks=[checkpoint])

-- 

⑨