shithub: opus

Info  •  Files  •  Log  •  Branches

Avoid rounding the prediction multiple times

--- a/dnn/train_lpcnet.py

+++ b/dnn/train_lpcnet.py

@@ -94,7 +94,7 @@

 # Note: the LPC predictor output is now calculated by the loop below, this code was

 # for an ealier version that implemented the prediction filter in C

-upred = np.zeros((nb_frames*pcm_chunk_size,), dtype='int16')

+upred = np.zeros((nb_frames*pcm_chunk_size,), dtype='float32')

 # Use 16th order LPC to generate LPC prediction output upred[] and (in

 # mu-law form) pred[]

@@ -136,8 +136,8 @@

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

 # dump models to disk as we go

-checkpoint = ModelCheckpoint('lpcnet9c_384_10_G16_{epoch:02d}.h5')

+checkpoint = ModelCheckpoint('lpcnet14_384_10_G16_{epoch:02d}.h5')

-#model.load_weights('wavenet4f2_30.h5')

+#model.load_weights('lpcnet9b_384_10_G16_01.h5')

 model.compile(optimizer=Adam(0.001, amsgrad=True, decay=5e-5), loss='sparse_categorical_crossentropy', metrics=['sparse_categorical_accuracy'])

 model.fit([in_data, in_exc, features, periods], out_data, batch_size=batch_size, epochs=nb_epochs, validation_split=0.0, callbacks=[checkpoint, lpcnet.Sparsify(2000, 40000, 400, 0.1)])

-- 

⑨