ref: 2d74d3189c04f3420994b4d02fb482e75bd85957
parent: c381db5688af77807561d3f525ef449c0bc0a2a3
author: Jean-Marc Valin <jmvalin@jmvalin.ca>
date: Tue Oct 2 14:26:42 EDT 2018
...
--- a/dnn/lpcnet.py
+++ b/dnn/lpcnet.py
@@ -10,7 +10,7 @@
import h5py
import sys
-rnn_units=512
+rnn_units=128
pcm_bits = 8
embed_size = 128
pcm_levels = 2**pcm_bits
--- a/dnn/test_wavenet_audio.py
+++ b/dnn/test_wavenet_audio.py
@@ -29,7 +29,7 @@
feature_file = sys.argv[2]
frame_size = 160
nb_features = 55
-nb_used_features = wavenet.nb_used_features
+nb_used_features = lpcnet.nb_used_features
feature_chunk_size = 15
pcm_chunk_size = frame_size*feature_chunk_size
@@ -66,7 +66,7 @@
out_data = np.reshape(data, (nb_frames*pcm_chunk_size, 1))
-model.load_weights('wavenet4b_30.h5')+model.load_weights('wavenet4d2_203.h5')order = 16
@@ -92,8 +92,8 @@
p, state = dec.predict([fexc, iexc, cfeat[:, fr:fr+1, :], state])
#p = p*p
#p = p/(1e-18 + np.sum(p))
- p = np.maximum(p-0.001, 0)
- p = p/(1e-5 + np.sum(p))
+ p = np.maximum(p-0.001, 0).astype('float64')+ p = p/(1e-8 + np.sum(p))
iexc[0, 0, 0] = np.argmax(np.random.multinomial(1, p[0,0,:], 1))
pcm[f*frame_size + i, 0] = pred + 32768*ulaw2lin(iexc[0, 0, 0]-128)
--- a/dnn/train_wavenet_audio.py
+++ b/dnn/train_wavenet_audio.py
@@ -31,7 +31,7 @@
pcm_file = sys.argv[4]
frame_size = 160
nb_features = 55
-nb_used_features = wavenet.nb_used_features
+nb_used_features = lpcnet.nb_used_features
feature_chunk_size = 15
pcm_chunk_size = frame_size*feature_chunk_size
@@ -46,7 +46,7 @@
features = features[:nb_frames*feature_chunk_size*nb_features]
in_data = np.concatenate([data[0:1], data[:-1]]);
-noise = np.concatenate([np.zeros((len(data)*2//5)), np.random.randint(-1, 1, len(data)*3//5)])
+noise = np.concatenate([np.zeros((len(data)*1//5)), np.random.randint(-3, 3, len(data)*1//5), np.random.randint(-2, 2, len(data)*1//5), np.random.randint(-1, 1, len(data)*2//5)])
in_data = in_data + noise
in_data = np.maximum(-127, np.minimum(127, in_data))
@@ -58,8 +58,8 @@
pred_in = 32768.*ulaw2lin(in_data)
for i in range(2, nb_frames*feature_chunk_size):
upred[i*frame_size:(i+1)*frame_size] = 0
- if i % 100000 == 0:
- print(i)
+ #if i % 100000 == 0:
+ # print(i)
for k in range(16):
upred[i*frame_size:(i+1)*frame_size] = upred[i*frame_size:(i+1)*frame_size] - \
pred_in[i*frame_size-k:(i+1)*frame_size-k]*features[i, nb_features-16+k]
@@ -103,7 +103,7 @@
# f.create_dataset('data', data=in_data[:50000, :, :]) # f.create_dataset('feat', data=features[:50000, :, :])-checkpoint = ModelCheckpoint('wavenet4b_{epoch:02d}.h5')+checkpoint = ModelCheckpoint('wavenet4d3_{epoch:02d}.h5') #model.load_weights('wavernn1c_01.h5')model.compile(optimizer=Adam(0.001, amsgrad=True, decay=2e-4), loss='sparse_categorical_crossentropy', metrics=['sparse_categorical_accuracy'])
--
⑨