diff --git a/dnn/dump_data.c b/dnn/dump_data.c
index 1db653d0..1d8f6b3c 100644
--- a/dnn/dump_data.c
+++ b/dnn/dump_data.c
@@ -313,7 +313,7 @@ int main(int argc, char **argv) {
       }
       last_silent = silent;
     }
-    if (count>=5000000 && one_pass_completed) break;
+    if (count*FRAME_SIZE_5MS>=10000000 && one_pass_completed) break;
     if (training && ++gain_change_count > 2821) {
       float tmp;
       speech_gain = pow(10., (-20+(rand()%40))/20.);
diff --git a/dnn/lpcnet.py b/dnn/lpcnet.py
index 532adc8e..6b1adf64 100644
--- a/dnn/lpcnet.py
+++ b/dnn/lpcnet.py
@@ -36,6 +36,7 @@ import numpy as np
 import h5py
 import sys
 
+frame_size = 160
 pcm_bits = 8
 embed_size = 128
 pcm_levels = 2**pcm_bits
@@ -139,7 +140,7 @@ def new_lpcnet_model(rnn_units1=384, rnn_units2=16, nb_used_features = 38, use_g
 
     cfeat = fdense2(fdense1(cfeat))
     
-    rep = Lambda(lambda x: K.repeat_elements(x, 160, 1))
+    rep = Lambda(lambda x: K.repeat_elements(x, frame_size, 1))
 
     if use_gpu:
         rnn = CuDNNGRU(rnn_units1, return_sequences=True, return_state=True, name='gru_a')
@@ -158,6 +159,7 @@ def new_lpcnet_model(rnn_units1=384, rnn_units2=16, nb_used_features = 38, use_g
     model.rnn_units1 = rnn_units1
     model.rnn_units2 = rnn_units2
     model.nb_used_features = nb_used_features
+    model.frame_size = frame_size
 
     encoder = Model([feat, pitch], cfeat)
     
diff --git a/dnn/test_lpcnet.py b/dnn/test_lpcnet.py
index 842ce56a..b7a57c3e 100755
--- a/dnn/test_lpcnet.py
+++ b/dnn/test_lpcnet.py
@@ -47,7 +47,7 @@ model.compile(optimizer='adam', loss='sparse_categorical_crossentropy', metrics=
 
 feature_file = sys.argv[1]
 out_file = sys.argv[2]
-frame_size = 160
+frame_size = model.frame_size
 nb_features = 55
 nb_used_features = model.nb_used_features
 
diff --git a/dnn/train_lpcnet.py b/dnn/train_lpcnet.py
index c3b64ff7..9d1fb181 100755
--- a/dnn/train_lpcnet.py
+++ b/dnn/train_lpcnet.py
@@ -58,7 +58,7 @@ model.summary()
 
 feature_file = sys.argv[1]
 pcm_file = sys.argv[2]     # 16 bit unsigned short PCM samples
-frame_size = 160
+frame_size = model.frame_size
 nb_features = 55
 nb_used_features = model.nb_used_features
 feature_chunk_size = 15
@@ -97,7 +97,7 @@ del sig
 del pred
 
 # dump models to disk as we go
-checkpoint = ModelCheckpoint('lpcnet20c_384_10_G16_{epoch:02d}.h5')
+checkpoint = ModelCheckpoint('lpcnet20g_384_10_G16_{epoch:02d}.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')