malar
/
speech-scoring
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

1. fixed softmax output and overfit the model for small sample 

2. updated to run on complete data
master
Malar Kannan 2017-12-12 11:38:27 +05:30
parent cc4fbe45b9
commit f44665e9b2
2 changed files with 24 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
segment_data.py
View File

@ -258,7 +258,7 @@ if __name__ == '__main__':
# plot_segments('story_test_segments') # plot_segments('story_test_segments')
# fix_csv('story_words') # fix_csv('story_words')
# pass # pass
create_segments_tfrecords('story_words.3', sample_count=3,train_test_ratio=0.33) create_segments_tfrecords('story_words.30', sample_count=36,train_test_ratio=0.1)
# record_generator,input_data,output_data,copy_read_consts = read_segments_tfrecords_generator('story_test') # record_generator,input_data,output_data,copy_read_consts = read_segments_tfrecords_generator('story_test')
# tr_gen = record_generator() # tr_gen = record_generator()
# for i in tr_gen: # for i in tr_gen:

42
segment_model.py
View File

@ -7,7 +7,7 @@ from keras.layers import Dense,Conv2D, LSTM, Bidirectional, GRU
from keras.layers import BatchNormalization,Activation from keras.layers import BatchNormalization,Activation
from keras.losses import categorical_crossentropy from keras.losses import categorical_crossentropy
from keras.utils import to_categorical from keras.utils import to_categorical
from keras.optimizers import RMSprop from keras.optimizers import RMSprop,Adadelta,Adagrad,Adam,Nadam
from keras.callbacks import TensorBoard, ModelCheckpoint from keras.callbacks import TensorBoard, ModelCheckpoint
from keras import backend as K from keras import backend as K
from keras.utils import plot_model from keras.utils import plot_model
@ -50,20 +50,24 @@ def segment_model(input_dim):
def simple_segment_model(input_dim): def simple_segment_model(input_dim):
inp = Input(shape=input_dim) inp = Input(shape=input_dim)
b_gr1 = Bidirectional(GRU(256, return_sequences=True),merge_mode='sum')(inp) b_gr1 = Bidirectional(LSTM(32, return_sequences=True))(inp)
b_gr1 = Bidirectional(LSTM(16, return_sequences=True),merge_mode='sum')(b_gr1)
b_gr1 = LSTM(1, return_sequences=True,activation='softmax')(b_gr1)
# b_gr1 = LSTM(4, return_sequences=True)(b_gr1)
# b_gr1 = LSTM(2, return_sequences=True)(b_gr1)
# bn_b_gr1 = BatchNormalization(momentum=0.98)(b_gr1) # bn_b_gr1 = BatchNormalization(momentum=0.98)(b_gr1)
b_gr2 = Bidirectional(GRU(64, return_sequences=True),merge_mode='sum')(b_gr1) # b_gr2 = GRU(64, return_sequences=True)(b_gr1)
# bn_b_gr2 = BatchNormalization(momentum=0.98)(b_gr2) # bn_b_gr2 = BatchNormalization(momentum=0.98)(b_gr2)
d1 = Dense(32)(b_gr2) # d1 = Dense(32)(b_gr2)
bn_d1 = BatchNormalization(momentum=0.98)(d1) # bn_d1 = BatchNormalization(momentum=0.98)(d1)
bn_da1 = Activation('relu')(bn_d1) # bn_da1 = Activation('relu')(bn_d1)
d2 = Dense(8)(bn_da1) # d2 = Dense(8)(bn_da1)
bn_d2 = BatchNormalization(momentum=0.98)(d2) # bn_d2 = BatchNormalization(momentum=0.98)(d2)
bn_da2 = Activation('relu')(bn_d2) # bn_da2 = Activation('relu')(bn_d2)
d3 = Dense(1)(bn_da2) # d3 = Dense(1)(b_gr1)
bn_d3 = BatchNormalization(momentum=0.98)(d3) # # bn_d3 = BatchNormalization(momentum=0.98)(d3)
bn_da3 = Activation('softmax')(bn_d3) # bn_da3 = Activation('softmax')(d3)
oup = Reshape(target_shape=(input_dim[0],))(bn_da3) oup = Reshape(target_shape=(input_dim[0],))(b_gr1)
return Model(inp, oup) return Model(inp, oup)
def write_model_arch(mod,mod_file): def write_model_arch(mod,mod_file):
@ -79,13 +83,13 @@ def load_model_arch(mod_file):
def train_segment(collection_name = 'test',resume_weights='',initial_epoch=0): def train_segment(collection_name = 'test',resume_weights='',initial_epoch=0):
# collection_name = 'story_test' # collection_name = 'story_test'
# batch_size = 32 batch_size = 128
batch_size = 1 # batch_size = 4
model_dir = './models/segment/'+collection_name model_dir = './models/segment/'+collection_name
create_dir(model_dir) create_dir(model_dir)
log_dir = './logs/segment/'+collection_name log_dir = './logs/segment/'+collection_name
create_dir(log_dir) create_dir(log_dir)
tr_gen_fn,te_x,te_y,copy_read_consts = read_segments_tfrecords_generator(collection_name,batch_size,batch_size) tr_gen_fn,te_x,te_y,copy_read_consts = read_segments_tfrecords_generator(collection_name,batch_size,2*batch_size)
tr_gen = tr_gen_fn() tr_gen = tr_gen_fn()
n_step,n_features,n_records = copy_read_consts(model_dir) n_step,n_features,n_records = copy_read_consts(model_dir)
input_dim = (n_step, n_features) input_dim = (n_step, n_features)
@ -115,8 +119,8 @@ def train_segment(collection_name = 'test',resume_weights='',initial_epoch=0):
mode='auto', mode='auto',
period=1) period=1)
# train # train
rms = RMSprop() opt = RMSprop()
model.compile(loss=categorical_crossentropy, optimizer=rms, metrics=[accuracy]) model.compile(loss=categorical_crossentropy, optimizer=opt, metrics=[accuracy])
write_model_arch(model,model_dir+'/speech_segment_model_arch.yaml') write_model_arch(model,model_dir+'/speech_segment_model_arch.yaml')
epoch_n_steps = step_count(n_records,batch_size) epoch_n_steps = step_count(n_records,batch_size)
if resume_weights != '': if resume_weights != '':
@ -137,4 +141,4 @@ def train_segment(collection_name = 'test',resume_weights='',initial_epoch=0):
if __name__ == '__main__': if __name__ == '__main__':
# pass # pass
train_segment('story_words.3')#,'./models/segment/story_phrases.1000/speech_segment_model-final.h5',1001) train_segment('story_words')#,'./models/segment/story_phrases.1000/speech_segment_model-final.h5',1001)