segmentation model wip

segment_data.py

@@ -0,0 +1,11 @@
+import pandas as pd
+
+def fix_csv(collection_name = 'test'):
+    seg_data = pd.read_csv('./outputs/'+collection_name+'.csv',names=['phrase','filename'
+                ,'start_phoneme','end_phoneme','start_time','end_time'])
+    seg_data.to_csv('./outputs/'+collection_name+'.fixed.csv')
+
+
+def segment_data_gen(collection_name = 'test'):
+    # collection_name = 'test'
+    seg_data = pd.read_csv('./outputs/'+collection_name+'.fixed.csv',index_col=0)

segment_model.py

@@ -0,0 +1,108 @@
+from __future__ import absolute_import
+from __future__ import print_function
+import numpy as np
+from keras.models import Model,load_model,model_from_yaml
+from keras.layers import Input,Concatenate,Lambda, BatchNormalization, Dropout
+from keras.layers import Dense, LSTM, Bidirectional, GRU
+from keras.losses import categorical_crossentropy
+from keras.utils import to_categorical
+from keras.optimizers import RMSprop
+from keras.callbacks import TensorBoard, ModelCheckpoint
+from keras import backend as K
+from keras.utils import plot_model
+from speech_tools import create_dir,step_count
+from speech_data import segment_data_gen
+
+
+def accuracy(y_true, y_pred):
+    '''Compute classification accuracy with a fixed threshold on distances.
+    '''
+    return K.mean(K.equal(y_true, K.cast(y_pred > 0.5, y_true.dtype)))
+
+def dense_classifier(processed):
+    conc_proc = Concatenate()(processed)
+    d1 = Dense(64, activation='relu')(conc_proc)
+    # dr1 = Dropout(0.1)(d1)
+    # d2 = Dense(128, activation='relu')(d1)
+    d3 = Dense(8, activation='relu')(d1)
+    # dr2 = Dropout(0.1)(d2)
+    return Dense(2, activation='softmax')(d3)
+
+def segment_model(input_dim):
+    inp = Input(shape=input_dim)
+    # ls0 = LSTM(512, return_sequences=True)(inp)
+    ls1 = LSTM(128, return_sequences=True)(inp)
+    ls2 = LSTM(64, return_sequences=True)(ls1)
+    # ls3 = LSTM(32, return_sequences=True)(ls2)
+    ls4 = LSTM(32)(ls2)
+    d1 = Dense(64, activation='relu')(ls4)
+    d3 = Dense(8, activation='relu')(d1)
+    oup = Dense(2, activation='softmax')(d3)
+    return Model(inp, oup)
+
+def write_model_arch(mod,mod_file):
+    model_f = open(mod_file,'w')
+    model_f.write(mod.to_yaml())
+    model_f.close()
+
+def load_model_arch(mod_file):
+    model_f = open(mod_file,'r')
+    mod = model_from_yaml(model_f.read())
+    model_f.close()
+    return mod
+
+def train_segment(collection_name = 'test'):
+    batch_size = 128
+    model_dir = './models/segment/'+collection_name
+    create_dir(model_dir)
+    log_dir = './logs/segment/'+collection_name
+    create_dir(log_dir)
+    tr_gen_fn = segment_data_gen()
+    tr_gen = tr_gen_fn()
+    input_dim = (n_step, n_features)
+
+    model = segment_model(input_dim)
+    plot_model(model,show_shapes=True, to_file=model_dir+'/model.png')
+
+    tb_cb = TensorBoard(
+        log_dir=log_dir,
+        histogram_freq=1,
+        batch_size=32,
+        write_graph=True,
+        write_grads=True,
+        write_images=True,
+        embeddings_freq=0,
+        embeddings_layer_names=None,
+        embeddings_metadata=None)
+    cp_file_fmt = model_dir+'/siamese_speech_model-{epoch:02d}-epoch-{val_loss:0.2f}\
+-acc.h5'
+
+    cp_cb = ModelCheckpoint(
+        cp_file_fmt,
+        monitor='val_loss',
+        verbose=0,
+        save_best_only=True,
+        save_weights_only=True,
+        mode='auto',
+        period=1)
+    # train
+    rms = RMSprop()
+    model.compile(loss=categorical_crossentropy, optimizer=rms, metrics=[accuracy])
+    write_model_arch(model,model_dir+'/siamese_speech_model_arch.yaml')
+    epoch_n_steps = step_count(n_records,batch_size)
+    model.fit_generator(tr_gen
+                        , epochs=1000
+                        , steps_per_epoch=epoch_n_steps
+                        , validation_data=([te_pairs[:, 0], te_pairs[:, 1]], te_y)
+                        , max_queue_size=32
+                        , callbacks=[tb_cb, cp_cb])
+    model.save(model_dir+'/speech_segment_model-final.h5')
+
+    y_pred = model.predict([te_pairs[:, 0], te_pairs[:, 1]])
+    te_acc = compute_accuracy(te_y, y_pred)
+    print('* Accuracy on test set: %0.2f%%' % (100 * te_acc))
+
+
+
+if __name__ == '__main__':
+    train_segment('test')

speech_model.py

@@ -3,7 +3,8 @@ from __future__ import print_function
 import numpy as np
 from speech_data import read_siamese_tfrecords_generator
 from keras.models import Model,load_model,model_from_yaml
-from keras.layers import Input, Dense, Dropout, LSTM, Lambda, Concatenate, Bidirectional
+from keras.layers import Input,Concatenate,Lambda, BatchNormalization, Dropout
+from keras.layers import Dense, LSTM, Bidirectional, GRU
 from keras.losses import categorical_crossentropy
 from keras.utils import to_categorical
 from keras.optimizers import RMSprop