import tflearn
import utils
learning_rate = 0.01
training_iters = 300000 # steps
batch_size = 50
width = 20 # mfcc features
height = 80 # (max) length of utterance
# classes = 10 # digits
speakers = ['adrian', 'zhanet']
number_classes = len(speakers)
batch = utils.mfcc_batch_generator(speakers,
batch_size=batch_size,
utterance_len=height)
# Network building
# net = tflearn.input_data([None, width, height])
net = tflearn.input_data([None, height, width])
net = tflearn.lstm(net, 800, dropout=0.5)
net = tflearn.fully_connected(net, number_classes, activation='softmax')
net = tflearn.regression(net,
optimizer='adam',
learning_rate=learning_rate,
loss='categorical_crossentropy')
model = tflearn.DNN(net, tensorboard_verbose=3)
# model.load('models/lstm.model_80_800.0.1')
# model.load('models/tflearn.lstm.model_48_128_4') # 86.32%
# model.load('models/tflearn.lstm.model_48_128') # 81.79%
# model.load('models/tflearn.lstm.model_80_800') # 81.18%
# ZIPPED MFCC NETS
# model.load('models/t2_lstm.model_2_80_128_3') # 31.25
# model.load('models/t2_lstm.model_2_48_128') # 27.27%
# model.load('models/t2_lstm.model_2_48_128_4') # 36.36
# model.load('models/tflearn.lstm.model_2_48_128_4') # 49.95%
# model.load('models/tflearn.lstm.model_2_48_128') # 31.81%
# model.load('models/tflearn.lstm.model_2_80_128_3') # 50.04%
batch = mfcc_batch_generator(speakers, batch_size = batch_size, utterance_len = height, path = path)
count = 0
correct = 0
for feats, labels in batch:
_y = model.predict(feats)
for i, val in enumerate(_y):
real_label = one_hot_to_item(labels[i], speakers)
label = one_hot_to_item(val, speakers)
if real_label == label:
correct += 1
count += 1