# Reg_de_train = tf.train.AdamOptimizer(0.00001).minimize(loss_Reg_de)
train_step = tf.train.AdamOptimizer(0.0003).minimize(loss)
batchSize = 128
src = 'SF1'
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
tS = time.time()
e=[]
saver = tf.train.Saver()
for i in range(10):
# if i>2:
# lamb = 1
for j in range(100):
x_batch,label_batch = nextbatch(trainData, Label, batchSize)
train_step.run(feed_dict = {source: x_batch, label: label_batch})
# Reg_de_train.run(feed_dict={source: x_batch, label: label_batch})
x_Evabatch, label_Evabatch = nextbatch(trainData, Label, batchSize)
eva_loss= sess.run([loss], feed_dict={source: x_Evabatch, label: label_Evabatch})
e.append(eva_loss)
print('epoch %d' %(i))
print('reconstruct loss: %f' %(np.mean(eva_loss)))
for i in range(10):
for j in range(100):
x_batch, label_batch = nextbatch(trainData, Label, batchSize)
de = sess.run(W, feed_dict={source: x_batch, label: label_batch})
Reg_en_train.run(feed_dict={latent: de, label: label_batch})
reg_loss, enacu = sess.run([loss_Reg_en, Acu_Reg_en], feed_dict={latent: de, label: label_batch})
train_MS = tf.train.AdamOptimizer(0.0001).minimize(loss_MS)
train_KL = tf.train.AdamOptimizer(0.00001).minimize(loss_KL)
train_Reg = tf.train.AdamOptimizer(0.0001).minimize(loss_Reg)
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
saver = tf.train.Saver()
time_start = time.time()
batchSize = 128
is_training = True
if is_training:
for epoch in range(15):
for iter in range(1000):
src_batch, label_batch = nextbatch(TrainDataPath, featureSize,
batchSize)
train_MS.run(feed_dict={source: src_batch, label: label_batch})
train_KL.run(feed_dict={source: src_batch, label: label_batch})
ms_value, kl_value = sess.run([loss_MS, loss_KL],
feed_dict={
source: src_batch,
label: label_batch
})
print('epoch: %d' % (epoch))
print('MSE: %f' % (ms_value))
print('KL: %f' % (np.mean(kl_value)))
# for epoch in range(10):
# for iter in range(1000):
# src_batch, label_batch = nextbatch(TrainDataPath, featureSize, batchSize)
# train_Reg.run(feed_dict={source: src_batch, label: label_batch})
trainables = tf.trainable_variables()
loss = tf.nn.l2_loss(recover - source)
train_L2_loss = tf.train.AdamOptimizer(0.0001).minimize(loss)
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
batchSize = 128
is_training = False
saver = tf.train.Saver()
if is_training:
for epoch in range(20):
for iter in range(100):
x = nextbatch(Trainfilepath, segmentLength, batchSize)
train_L2_loss.run(feed_dict={source: x})
loss_value = sess.run(loss, feed_dict={source: x})
print('epoch: %d' % (epoch))
print('loss: %f' % (loss_value))
saver.save(sess, 'ckpt/model.ckpt')
else:
saver.restore(sess, tf.train.latest_checkpoint('ckpt/'))
print('Model restored.')
voice, filename = pickOne(Testfilepath, 'SF1', segmentLength)
output = sess.run(recover, feed_dict={source: voice})
output = output.reshape([-1])
scipy.io.wavfile.write(filename + '.wav', sampleRate, output)
loss = tf.reduce_mean(
tf.nn.softmax_cross_entropy_with_logits(labels=label, logits=y))
train_step = tf.train.AdamOptimizer(0.0001).minimize(loss)
correct_prediction = tf.equal(tf.argmax(y, 1), tf.argmax(label, 1))
accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
batchSize = 128
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
for i in range(20):
for j in range(200):
x_batch, y_batch = nextbatch(TrainDataPath, segmentLength, batchSize)
train_step.run(feed_dict={source: x_batch, label: y_batch})
x_Evabatch, y_Evabatch = nextbatch(TestDataPath, segmentLength, batchSize)
tr_acu = sess.run(accuracy, feed_dict={source: x_batch, label: y_batch})
te_acu = sess.run(accuracy,
feed_dict={
source: x_Evabatch,
label: y_Evabatch
})
print('epoch %d' % (i))
print('Train accuracy: %f' % (tr_acu))
print('Test accuracy: %f' % (te_acu))
tE = time.time()
print("training time: %f" % (tE - tS))
train_KL = tf.train.AdamOptimizer(0.00001).minimize(KL)
train_Regloss = tf.train.AdamOptimizer(0.0001).minimize(speakReg_loss)
sess = tf.InteractiveSession()
tf.global_variables_initializer().run()
tS = time.time()
saver = tf.train.Saver()
batchSize = 64
is_training = False
if is_training:
for epoch in range(5):
for iter in range(100):
x_batch, label_batch = nextbatch(TrainDataPath, segmentLength,
batchSize)
train_re_loss.run(feed_dict={source: x_batch, label: label_batch})
train_trans_loss.run(feed_dict={
source: x_batch,
label: label_batch
})
train_KL.run(feed_dict={source: x_batch, label: label_batch})
loss_value, loss_KL = sess.run([recover_loss, KL],
feed_dict={
source: x_batch,
label: label_batch
})
print('epoch: %d' % (epoch))
print('l2_loss: %f' % (loss_value))
saver.save(sess, 'ckpt/model')
# for epoch in range(50):