def _gta_forward(self, inp, tar, stop_prob, xvectors, training):
#add xvector
tar_inp = tar[:, :-1]
tar_real = tar[:, 1:]
tar_stop_prob = stop_prob[:, 1:]
mel_len = int(tf.shape(tar_inp)[1])
tar_mel = tar_inp[:, 0::self.r, :]
with tf.GradientTape() as tape:
#add xvector in inputs
model_out = self.__call__(inputs=inp,
targets=tar_mel,
xvectors=xvectors,
training=training)
loss, loss_vals = weighted_sum_losses(
(tar_real, tar_stop_prob, tar_real),
(model_out['final_output'][:, :mel_len, :],
model_out['stop_prob'][:, :mel_len, :],
model_out['mel_linear'][:, :mel_len, :]), self.loss,
self.loss_weights)
model_out.update({'loss': loss})
model_out.update({
'losses': {
'output': loss_vals[0],
'stop_prob': loss_vals[1],
'mel_linear': loss_vals[2]
}
})
model_out.update({'reduced_target': tar_mel})
return model_out, tape
def _val_step(self, input_sequence, target_sequence, target_durations):
target_durations = tf.expand_dims(target_durations, -1)
mel_len = int(tf.shape(target_sequence)[1])
model_out = self.__call__(input_sequence, target_durations, training=False)
loss, loss_vals = weighted_sum_losses((target_sequence,
target_durations),
(model_out['mel'][:, :mel_len, :],
model_out['duration']),
self.loss,
self.loss_weights)
model_out.update({'loss': loss})
model_out.update({'losses': {'mel': loss_vals[0], 'duration': loss_vals[1]}})
return model_out
def _train_step(self, input_sequence, target_sequence, target_durations):
target_durations = tf.expand_dims(target_durations, -1)
mel_len = int(tf.shape(target_sequence)[1])
with tf.GradientTape() as tape:
model_out = self.__call__(input_sequence, target_durations, training=True)
loss, loss_vals = weighted_sum_losses((target_sequence,
target_durations),
(model_out['mel'][:, :mel_len, :],
model_out['duration']),
self.loss,
self.loss_weights)
model_out.update({'loss': loss})
model_out.update({'losses': {'mel': loss_vals[0], 'duration': loss_vals[1]}})
gradients = tape.gradient(model_out['loss'], self.trainable_variables)
self.optimizer.apply_gradients(zip(gradients, self.trainable_variables))
return model_out
