def model(self):
"""
The main model function, takes and returns tensors.
Defined in modules.
"""
with tf.variable_scope('encoder') as scope:
self.content_embedding_1 = modules_autovc.content_encoder(
self.input_placeholder, self.speaker_onehot_labels,
self.is_train)
with tf.variable_scope('decoder') as scope:
self.output_1 = modules_autovc.decoder(
self.content_embedding_1, self.speaker_onehot_labels_1,
self.is_train)
with tf.variable_scope('post_net') as scope:
self.residual = modules_autovc.post_net(self.output_1,
self.is_train)
self.output = self.output_1 + self.residual
with tf.variable_scope('encoder') as scope:
scope.reuse_variables()
self.content_embedding_2 = modules_autovc.content_encoder(
self.output, self.speaker_onehot_labels, self.is_train)
with tf.variable_scope('stft_encoder') as scope:
self.content_embedding_stft = modules_SDN.content_encoder_stft(
self.stft_placeholder, self.is_train)
with tf.variable_scope('stft_decoder') as scope:
self.output_stft_1 = modules_autovc.decoder(
self.content_embedding_stft, self.speaker_onehot_labels_1,
self.is_train)
with tf.variable_scope('stft_post_net') as scope:
self.residual_stft = modules_autovc.post_net(
self.output_stft_1, self.is_train)
self.output_stft = self.output_stft_1 + self.residual_stft
with tf.variable_scope('encoder') as scope:
scope.reuse_variables()
self.content_embedding_stft_2 = modules_autovc.content_encoder(
self.output_stft, self.speaker_onehot_labels, self.is_train)
with tf.variable_scope('F0_Model') as scope:
self.f0 = modules_SDN.enc_dec_f0(self.stft_placeholder,
self.output_stft[:, :, :-4],
self.output_stft[:, :, -4:],
self.is_train)
with tf.variable_scope('Vuv_Model') as scope:
self.vuv = modules_SDN.enc_dec_vuv(self.stft_placeholder,
self.output_stft[:, :, :-4],
self.output_stft[:, :, -4:],
self.f0, self.is_train)
def model(self):
"""
The main model function, takes and returns tensors.
Defined in modules.
"""
with tf.variable_scope('encoder') as scope:
self.content_embedding_1 = modules.content_encoder(self.input_placeholder, self.speaker_labels, self.is_train)
with tf.variable_scope('decoder') as scope:
self.output_1 = modules.decoder(self.content_embedding_1, self.speaker_labels_1, self.is_train)
with tf.variable_scope('post_net') as scope:
self.residual = modules.post_net(self.output_1, self.is_train)
self.output = self.output_1 + self.residual
with tf.variable_scope('encoder') as scope:
scope.reuse_variables()
self.content_embedding_2 = modules.content_encoder(self.output, self.speaker_labels_1, self.is_train)
def model(self):
"""
The main model function, takes and returns tensors.
Defined in modules.
"""
with tf.variable_scope('encoder') as scope:
self.content_embedding_1 = modules.content_encoder(
self.input_placeholder, self.speaker_onehot_labels,
self.is_train)
with tf.variable_scope('decoder') as scope:
self.output_1 = modules.decoder(self.content_embedding_1,
self.speaker_onehot_labels_1,
self.is_train)
with tf.variable_scope('post_net') as scope:
self.residual = modules.post_net(self.output_1, self.is_train)
self.output = self.output_1 + self.residual
with tf.variable_scope('encoder') as scope:
scope.reuse_variables()
self.content_embedding_2 = modules.content_encoder(
self.output, self.speaker_onehot_labels, self.is_train)
with tf.variable_scope('F0_Model') as scope:
self.f0 = modules_notes.f0_model(self.content_embedding_1,
self.notes_placeholder,
self.speaker_onehot_labels_1,
self.is_train)
self.f0 = (self.f0 +
self.notes_placeholder[:, :, 0:1]) * tf.clip_by_value(
self.notes_placeholder[:, :, 0:1] * 500,
clip_value_min=0,
clip_value_max=1)
with tf.variable_scope('Vuv_Model') as scope:
self.vuv = modules_notes.vuv_model(self.output,
self.notes_placeholder, self.f0,
self.is_train)