set teacher model and initialize duration_calculator before training Args: teacher_model: the loaded teacher model

Deep copy x and do spec augmentation then return it Args: x: input feature, T * F 2D num_t_mask: number of time mask

Splice a tensor along the last dimension with context. Example: >>> t = [[[1, 2, 3], >>> [4, 5, 6], >>> [7, 8, 9]]]

splice features according to input_left_context and input_right_context input_left_context: the left features to be spliced, repea

splice features according to input_left_context and input_right_context input_left_context: the left features to be spliced, repea

hard mode can only be used in the synthesis step Args: sigmoid_probs: sigmoid probabilities, shape: [batch, x_steps]

If epoch > 0, return a summary of loss and metrics on dev set and write on Tensorboard Otherwise, just return evaluate loss and metrics

generate summary of learning_rate, loss, metrics, speed and write on Tensorboard

Synthesize acoustic features from the input texts Args: samples: the data source to be synthesized Returns::

Return a list with tensor shape. For each dimension, use tensor.get_shape() first. If not available, use tf.shape().

Synthesize one step frames Args: encoder_output: the encoder output, shape: [batch, x_steps, eunits] memory_

* @return A string representing dt * */

* @return A string representing dt * */

Update the model in 1 epoch

append new predictions and labels

load video list from scp file return a dic

load video list from scp file return a dic

load video list from scp file return a dic