Conv1D 1x1 is literally a causal convolution with kernel_size = 1

Projection layer to r * num_mels dimensions or num_mels dimensions

Simple class to wrap relu activation function in class for later call.

`namedtuple` storing the state of a `TacotronDecoderCell`. Contains: - `cell_state`: The state of the wrapped `RNNCell` at the previous time

Wrapper for tf LSTM to create Zoneout LSTM Cell inspired by: https://github.com/teganmaharaj/zoneout/blob/master/zoneout_tensorflow.py P

This wrapper reparameterizes a layer by decoupling the weight's magnitude and direction. This speeds up convergence by improving the conditioning o

Custom sampling decoder. Allows for stop token prediction at inference time and returns equivalent loss in training time. Note: Only use this de

Decoder two uni directional LSTM Cells

Embedding class for global conditions.

Encoder convolutional layers used to find local dependencies in inputs characters.

Encoder bidirectional one layer LSTM

Feeds batches of data into queue in a background thread.

Feeds batches of data into queue on a background thread.

Simple class to wrap leaky relu activation function in class for later call.

Impelements Bahdanau-style (cumulative) scoring function. Usually referred to as "hybrid" attention (content-based + location-based) Extends the add

Postnet that takes final decoder output and fine tunes it (using vision on past and future frames)

Two fully connected layers used as an information bottleneck for the attention.

Dilated conv1d + Gated Linear Unit + condition convolutions + residual and skip convolutions Dilated convolution is considered to be the most import

Projection to a scalar and through a sigmoid activation

Tacotron-2 Feature prediction Model.

Tactron 2 Decoder Cell Decodes encoder output and previous mel frames into next r frames Decoder Step i: 1) Prenet to compress last o

Tacotron 2 Encoder Cell Passes inputs through a stack of convolutional layers then through a bidirectional LSTM layer to predict the hidden re

Tacotron-2 Wavenet Vocoder model.

Thin wrapper around CMUDict data. http://www.speech.cs.cmu.edu/cgi-bin/cmudict