athena.models.vad.vad_marblenet

Module Contents

Classes

TFReflectionPad1d	Tensorflow ReflectionPad1d module.
BNResidualBlock	Tensorflow BN_Residual_Block module.
BNResidualBlocks
VadMarbleNet	implementation of a frame level or segment speech classification

class athena.models.vad.vad_marblenet.TFReflectionPad1d(padding_size, padding_type='CONSTANT', **kwargs)

Bases: tensorflow.keras.layers.Layer

Tensorflow ReflectionPad1d module.

call(x)

Calculate forward propagation. :param x: Input tensor (B, T, C). :type x: Tensor

Returns

Padded tensor (B, T + 2 * padding_size, C).

Return type

Tensor

class athena.models.vad.vad_marblenet.BNResidualBlock(repeat_times, kernel_size, filters, dilation_rate, strides, zoneout_rate, nonlinear_activation, nonlinear_activation_params, is_weight_norm, padding_type='REFLECT', **kwargs)

Bases: tensorflow.keras.layers.Layer

Tensorflow BN_Residual_Block module.

call(x)

Calculate forward propagation. :param x: Input tensor (B, T, C). :type x: Tensor

Returns

Output tensor (B, T, C).

Return type

Tensor

_apply_weightnorm(list_layers)

Try apply weightnorm for all layer in list_layers.

class athena.models.vad.vad_marblenet.BNResidualBlocks(config)

Bases: tensorflow.keras.layers.Layer

default_config

call(x)

Calculate forward propagation. :param x: Input tensor (B, T, C). :type x: Tensor

Returns

Output tensor (B, T, C).

Return type

Tensor

class athena.models.vad.vad_marblenet.VadMarbleNet(data_descriptions, config=None)

Bases: athena.models.base.BaseModel

implementation of a frame level or segment speech classification

default_config

call(samples, training=None)

call model

build_model(data_descriptions)

get_loss(outputs, samples, training=None)

get loss