paddlespeech.t2s.modules.tacotron2.attentions module

Attention modules for RNN.

class paddlespeech.t2s.modules.tacotron2.attentions.AttForward(eprojs, dunits, att_dim, aconv_chans, aconv_filts)

Bases: Layer

Forward attention module. Reference ---------- Forward attention in sequence-to-sequence acoustic modeling for speech synthesis

(https://arxiv.org/pdf/1807.06736.pdf)

Args:

eprojs (int):

projection-units of encoder

dunits (int):

units of decoder

att_dim (int):

attention dimension

aconv_chans (int):

channels of attention convolution

aconv_filts (int):

filter size of attention convolution

Methods

__call__(*inputs, **kwargs)	Call self as a function.
add_parameter(name, parameter)	Adds a Parameter instance.
add_sublayer(name, sublayer)	Adds a sub Layer instance.
apply(fn)	Applies fn recursively to every sublayer (as returned by .sublayers()) as well as self.
buffers([include_sublayers])	Returns a list of all buffers from current layer and its sub-layers.
children()	Returns an iterator over immediate children layers.
clear_gradients()	Clear the gradients of all parameters for this layer.
create_parameter(shape[, attr, dtype, ...])	Create parameters for this layer.
create_tensor([name, persistable, dtype])	Create Tensor for this layer.
create_variable([name, persistable, dtype])	Create Tensor for this layer.
eval()	Sets this Layer and all its sublayers to evaluation mode.
extra_repr()	Extra representation of this layer, you can have custom implementation of your own layer.
forward(enc_hs_pad, enc_hs_len, dec_z, att_prev)	Calculate AttForward forward propagation.
full_name()	Full name for this layer, composed by name_scope + "/" + MyLayer.__class__.__name__
load_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
named_buffers([prefix, include_sublayers])	Returns an iterator over all buffers in the Layer, yielding tuple of name and Tensor.
named_children()	Returns an iterator over immediate children layers, yielding both the name of the layer as well as the layer itself.
named_parameters([prefix, include_sublayers])	Returns an iterator over all parameters in the Layer, yielding tuple of name and parameter.
named_sublayers([prefix, include_self, ...])	Returns an iterator over all sublayers in the Layer, yielding tuple of name and sublayer.
parameters([include_sublayers])	Returns a list of all Parameters from current layer and its sub-layers.
register_buffer(name, tensor[, persistable])	Registers a tensor as buffer into the layer.
register_forward_post_hook(hook)	Register a forward post-hook for Layer.
register_forward_pre_hook(hook)	Register a forward pre-hook for Layer.
reset()	reset states
set_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
set_state_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
state_dict([destination, include_sublayers, ...])	Get all parameters and persistable buffers of current layer and its sub-layers.
sublayers([include_self])	Returns a list of sub layers.
to([device, dtype, blocking])	Cast the parameters and buffers of Layer by the give device, dtype and blocking.
to_static_state_dict([destination, ...])	Get all parameters and buffers of current layer and its sub-layers.
train()	Sets this Layer and all its sublayers to training mode.

backward
register_state_dict_hook

forward(enc_hs_pad, enc_hs_len, dec_z, att_prev, scaling=1.0, last_attended_idx=None, backward_window=1, forward_window=3)

Calculate AttForward forward propagation.

Args:

enc_hs_pad(Tensor):

padded encoder hidden state (B, T_max, D_enc)

enc_hs_len(list):

padded encoder hidden state length (B,)

dec_z(Tensor):

decoder hidden state (B, D_dec)

att_prev(Tensor):

attention weights of previous step (B, T_max)

scaling(float, optional):

scaling parameter before applying softmax (Default value = 1.0)

last_attended_idx(int, optional):

index of the inputs of the last attended (Default value = None)

backward_window(int, optional):

backward window size in attention constraint (Default value = 1)

forward_window(int, optional):

(Default value = 3)

Returns:

Tensor:

attention weighted encoder state (B, D_enc)

Tensor:

previous attention weights (B, T_max)

reset()

reset states

class paddlespeech.t2s.modules.tacotron2.attentions.AttForwardTA(eunits, dunits, att_dim, aconv_chans, aconv_filts, odim)

Bases: Layer

Forward attention with transition agent module. Reference:

Forward attention in sequence-to-sequence acoustic modeling for speech synthesis

(https://arxiv.org/pdf/1807.06736.pdf)

Args:

eunits (int):

units of encoder

dunits (int):

units of decoder

att_dim (int):

attention dimension

aconv_chans (int):

channels of attention convolution

aconv_filts (int):

filter size of attention convolution

odim (int):

output dimension

Methods

__call__(*inputs, **kwargs)	Call self as a function.
add_parameter(name, parameter)	Adds a Parameter instance.
add_sublayer(name, sublayer)	Adds a sub Layer instance.
apply(fn)	Applies fn recursively to every sublayer (as returned by .sublayers()) as well as self.
buffers([include_sublayers])	Returns a list of all buffers from current layer and its sub-layers.
children()	Returns an iterator over immediate children layers.
clear_gradients()	Clear the gradients of all parameters for this layer.
create_parameter(shape[, attr, dtype, ...])	Create parameters for this layer.
create_tensor([name, persistable, dtype])	Create Tensor for this layer.
create_variable([name, persistable, dtype])	Create Tensor for this layer.
eval()	Sets this Layer and all its sublayers to evaluation mode.
extra_repr()	Extra representation of this layer, you can have custom implementation of your own layer.
forward(enc_hs_pad, enc_hs_len, dec_z, ...)	Calculate AttForwardTA forward propagation.
full_name()	Full name for this layer, composed by name_scope + "/" + MyLayer.__class__.__name__
load_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
named_buffers([prefix, include_sublayers])	Returns an iterator over all buffers in the Layer, yielding tuple of name and Tensor.
named_children()	Returns an iterator over immediate children layers, yielding both the name of the layer as well as the layer itself.
named_parameters([prefix, include_sublayers])	Returns an iterator over all parameters in the Layer, yielding tuple of name and parameter.
named_sublayers([prefix, include_self, ...])	Returns an iterator over all sublayers in the Layer, yielding tuple of name and sublayer.
parameters([include_sublayers])	Returns a list of all Parameters from current layer and its sub-layers.
register_buffer(name, tensor[, persistable])	Registers a tensor as buffer into the layer.
register_forward_post_hook(hook)	Register a forward post-hook for Layer.
register_forward_pre_hook(hook)	Register a forward pre-hook for Layer.
set_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
set_state_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
state_dict([destination, include_sublayers, ...])	Get all parameters and persistable buffers of current layer and its sub-layers.
sublayers([include_self])	Returns a list of sub layers.
to([device, dtype, blocking])	Cast the parameters and buffers of Layer by the give device, dtype and blocking.
to_static_state_dict([destination, ...])	Get all parameters and buffers of current layer and its sub-layers.
train()	Sets this Layer and all its sublayers to training mode.

backward
register_state_dict_hook
reset

forward(enc_hs_pad, enc_hs_len, dec_z, att_prev, out_prev, scaling=1.0, last_attended_idx=None, backward_window=1, forward_window=3)

Calculate AttForwardTA forward propagation.

Args:

enc_hs_pad(Tensor):

padded encoder hidden state (B, Tmax, eunits)

enc_hs_len(list Tensor):

padded encoder hidden state length (B,)

dec_z(Tensor):

decoder hidden state (B, dunits)

att_prev(Tensor):

attention weights of previous step (B, T_max)

out_prev(Tensor):

decoder outputs of previous step (B, odim)

scaling(float, optional):

scaling parameter before applying softmax (Default value = 1.0)

last_attended_idx(int, optional):

index of the inputs of the last attended (Default value = None)

backward_window(int, optional):

backward window size in attention constraint (Default value = 1)

forward_window(int, optional):

(Default value = 3)

Returns:

Tensor:

attention weighted encoder state (B, dunits)

Tensor:

previous attention weights (B, Tmax)

reset()

class paddlespeech.t2s.modules.tacotron2.attentions.AttLoc(eprojs, dunits, att_dim, aconv_chans, aconv_filts, han_mode=False)

Bases: Layer

location-aware attention module.

Reference: Attention-Based Models for Speech Recognition

(https://arxiv.org/pdf/1506.07503.pdf)

Args:

eprojs (int):

projection-units of encoder

dunits (int):

units of decoder

att_dim (int):

attention dimension

aconv_chans (int):

channels of attention convolution

aconv_filts (int):

filter size of attention convolution

han_mode (bool):

flag to swith on mode of hierarchical attention and not store pre_compute_enc_h

Methods

__call__(*inputs, **kwargs)	Call self as a function.
add_parameter(name, parameter)	Adds a Parameter instance.
add_sublayer(name, sublayer)	Adds a sub Layer instance.
apply(fn)	Applies fn recursively to every sublayer (as returned by .sublayers()) as well as self.
buffers([include_sublayers])	Returns a list of all buffers from current layer and its sub-layers.
children()	Returns an iterator over immediate children layers.
clear_gradients()	Clear the gradients of all parameters for this layer.
create_parameter(shape[, attr, dtype, ...])	Create parameters for this layer.
create_tensor([name, persistable, dtype])	Create Tensor for this layer.
create_variable([name, persistable, dtype])	Create Tensor for this layer.
eval()	Sets this Layer and all its sublayers to evaluation mode.
extra_repr()	Extra representation of this layer, you can have custom implementation of your own layer.
forward(enc_hs_pad, enc_hs_len, dec_z, att_prev)	Calculate AttLoc forward propagation. Args: enc_hs_pad(Tensor): padded encoder hidden state (B, T_max, D_enc) enc_hs_len(Tensor): padded encoder hidden state length (B) dec_z(Tensor dec_z): decoder hidden state (B, D_dec) att_prev(Tensor): previous attention weight (B, T_max) scaling(float, optional): scaling parameter before applying softmax (Default value = 2.0) forward_window(Tensor, optional): forward window size when constraining attention (Default value = 3) last_attended_idx(int, optional): index of the inputs of the last attended (Default value = None) backward_window(int, optional): backward window size in attention constraint (Default value = 1) forward_window(int, optional): forward window size in attetion constraint (Default value = 3) Returns: Tensor: attention weighted encoder state (B, D_enc) Tensor: previous attention weights (B, T_max).
full_name()	Full name for this layer, composed by name_scope + "/" + MyLayer.__class__.__name__
load_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
named_buffers([prefix, include_sublayers])	Returns an iterator over all buffers in the Layer, yielding tuple of name and Tensor.
named_children()	Returns an iterator over immediate children layers, yielding both the name of the layer as well as the layer itself.
named_parameters([prefix, include_sublayers])	Returns an iterator over all parameters in the Layer, yielding tuple of name and parameter.
named_sublayers([prefix, include_self, ...])	Returns an iterator over all sublayers in the Layer, yielding tuple of name and sublayer.
parameters([include_sublayers])	Returns a list of all Parameters from current layer and its sub-layers.
register_buffer(name, tensor[, persistable])	Registers a tensor as buffer into the layer.
register_forward_post_hook(hook)	Register a forward post-hook for Layer.
register_forward_pre_hook(hook)	Register a forward pre-hook for Layer.
reset()	reset states
set_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
set_state_dict(state_dict[, use_structured_name])	Set parameters and persistable buffers from state_dict.
state_dict([destination, include_sublayers, ...])	Get all parameters and persistable buffers of current layer and its sub-layers.
sublayers([include_self])	Returns a list of sub layers.
to([device, dtype, blocking])	Cast the parameters and buffers of Layer by the give device, dtype and blocking.
to_static_state_dict([destination, ...])	Get all parameters and buffers of current layer and its sub-layers.
train()	Sets this Layer and all its sublayers to training mode.

backward
register_state_dict_hook

forward(enc_hs_pad, enc_hs_len, dec_z, att_prev, scaling=2.0, last_attended_idx=-1, backward_window=1, forward_window=3)

Calculate AttLoc forward propagation. Args:

enc_hs_pad(Tensor):

padded encoder hidden state (B, T_max, D_enc)

enc_hs_len(Tensor):

padded encoder hidden state length (B)

dec_z(Tensor dec_z):

decoder hidden state (B, D_dec)

att_prev(Tensor):

previous attention weight (B, T_max)

scaling(float, optional):

scaling parameter before applying softmax (Default value = 2.0)

forward_window(Tensor, optional):

forward window size when constraining attention (Default value = 3)

last_attended_idx(int, optional):

index of the inputs of the last attended (Default value = None)

backward_window(int, optional):

backward window size in attention constraint (Default value = 1)

forward_window(int, optional):

forward window size in attetion constraint (Default value = 3)

Returns:

Tensor:

attention weighted encoder state (B, D_enc)

Tensor:

previous attention weights (B, T_max)

reset()

reset states