speechbrain.lobes.models.transformer.TransformerASR module

Transformer for ASR in the SpeechBrain style.

Authors * Jianyuan Zhong 2020

Summary

Classes:

EncoderWrapper	This is a wrapper of any ASR transformer encoder.
TransformerASR	This is an implementation of transformer model for ASR.

Reference

class speechbrain.lobes.models.transformer.TransformerASR.TransformerASR(tgt_vocab, input_size, d_model=512, nhead=8, num_encoder_layers=6, num_decoder_layers=6, d_ffn=2048, dropout=0.1, activation=<class 'torch.nn.modules.activation.ReLU'>, positional_encoding='fixed_abs_sine', normalize_before=False, kernel_size: int | None = 31, bias: bool | None = True, encoder_module: str | None = 'transformer', conformer_activation: ~torch.nn.modules.module.Module | None = <class 'speechbrain.nnet.activations.Swish'>, attention_type: str | None = 'regularMHA', max_length: int | None = 2500, causal: bool | None = True)

Bases: TransformerInterface

This is an implementation of transformer model for ASR.

The architecture is based on the paper “Attention Is All You Need”: https://arxiv.org/pdf/1706.03762.pdf

Parameters:

• tgt_vocab (int) – Size of vocabulary.

• input_size (int) – Input feature size.

• d_model (int, optional) – Embedding dimension size. (default=512).

• nhead (int, optional) – The number of heads in the multi-head attention models (default=8).

• num_encoder_layers (int, optional) – The number of sub-encoder-layers in the encoder (default=6).

• num_decoder_layers (int, optional) – The number of sub-decoder-layers in the decoder (default=6).

• dim_ffn (int, optional) – The dimension of the feedforward network model (default=2048).

• dropout (int, optional) – The dropout value (default=0.1).

• activation (torch.nn.Module, optional) – The activation function of FFN layers. Recommended: relu or gelu (default=relu).

• positional_encoding (str, optional) – Type of positional encoding used. e.g. ‘fixed_abs_sine’ for fixed absolute positional encodings.

• normalize_before (bool, optional) – Whether normalization should be applied before or after MHA or FFN in Transformer layers. Defaults to True as this was shown to lead to better performance and training stability.

• kernel_size (int, optional) – Kernel size in convolutional layers when Conformer is used.

• bias (bool, optional) – Whether to use bias in Conformer convolutional layers.

• encoder_module (str, optional) – Choose between Conformer and Transformer for the encoder. The decoder is fixed to be a Transformer.

• conformer_activation (torch.nn.Module, optional) – Activation module used after Conformer convolutional layers. E.g. Swish, ReLU etc. it has to be a torch Module.

• attention_type (str, optional) – Type of attention layer used in all Transformer or Conformer layers. e.g. regularMHA or RelPosMHA.

• max_length (int, optional) – Max length for the target and source sequence in input. Used for positional encodings.

• causal (bool, optional) – Whether the encoder should be causal or not (the decoder is always causal). If causal the Conformer convolutional layer is causal.

Example

>>> src = torch.rand([8, 120, 512])
>>> tgt = torch.randint(0, 720, [8, 120])
>>> net = TransformerASR(
...     720, 512, 512, 8, 1, 1, 1024, activation=torch.nn.GELU
... )
>>> enc_out, dec_out = net.forward(src, tgt)
>>> enc_out.shape
torch.Size([8, 120, 512])
>>> dec_out.shape
torch.Size([8, 120, 512])

forward(src, tgt, wav_len=None, pad_idx=0)

Parameters:

• src (torch.Tensor) – The sequence to the encoder.

• tgt (torch.Tensor) – The sequence to the decoder.

• wav_len (torch.Tensor, optional) – Torch Tensor of shape (batch, ) containing the relative length to padded length for each example.

• pad_idx (int, optional) – The index for <pad> token (default=0).

make_masks(src, tgt, wav_len=None, pad_idx=0)

This method generates the masks for training the transformer model.

Parameters:

• src (tensor) – The sequence to the encoder (required).

• tgt (tensor) – The sequence to the decoder (required).

• pad_idx (int) – The index for <pad> token (default=0).

decode(tgt, encoder_out, enc_len=None)

This method implements a decoding step for the transformer model.

Parameters:

• tgt (torch.Tensor) – The sequence to the decoder.

• encoder_out (torch.Tensor) – Hidden output of the encoder.

• enc_len (torch.LongTensor) – The actual length of encoder states.

encode(src, wav_len=None)

Encoder forward pass

Parameters:

• src (torch.Tensor) – The sequence to the encoder.

• wav_len (torch.Tensor, optional) – Torch Tensor of shape (batch, ) containing the relative length to padded length for each example.

training: bool

class speechbrain.lobes.models.transformer.TransformerASR.EncoderWrapper(transformer, *args, **kwargs)

Bases: Module

This is a wrapper of any ASR transformer encoder. By default, the TransformerASR .forward() function encodes and decodes. With this wrapper the .forward() function becomes .encode() only.

Important: The TransformerASR class must contain a .encode() function.

Parameters:

transformer (sb.lobes.models.TransformerInterface) – A Transformer instance that contains a .encode() function.

Example

>>> src = torch.rand([8, 120, 512])
>>> tgt = torch.randint(0, 720, [8, 120])
>>> net = TransformerASR(
...     720, 512, 512, 8, 1, 1, 1024, activation=torch.nn.GELU
... )
>>> encoder = EncoderWrapper(net)
>>> enc_out = encoder(src)
>>> enc_out.shape
torch.Size([8, 120, 512])

forward(x, wav_lens=None)

Processes the input tensor x and returns an output tensor.

training: bool