Method __init__

espnet2/asr/decoder/transformer_decoder.py:721–775

(
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
        use_speech_attn: bool = True,
    )

class TransformerMDDecoder(BaseTransformerDecoder):
    @typechecked
    def __init__(
        self,
        vocab_size: int,
        encoder_output_size: int,
        attention_heads: int = 4,
        linear_units: int = 2048,
        num_blocks: int = 6,
        dropout_rate: float = 0.1,
        positional_dropout_rate: float = 0.1,
        self_attention_dropout_rate: float = 0.0,
        src_attention_dropout_rate: float = 0.0,
        input_layer: str = "embed",
        use_output_layer: bool = True,
        pos_enc_class=PositionalEncoding,
        normalize_before: bool = True,
        concat_after: bool = False,
        use_speech_attn: bool = True,
    ):
        super().__init__(
            vocab_size=vocab_size,
            encoder_output_size=encoder_output_size,
            dropout_rate=dropout_rate,
            positional_dropout_rate=positional_dropout_rate,
            input_layer=input_layer,
            use_output_layer=use_output_layer,
            pos_enc_class=pos_enc_class,
            normalize_before=normalize_before,
        )

        attention_dim = encoder_output_size
        self.decoders = repeat(
            num_blocks,
            lambda lnum: DecoderLayer(
                attention_dim,
                MultiHeadedAttention(
                    attention_heads, attention_dim, self_attention_dropout_rate
                ),
                MultiHeadedAttention(
                    attention_heads, attention_dim, src_attention_dropout_rate
                ),
                PositionwiseFeedForward(attention_dim, linear_units, dropout_rate),
                dropout_rate,
                normalize_before,
                concat_after,
                (
                    MultiHeadedAttention(
                        attention_heads, attention_dim, src_attention_dropout_rate
                    )
                    if use_speech_attn
                    else None
                ),
            ),
        )

        self.use_speech_attn = use_speech_attn

    def forward(
        self,