o i@sDdZddlmZmZmZddlZddlmZGdddejj Z dS)z'ConvInput block for Transducer encoder.)OptionalTupleUnionN)sub_factor_to_paramsc seZdZdZ   ddedeeefdeded eed df fd d Z d e j dee j d ee j e j ffddZ de j d e j fddZ de j d e j fddZded efddZZS) ConvInputaConvInput module definition. Args: input_size: Input size. conv_size: Convolution size. subsampling_factor: Subsampling factor. vgg_like: Whether to use a VGG-like network. output_size: Block output dimension. TN input_size conv_sizesubsampling_factorvgg_like output_sizereturnc s~t|rc|\}}tjtjjd|ddddtjtjj||ddddtjtjdtjj||ddddtjtjj||ddddtjtjd |_||dd}d|_ |j |_ n7t ||\} } } tjtjd|ddtjtj||| | tj|_|| }||_ | |_ | |_|j|_ ||_|dkrd nd |_|d urtj|||_||_d Sd |_||_d S) zConstruct a ConvInput object.)stridepadding)r)rrrr N)super__init__torchnn SequentialConv2dReLU MaxPool2dconvr create_new_vgg_maskcreate_new_maskrkernel_2stride_2create_new_conv2d_maskr min_frame_lengthLinearoutputr ) selfrr r r r conv_size1 conv_size2 output_projr!r"conv_2_output_size __class__d/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/asr_transducer/encoder/blocks/conv_input.pyrsP        zConvInput.__init__xmaskcCsl||d}|\}}}}|dd||||}|jdur)||}|dur2||}||fS)a*Encode input sequences. Args: x: ConvInput input sequences. (B, T, D_feats) mask: Mask of input sequences. (B, 1, T) Returns: x: ConvInput output sequences. (B, sub(T), D_out) mask: Mask of output sequences. (B, 1, sub(T)) rrN)r unsqueezesize transpose contiguousviewr&r )r'r0r1bctfr.r.r/forwardUs   zConvInput.forwardcCs|d|dd}|ddd|fdddddf}|d|dd}|ddd|fdddddf}|S)zCreate a new mask for VGG output sequences. Args: mask: Mask of input sequences. (B, T) Returns: mask: Mask of output sequences. (B, sub(T)) rrNr)r3)r'r1 vgg1_t_len vgg2_t_lenr.r.r/rps  &&zConvInput.create_new_vgg_maskcCs2|dddddfddd|jd |jfS)zCreate new conformer mask for Conv2d output sequences. Args: mask: Mask of input sequences. (B, T) Returns: mask: Mask of output sequences. (B, sub(T)) Nrr)r!r")r'r1r.r.r/r#s2 z ConvInput.create_new_conv2d_maskr3cCs2|jr |dddS|dd|jd|jS)zReturn the original size before subsampling for a given size. Args: size: Number of frames after subsampling. Returns: : Number of frames before subsampling. rrr)r r!r")r'r3r.r.r/get_size_before_subsamplings z%ConvInput.get_size_before_subsampling)rTN)__name__ __module__ __qualname____doc__intrrboolrrrTensorr;rr#r? __classcell__r.r.r,r/r s8 ?  r) rCtypingrrrrespnet2.asr_transducer.utilsrrModulerr.r.r.r/s