o i@sLdZddlmZmZmZddlZGdddejjZGdddejjZ dS)z(Convolution modules for X-former blocks.)DictOptionalTupleNc seZdZdZejidfdededejjde de dd f fd d Z dd ej de ej dedeej ej ffddZZS)ConformerConvolutionaHConformerConvolution module definition. Args: channels: The number of channels. kernel_size: Size of the convolving kernel. activation: Type of activation function. norm_args: Normalization module arguments. causal: Whether to use causal convolution (set to True if streaming). Fchannels kernel_size activation norm_argscausalreturnNcst|dddksJ||_tjj|d|dddd|_|r*|d|_d}n d|_|dd}tjj|||d||d|_tjj |fi||_ tjj||dddd|_ ||_ dS)z)Construct an ConformerConvolution object.r)rstridepaddingrrgroupsN) super__init__rtorchnnConv1dpointwise_conv1lorderdepthwise_conv BatchNorm1dnormpointwise_conv2r)selfrrrr r r __class__f/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/asr_transducer/encoder/modules/convolution.pyrs@    zConformerConvolution.__init__rxcache right_contextcCs||dd}tjjj|dd}|jdkrX|dur)tjj||jdfdd}n/tj||gdd}|dkrI|dddd|j| | f}n|dddd|j df}| |}| | |}| |dd}||fS)aCompute convolution module. Args: x: ConformerConvolution input sequences. (B, T, D_hidden) cache: ConformerConvolution input cache. (1, conv_kernel, D_hidden) right_context: Number of frames in right context. Returns: x: ConformerConvolution output sequences. (B, T, D_hidden) cache: ConformerConvolution output cache. (1, conv_kernel, D_hidden) r r dimrNconstant) r transposerr functionalglurpadcatrrrr)rr"r#r$r r r!forwardEs & zConformerConvolution.forwardNr)__name__ __module__ __qualname____doc__rrReLUintModulerboolrTensorrrr. __classcell__r r rr!rs:4rcseZdZdZejjiddfdededejjde de d e d d ffd d Z ddej deej ded eej ej ffddZZS)ConvolutionalSpatialGatingUnitaConvolutional Spatial Gating Unit module definition. Args: size: Initial size to determine the number of channels. kernel_size: Size of the convolving kernel. norm_class: Normalization module class. norm_args: Normalization module arguments. dropout_rate: Dropout rate. causal: Whether to use causal convolution (set to True if streaming). r(Fsizer norm_classr dropout_rater r Nc st|d}||_|r|d|_d}n d|_|dd}tjj|||d||d|_||fi||_tj |_ tj ||_ dS)z2Construct a ConvolutionalSpatialGatingUnit object.r r rrN) rrrrrrrconvrIdentityrDropoutdropout) rr;rr<r r=r rrrr r!rzs&    z'ConvolutionalSpatialGatingUnit.__init__rr"r#r$cCs|jddd\}}||dd}|jdkrX|dur)tjj||jdfdd}n/tj||gdd}|dkrI|dddd|j| | f}n|dddd|j df}| |dd}| || |}||fS) aCompute convolution module. Args: x: ConvolutionalSpatialGatingUnit input sequences. (B, T, D_hidden) cache: ConvolutionalSpationGatingUnit input cache. (1, conv_kernel, D_hidden) right_context: Number of frames in right context. Returns: x: ConvolutionalSpatialGatingUnit output sequences. (B, T, D_hidden // 2) r r%r rNr'r() chunkrr)rrrr*r,r-r>rAr)rr"r#r$x_rx_gr r r!r.s &z&ConvolutionalSpatialGatingUnit.forwardr/)r0r1r2r3rr LayerNormr5r6rfloatr7rr8rrr.r9r r rr!r:ms@(r:) r3typingrrrrrr6rr:r r r r!s e