o iR@s8dZddlmZmZmZddlZGdddejjZdS)z$Conv1d block for Transducer encoder.)OptionalTupleUnionNcsFeZdZdZ        d'dededeeefd eeefd eeefd eeefd ed ededededdffdd Z dede j ddfddZ  d(de j de j dee j dee j dee j e j e j ff ddZ  d)de j de j de j ded edee j e j ff d!d"Zde j de j fd#d$Zde j de j fd%d&ZZS)*Conv1daConv1d module definition. Args: input_size: Input dimension. output_size: Output dimension. kernel_size: Size of the convolving kernel. stride: Stride of the convolution. dilation: Spacing between the kernel points. groups: Number of blocked connections from input channels to output channels. bias: Whether to add a learnable bias to the output. batch_norm: Whether to use batch normalization after convolution. relu: Whether to use a ReLU activation after convolution. causal: Whether to use causal convolution (set to True if streaming). dropout_rate: Dropout rate. TF input_size output_size kernel_sizestridedilationgroupsbias batch_normrelucausal dropout_ratereturnNc st| r|d|_d}nd|_|}tjj|||||||d|_tjj| d|_| r2tj |_ |r;tj ||_ tj |||_||_||_| |_||_| |_||_||d|_||_d|_dS)zConstruct a Conv1d object.rr)r r r r)pN)super__init__lordertorchnnrconvDropoutdropoutReLU relu_func BatchNorm1dbnLinearout_posrr rrrr paddingr cache) selfrr r r r r rrrrr __class__`/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/asr_transducer/encoder/blocks/conv1d.pyrs<     zConv1d.__init__ left_contextdevicecCs"tjd|j|jdf|d|_dS)zInitialize/Reset Conv1d cache for streaming. Args: left_context: Number of left frames during chunk-by-chunk inference. device: Device to use for cache tensor. r)r+N)rzerosrr r$)r%r*r+r(r(r)reset_streaming_cacheSs zConv1d.reset_streaming_cachexpos_encmask chunk_maskcCs|dd}|jdkrtjj||jdfdd}n ||}||}||}|j r0| |}| |}|j r=| |}|dd}||||fS)a Encode input sequences. Args: x: Conv1d input sequences. (B, T, D_in) pos_enc: Positional embedding sequences. (B, 2 * (T - 1), D_in) mask: Source mask. (B, T) chunk_mask: Chunk mask. (T_2, T_2) Returns: x: Conv1d output sequences. (B, sub(T), D_out) mask: Source mask. (B, T) or (B, sub(T)) pos_enc: Positional embedding sequences. (B, 2 * (T - 1), D_att) or (B, 2 * (sub(T) - 1), D_out) rrconstantr) transposerrr functionalpadcreate_new_maskcreate_new_pos_encrrr rrrr")r%r.r/r0r1r(r(r)forward_s         zConv1d.forwardr right_contextcCstj|j|ddgdd}|dkr&|dddd|j| | f|_n|dddd|j df|_||}|jrC||}||}|j rP| |}|dd}|| |fS)aEncode chunk of input sequence. Args: x: Conv1d input sequences. (B, T, D_in) pos_enc: Positional embedding sequences. (B, 2 * (T - 1), D_in) mask: Source mask. (B, T) left_context: Number of frames in left context. right_context: Number of frames in right context. Returns: x: Conv1d output sequences. (B, T, D_out) pos_enc: Positional embedding sequences. (B, 2 * (T - 1), D_out) rr2dimrN) rcatr$r4rrrr rrrr")r%r.r/r0r*r:r(r(r) chunk_forwards(      zConv1d.chunk_forwardcCs:|jdkr|ddd|j f}|dddd|jfS)zCreate new mask for output sequences. Args: mask: Mask of input sequences. (B, T) Returns: mask: Mask of output sequences. (B, sub(T)) rN)r#r )r%r0r(r(r)r7s zConv1d.create_new_maskcCs|ddd|dddddf}|dd|dddddf}|jdkrI|ddd|j ddf}|ddd|j ddf}|dddd|jddf}|dddd|jddf}tj||ddddddfgdd}|S)a&Create new positional embedding vector. Args: pos_enc: Input sequences positional embedding. (B, 2 * (T - 1), D_in) Returns: pos_enc: Output sequences positional embedding. (B, 2 * (sub(T) - 1), D_in) Nrr2rr;)sizer#r rr=)r%r/pos_enc_positivepos_enc_negativer(r(r)r8s( $ (zConv1d.create_new_pos_enc)rrrTFTFr)NN)rr)__name__ __module__ __qualname____doc__intrrboolfloatrrr+r-Tensorrr9r>r7r8 __classcell__r(r(r&r)rs         9 1 +r) rEtypingrrrrrModulerr(r(r(r)s