o i@sLddlmZddlmZddlmZGdddejZGdddejZdS)N) choose_norm)get_activationcs4eZdZdZ    d fdd Zd d d ZZS) ImprovedTransformerLayeraContainer module of the (improved) Transformer proposed in [1]. Reference: Dual-path transformer network: Direct context-aware modeling for end-to-end monaural speech separation; Chen et al, Interspeech 2020. Args: rnn_type (str): select from 'RNN', 'LSTM' and 'GRU'. input_size (int): Dimension of the input feature. att_heads (int): Number of attention heads. hidden_size (int): Dimension of the hidden state. dropout (float): Dropout ratio. Default is 0. activation (str): activation function applied at the output of RNN. bidirectional (bool, optional): True for bidirectional Inter-Chunk RNN (Intra-Chunk is always bidirectional). norm (str, optional): Type of normalization to use. reluTgLNc st|}|dvsJd|||_||_tj|||d|_tj|d|_ t |||_ t t|||dd|d|_ t|}|rGd|n|} t|tj|dt| ||_t |||_dS) N)RNNLSTMGRUz4Only support 'RNN', 'LSTM' and 'GRU', current type: )dropout)pT) batch_first bidirectional)super__init__upperrnn_type att_headsnnMultiheadAttention self_attnDropoutr r norm_attngetattrrnnr SequentialLinear feed_forwardnorm_ff) selfr input_sizer hidden_sizer activationrnormhdim __class__M/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/enh/layers/dptnet.pyr!s.   z!ImprovedTransformerLayer.__init__NcCs|ddd}|j||||ddddd}|||}||dddd}|||d}|||}||ddddS)Nr rr) attn_mask)permuterr r transposerrr )r!xr+srcoutout2r)r)r*forwardKs z ImprovedTransformerLayer.forward)rrTrN)__name__ __module__ __qualname____doc__rr4 __classcell__r)r)r'r*rs*rcsFeZdZdZ      dfdd Zd d Zd d ZddZZS)DPTNetaDual-path transformer network. args: rnn_type (str): select from 'RNN', 'LSTM' and 'GRU'. input_size (int): dimension of the input feature. Input size must be a multiple of `att_heads`. hidden_size (int): dimension of the hidden state. output_size (int): dimension of the output size. att_heads (int): number of attention heads. dropout (float): dropout ratio. Default is 0. activation (str): activation function applied at the output of RNN. num_layers (int): number of stacked RNN layers. Default is 1. bidirectional (bool): whether the RNN layers are bidirectional. Default is True. norm_type (str): type of normalization to use after each inter- or intra-chunk Transformer block. rrr Trc st||_||_||_t|_t|_t |D]"} |j t ||||||d| d|j t ||||||| | dqt t t||d|_dS)NT)r r$rr%r )rrr"r# output_sizer ModuleListrow_transformercol_transformerrangeappendrrPReLUConv2doutput) r!rr"r#r=rr r$ num_layersr norm_typeir'r)r*rks@      zDPTNet.__init__cCs>|}tt|jD]}|||}|||}q ||}|Sr5)rAlenr?intra_chunk_processinter_chunk_processrE)r!inputrErHr)r)r*r4s   zDPTNet.forwardcCsX|\}}}}|dd||||}|j||}|||||dddd}|S)Nr r,rr)sizer/reshaper?r.r!r0 layer_indexbatchN chunk_sizen_chunksr)r)r*rJs zDPTNet.intra_chunk_processcCs\|\}}}}|dddd||||}|j||}|||||dddd}|S)NrrrMr )rNr.rOr@rPr)r)r*rKs zDPTNet.inter_chunk_process)r<rrr Tr) r6r7r8r9rr4rJrKr:r)r)r'r*r;Ys3r;) torch.nnrespnet2.enh.layers.tcnr&espnet.nets.pytorch_backend.nets_utilsrModulerr;r)r)r)r*s   K