o i@srdZddlZddlmZmZddlZddlmmZ Gdddejj Z Gddde Z Gdd d ejj Z dS) zeResidual block modules. This code is modified from https://github.com/kan-bayashi/ParallelWaveGAN. N)OptionalTuplecs(eZdZdZfddZddZZS)Conv1dz-Conv1d module with customized initialization.cstj|i|dS)zInitialize Conv1d module.Nsuper__init__)selfargskwargs __class__Z/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/gan_tts/wavenet/residual_block.pyrszConv1d.__init__cCs8tjjj|jdd|jdurtjj|jddSdS)zReset parameters.relu) nonlinearityN)torchnninitkaiming_normal_weightbias constant_)rr r rreset_parameterss zConv1d.reset_parameters)__name__ __module__ __qualname____doc__rr __classcell__r r r rrs rcs.eZdZdZdededeffdd ZZS) Conv1d1x1z*1x1 Conv1d with customized initialization. in_channels out_channelsrcstj||ddd|ddS)zInitialize 1x1 Conv1d module.r) kernel_sizepaddingdilationrNr)rr r!rr r rr"s  zConv1d1x1.__init__)rrrrintboolrrr r r rrs"rcseZdZdZ         dd ed ed edededededededeffdd Z   d dej de ej de ej de ej de ej ej ff ddZ Z S)! ResidualBlockz!Residual block module in WaveNet.@Prr"TFr#residual_channels gate_channels skip_channels aux_channelsglobal_channels dropout_rater%rscale_residualc st||_||_||_| |_|dddksJd|ddks%J|dd|} t|||| || d|_|dkrEt||dd|_ nd|_ |dkrUt||dd|_ nd|_ |d} t| ||| d|_ dS) a[Initialize ResidualBlock module. Args: kernel_size (int): Kernel size of dilation convolution layer. residual_channels (int): Number of channels for residual connection. skip_channels (int): Number of channels for skip connection. aux_channels (int): Number of local conditioning channels. dropout (float): Dropout probability. dilation (int): Dilation factor. bias (bool): Whether to add bias parameter in convolution layers. scale_residual (bool): Whether to scale the residual outputs. r"rz$Not support even number kernel size.)r$r%rF)rN) rrr3r.r0r4rconvr conv1x1_aux conv1x1_glo conv1x1_out) rr#r.r/r0r1r2r3r%rr4r$gate_out_channelsr r rr,s4    zResidualBlock.__init__Nxx_maskcgreturncCs.|}tj||j|jd}||}d}|j||d|d\}}|durA||}|j||d|d\} } || || }}|dura||}|j||d|d\} } || || }}t |t |}| |}|durx||}|j|j |jgdd\}} ||}|jr|td}|| fS)aCalculate forward propagation. Args: x (Tensor): Input tensor (B, residual_channels, T). x_mask Optional[torch.Tensor]: Mask tensor (B, 1, T). c (Optional[Tensor]): Local conditioning tensor (B, aux_channels, T). g (Optional[Tensor]): Global conditioning tensor (B, global_channels, 1). Returns: Tensor: Output tensor for residual connection (B, residual_channels, T). Tensor: Output tensor for skip connection (B, skip_channels, T). )ptrainingr"r5)dimNg?)Fdropoutr3rAr6splitsizer7r8rtanhsigmoidr9r.r0r4mathsqrt)rr;r<r=r>residualsplitdimxaxbcacbgagbsr r rforwardps,    zResidualBlock.forward) r)r*r+r*r,r-rr"TF)NNN)rrrrr&floatr'rrTensorrrrTrr r r rr()s^   Gr()rrItypingrrrtorch.nn.functionalr functionalrCrrModuler(r r r rs