o i@sdZddlmZmZmZmZddlZddlZddl m m Z ddl mZGdddej jZGdddej jZGd d d ej jZGd d d ej jZdS) z`Upsampling module. This code is modified from https://github.com/kan-bayashi/ParallelWaveGAN. )AnyDictListOptionalN)Conv1dcsFeZdZdZd dededeffdd Zdejd ejfd d Z Z S) Stretch2dzStretch2d module.nearestx_scaley_scalemodecs t||_||_||_dS)zInitialize Stretch2d module. Args: x_scale (int): X scaling factor (Time axis in spectrogram). y_scale (int): Y scaling factor (Frequency axis in spectrogram). mode (str): Interpolation mode. N)super__init__r r r )selfr r r  __class__]/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/gan_tts/parallel_wavegan/upsample.pyr s  zStretch2d.__init__xreturncCstj||j|jf|jdS)zCalculate forward propagation. Args: x (Tensor): Input tensor (B, C, F, T). Returns: Tensor: Interpolated tensor (B, C, F * y_scale, T * x_scale), ) scale_factorr )F interpolater r r )rrrrrforward$s zStretch2d.forward)r) __name__ __module__ __qualname____doc__intstrr torchTensorr __classcell__rrrrrsrcs(eZdZdZfddZddZZS)Conv2dz-Conv2d module with customized initialization.cstj|i|dS)zInitialize Conv2d module.N)r r )rargskwargsrrrr 6szConv2d.__init__cCs>|jjdt|j|jdurtjj |jddSdS)zReset parameters.g?Ng) weightdatafill_npprod kernel_sizebiasrnninit constant_)rrrrreset_parameters:s zConv2d.reset_parameters)rrrrr r/r!rrrrr"3s r"c sfeZdZdZdiddfdeedeedeee fded ef fd d Z d e j d e j fddZ ZS)UpsampleNetworkzUpsampling network module.Nrupsample_scalesnonlinear_activationnonlinear_activation_paramsinterpolate_modefreq_axis_kernel_sizec sttj|_|D]V}t|d|}|j|g7_|dddks)Jd|dd}||ddf} ||f} tdd| | dd} |j| g7_|durcttj|di|} |j| g7_q dS) aInitialize UpsampleNetwork module. Args: upsample_scales (List[int]): List of upsampling scales. nonlinear_activation (Optional[str]): Activation function name. nonlinear_activation_params (Dict[str, Any]): Arguments for the specified activation function. interpolate_mode (str): Interpolation mode. freq_axis_kernel_size (int): Kernel size in the direction of frequency axis. r1rz.Not support even number freq axis kernel size.F)r*paddingr+Nr) r r rr, ModuleList up_layersrr"getattr) rr2r3r4r5r6scalestretchfreq_axis_paddingr*r8conv nonlinearrrrr Ds0    zUpsampleNetwork.__init__crcCs(|d}|jD]}||}q|dS)zCalculate forward propagation. Args: c : Input tensor (B, C, T_feats). Returns: Tensor: Upsampled tensor (B, C, T_wav). r1) unsqueezer:squeeze)rrAfrrrros   zUpsampleNetwork.forwardrrrrrrrrrrr rr rr!rrrrr0As$ +r0csreZdZdZdiddddfdeedeed eee fd ed ed ed effdd Z de j de j fddZ ZS)ConvInUpsampleNetworkz(Convolution + upsampling network module.Nrr1Prr2r3r4r5r6 aux_channelsaux_context_windowc sTt||_d|d}tj||_t|||dd|_t |||||d|_ dS)asInitialize ConvInUpsampleNetwork module. Args: upsample_scales (list): List of upsampling scales. nonlinear_activation (Optional[str]): Activation function name. nonlinear_activation_params (Dict[str, Any]): Arguments for the specified activation function. mode (str): Interpolation mode. freq_axis_kernel_size (int): Kernel size in the direction of frequency axis. aux_channels (int): Number of channels of pre-conv layer. aux_context_window (int): Context window size of the pre-conv layer. r7r1F)r*r+)r2r3r4r5r6N) r r rIrr,ReplicationPad1dpadrconv_inr0upsample) rr2r3r4r5r6rHrIr*rrrr s"   zConvInUpsampleNetwork.__init__rArcCs|||}||S)zCalculate forward propagation. Args: c (Tensor): Input tensor (B, C, T_feats). Returns: Tensor: Upsampled tensor (B, C, T_wav), where T_wav = T_feats * prod(upsample_scales). )rLrKrM)rrArrrrs zConvInUpsampleNetwork.forwardrErrrrrFs0 ,rF)rtypingrrrrnumpyr(rtorch.nn.functionalr, functionalr&espnet2.gan_tts.wavenet.residual_blockrModulerr"r0rFrrrrs  >