o i@sddZddlmZmZmZddlZddlmZddlm Z Gdddej j Z Gdd d ej j Z dS) zhResidual affine coupling modules in VITS. This code is based on https://github.com/jaywalnut310/vits. )OptionalTupleUnionN)FlipFlow)WaveNetcseZdZdZ           dd ed ed ed ed ededededededeffdd Z  d dej dej de ej dedej f ddZ Z S)!ResidualAffineCouplingBlockavResidual affine coupling block module. This is a module of residual affine coupling block, which used as "Flow" in `Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech`_. .. _`Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech`: https://arxiv.org/abs/2006.04558 T in_channelshidden_channelsflows kernel_size base_dilationlayersglobal_channels dropout_rateuse_weight_normbias use_only_meanc sdttj|_t|D] } |jt|||||d||| | | d g7_|jtg7_qdS)aInitilize ResidualAffineCouplingBlock module. Args: in_channels (int): Number of input channels. hidden_channels (int): Number of hidden channels. flows (int): Number of flows. kernel_size (int): Kernel size for WaveNet. base_dilation (int): Base dilation factor for WaveNet. layers (int): Number of layers of WaveNet. stacks (int): Number of stacks of WaveNet. global_channels (int): Number of global channels. dropout_rate (float): Dropout rate. use_weight_norm (bool): Whether to use weight normalization in WaveNet. bias (bool): Whether to use bias paramters in WaveNet. use_only_mean (bool): Whether to estimate only mean. r ) rrrrrstacksrrrrrN) super__init__torchnn ModuleListrrangeResidualAffineCouplingLayerr) selfrrrrrrrrrrri __class__Z/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/gan_tts/vits/residual_coupling.pyrs(   z$ResidualAffineCouplingBlock.__init__NFxx_maskginversereturncCsL|s|jD] }|||||d\}}q|St|jD] }|||||d}q|S)a}Calculate forward propagation. Args: x (Tensor): Input tensor (B, in_channels, T). x_lengths (Tensor): Length tensor (B,). g (Optional[Tensor]): Global conditioning tensor (B, global_channels, 1). inverse (bool): Whether to inverse the flow. Returns: Tensor: Output tensor (B, in_channels, T). )r)r*)rreversed)r!r'r(r)r*flow_r%r%r&forwardRs z#ResidualAffineCouplingBlock.forward) rrr r r r r r TTTNF) __name__ __module__ __qualname____doc__intfloatboolrrTensorrr/ __classcell__r%r%r#r&rsb     8rcseZdZdZ           dded ed ed ed ed edededededeffdd Z  ddej dej de ej dede ej e ej ej fff ddZ ZS) r zResidual affine coupling layer.rr r r r Trrrrrrrrrrrc s|ddks Jdt|d|_| |_tj|j|d|_tdiddddd|d |d |d |d |d dd|dd|d|d|d| d| dddddddd|_ | rltj||jd|_ n tj||jdd|_ |j j j |j jj dS)aInitialzie ResidualAffineCouplingLayer module. Args: in_channels (int): Number of input channels. hidden_channels (int): Number of hidden channels. kernel_size (int): Kernel size for WaveNet. base_dilation (int): Base dilation factor for WaveNet. layers (int): Number of layers of WaveNet. stacks (int): Number of stacks of WaveNet. global_channels (int): Number of global channels. dropout_rate (float): Dropout rate. use_weight_norm (bool): Whether to use weight normalization in WaveNet. bias (bool): Whether to use bias paramters in WaveNet. use_only_mean (bool): Whether to estimate only mean. rz$in_channels should be divisible by 2r rr out_channelsrrrrresidual_channels aux_channels gate_channels skip_channelsrrrruse_first_convF use_last_convscale_residualscale_skip_connectTNr%)rr half_channelsrrrConv1d input_convrencoderprojweightdatazero_r) r!rrrrrrrrrrrr#r%r&rqsv       z$ResidualAffineCouplingLayer.__init__NFr'r(r)r*r+c Cs|j|dddd\}}|||}|j|||d}|||}|js6|j|dddd\} } n|} t| } |s^| |t| |}t ||gd}t | ddg} || fS|| t| |}t ||gd}|S)aCalculate forward propagation. Args: x (Tensor): Input tensor (B, in_channels, T). x_lengths (Tensor): Length tensor (B,). g (Optional[Tensor]): Global conditioning tensor (B, global_channels, 1). inverse (bool): Whether to inverse the flow. Returns: Tensor: Output tensor (B, in_channels, T). Tensor: Log-determinant tensor for NLL (B,) if not inverse. r r:)dim)r)) splitsizerFrGrHrr zeros_likeexpcatsum) r!r'r(r)r*xaxbhstatsmlogslogdetr%r%r&r/s  z#ResidualAffineCouplingLayer.forward) rrr r r r r r TTTr0)r1r2r3r4r5r6r7rrr8rrrr/r9r%r%r#r&r nsb    Pr )r4typingrrrrespnet2.gan_tts.vits.flowrespnet2.gan_tts.wavenetrrModulerr r%r%r%r&s  \