o i)@sfdZddlZddlmZddlZddlmmZddl m Z m Z m Z m Z mZGdddejjZdS)zmStochastic duration predictor modules in VITS. This code is based on https://github.com/jaywalnut310/vits. N)Optional)ConvFlowDilatedDepthSeparableConvElementwiseAffineFlowFlipFlowLogFlowcseZdZdZ      ddeded ed ed ed ef fd d Z    ddejdejde ejde ejde dedejfddZ Z S)StochasticDurationPredictoradStochastic duration predictor module. This is a module of stochastic duration predictor described 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 ?channels kernel_size dropout_rateflowsdds_conv_layersglobal_channelsc sXttj||d|_t||||d|_tj||d|_t |_ tj |_ |j t dg7_ t|D]}|j td|||dg7_ |j tg7_ q8tjd|d|_t||||d|_tj||d|_tj |_|jt dg7_t|D]}|jtd|||dg7_|jtg7_q|dkrtj||d|_dSdS)aInitialize StochasticDurationPredictor module. Args: channels (int): Number of channels. kernel_size (int): Kernel size. dropout_rate (float): Dropout rate. flows (int): Number of flows. dds_conv_layers (int): Number of conv layers in DDS conv. global_channels (int): Number of global conditioning channels. )layersr)rrN)super__init__torchnnConv1dprerddsprojrlog_flow ModuleListrrrangerrpost_prepost_dds post_proj post_flows global_conv)selfrrrrrri __class__[/home/ubuntu/.local/lib/python3.10/site-packages/espnet2/gan_tts/vits/duration_predictor.pyr$sZ     z$StochasticDurationPredictor.__init__NF?xx_maskwginverse noise_scalereturncCsv|}||}|dur|||}|||}|||}|s|dus-Jd||}|||}|||}t | dd| dj |j |j d|}|} d} |jD]} | | |||d\} } | | 7} q\t| ddgd\} }t| |}|||}| tt| t| |ddg7} td tdtj|d|ddg| }d}|||\}}||7}t||gd}|jD]} | ||||d \}}||}qtd tdtj|d|ddg|}||Stt|j}|dd |d g}t | dd| dj |j |j d|}|D] } | ||||d }q#|dd\}}|}|S)aJCalculate forward propagation. Args: x (Tensor): Input tensor (B, channels, T_text). x_mask (Tensor): Mask tensor (B, 1, T_text). w (Optional[Tensor]): Duration tensor (B, 1, T_text). g (Optional[Tensor]): Global conditioning tensor (B, channels, 1) inverse (bool): Whether to inverse the flow. noise_scale (float): Noise scale value. Returns: Tensor: If not inverse, negative log-likelihood (NLL) tensor (B,). If inverse, log-duration tensor (B, 1, T_text). Nzw must be provided.rr)devicedtypeg)r1rg)r1r2r r )detachrr&rrr"r#r$rrandnsizetor5r6r%splitsigmoidsumF logsigmoidmathlogpircatrlistreversed)r'r.r/r0r1r2r3h_we_qz_q logdet_tot_qflowlogdet_qz_uz1uz0logq logdet_totlogdetznllrlogwr+r+r,forwardis        *  * z#StochasticDurationPredictor.forward)r r r r r r )NNFr-) __name__ __module__ __qualname____doc__intfloatrrTensorrboolrW __classcell__r+r+r)r,rsP Ir)r[rAtypingrrtorch.nn.functionalr functionalr?espnet2.gan_tts.vits.flowrrrrrModulerr+r+r+r,s