o jz)@sddlZddlmZddlmZmZddlmZddlmZddl m Z ddl m Z dZ d d ZGd d d ejjZGd ddejjZGdddejjZdS)N)nn)Conv1dConvTranspose1d) functional) weight_norm)remove_parametrizations) load_fsspecg?cCst|||dS)N)int)kdr W/home/kuhnn/.local/lib/python3.10/site-packages/TTS/vocoder/models/hifigan_generator.py get_paddingsrc2eZdZdZd fdd ZddZdd ZZS) ResBlock1a;Residual Block Type 1. It has 3 convolutional layers in each convolutional block. Network:: x -> lrelu -> conv1_1 -> conv1_2 -> conv1_3 -> z -> lrelu -> conv2_1 -> conv2_2 -> conv2_3 -> o -> + -> o |--------------------------------------------------------------------------------------------------| Args: channels (int): number of hidden channels for the convolutional layers. kernel_size (int): size of the convolution filter in each layer. dilations (list): list of dilation value for each conv layer in a block. rcstttt|||d|dt||ddtt|||d|dt||ddtt|||d|dt||ddg|_ttt|||ddt|ddtt|||ddt|ddtt|||ddt|ddg|_dS)Nrrdilationpaddingr ) super__init__r ModuleListrrrconvs1convs2selfchannels kernel_sizer __class__r rr!s\      # zResBlock1.__init__cCsLt|j|jD]\}}t|t}||}t|t}||}||}q|S)z Args: x (Tensor): input tensor. Returns: Tensor: output tensor. Shapes: x: [B, C, T] )ziprrF leaky_relu LRELU_SLOPE)rxc1c2xtr r rforwardTs   zResBlock1.forwardcCs0|jD]}t|dq|jD]}t|dqdSNweight)rrrrlr r rremove_weight_normes    zResBlock1.remove_weight_norm)rr__name__ __module__ __qualname____doc__rr,r1 __classcell__r r r"rrs 3rcr) ResBlock2aResidual Block Type 2. It has 1 convolutional layers in each convolutional block. Network:: x -> lrelu -> conv1-> -> z -> lrelu -> conv2-> o -> + -> o |---------------------------------------------------| Args: channels (int): number of hidden channels for the convolutional layers. kernel_size (int): size of the convolution filter in each layer. dilations (list): list of dilation value for each conv layer in a block. rrrcsbtttt|||d|dt||ddtt|||d|dt||ddg|_dS)Nrrr)rrrrrrrconvsrr"r rr{s0     zResBlock2.__init__cCs,|jD]}t|t}||}||}q|SN)r:r%r&r')rr(cr+r r rr,s   zResBlock2.forwardcCs|jD]}t|dqdSr-)r:rr/r r rr1s  zResBlock2.remove_weight_norm)rr9r2r r r"rr8ls r8csVeZdZ     dfdd ZdddZed d Zd d Z dddZ Z S)HifiganGeneratorrrTc sdt| |_t||_t||_tt||dddd|_|dkr$t nt }t |_ tt||D]#\}\}}|j tt|d||d|d||||ddq2t |_tt|j D]"}|d|d}tt||D]\}\}}|j||||qsqbtt||ddd| d|_| dkrt | |d|_| st|jd | st|jd d Sd S) aHiFiGAN Generator with Multi-Receptive Field Fusion (MRF) Network: x -> lrelu -> upsampling_layer -> resblock1_k1x1 -> z1 -> + -> z_sum / #resblocks -> lrelu -> conv_post_7x1 -> tanh -> o .. -> zI ---| resblockN_kNx1 -> zN ---' Args: in_channels (int): number of input tensor channels. out_channels (int): number of output tensor channels. resblock_type (str): type of the `ResBlock`. '1' or '2'. resblock_dilation_sizes (List[List[int]]): list of dilation values in each layer of a `ResBlock`. resblock_kernel_sizes (List[int]): list of kernel sizes for each `ResBlock`. upsample_kernel_sizes (List[int]): list of kernel sizes for each transposed convolution. upsample_initial_channel (int): number of channels for the first upsampling layer. This is divided by 2 for each consecutive upsampling layer. upsample_factors (List[int]): upsampling factors (stride) for each upsampling layer. inference_padding (int): constant padding applied to the input at inference time. Defaults to 5. rr)r1r )rbiasrr.N)rrinference_paddinglen num_kernels num_upsamplesrrconv_prerr8rrups enumerater$appendr resblocksrange conv_post cond_layerr)r in_channels out_channels resblock_typeresblock_dilation_sizesresblock_kernel_sizesupsample_kernel_sizesupsample_initial_channelupsample_factorsrA cond_channelsconv_pre_weight_normconv_post_weight_normconv_post_biasresblockiur ch_r r"r rrsB #       zHifiganGenerator.__init__NcCs||}t|dr|||}t|jD]=}t|t}|j||}d}t|j D]!}|dur?|j ||j ||}q,||j ||j ||7}q,||j }qt|}| |}t |}|S)z Args: x (Tensor): feature input tensor. g (Tensor): global conditioning input tensor. Returns: Tensor: output waveform. Shapes: x: [B, C, T] Tensor: [B, 1, T] rLN)rEhasattrrLrJrDr%r&r'rFrCrIrKtorchtanh)rr(gorZz_sumjr r rr,s      zHifiganGenerator.forwardcCs4||jjj}tjj||j|jfd}| |S)z Args: x (Tensor): conditioning input tensor. Returns: Tensor: output waveform. Shapes: x: [B, C, T] Tensor: [B, 1, T] replicate) torEr.devicer_rrpadrAr,)rr<r r r inference s  zHifiganGenerator.inferencecCsNtd|jD]}t|dq|jD]}|qt|jdt|jddS)NzRemoving weight norm...r.)printrFrrIr1rErKr/r r rr1s     z#HifiganGenerator.remove_weight_normFcCsHt|td|d}||d|r"||jrJ|dSdS)Ncpu) map_locationcachemodel)rr_rgload_state_dictevaltrainingr1)rconfigcheckpoint_pathrprmstater r rload_checkpoint%s  z HifiganGenerator.load_checkpoint)rrTTTr;)FF) r3r4r5rr,r_no_gradrir1rur7r r r"rr=s  I  r=)r_rtorch.nnrrrr%torch.nn.utils.parametrizationsrtorch.nn.utils.parametrizer TTS.utils.iorr'rModulerr8r=r r r rs     Z6