o j@sjddlmZddlmZddlmZddlmZGdddejZ GdddejZ Gd d d ejZ d S) )nn)ResidualConv1dBNBlock)FFTransformerBlock)RelativePositionTransformerc*eZdZdZfddZdddZZS)"RelativePositionTransformerEncoderawSpeedy speech encoder built on Transformer with Relative Position encoding. TODO: Integrate speaker conditioning vector. Args: in_channels (int): number of input channels. out_channels (int): number of output channels. hidden_channels (int): number of hidden channels params (dict): dictionary for residual convolutional blocks. c s@tt|||dddgdd|_t|||fi||_dS)N)r r r ) kernel_sizenum_res_blocksnum_conv_blocks dilations)super__init__rprenetrrel_pos_transformerself in_channels out_channelshidden_channelsparams __class__V/home/kuhnn/.local/lib/python3.10/site-packages/TTS/tts/layers/feed_forward/encoder.pyrs  z+RelativePositionTransformerEncoder.__init__NcCs*|durd}|||}|||}|SNr )rrrxx_maskgorrrforward!s  z*RelativePositionTransformerEncoder.forwardNN__name__ __module__ __qualname____doc__rr# __classcell__rrrrrs  rcr)ResidualConv1dBNEncoderaqResidual Convolutional Encoder as in the original Speedy Speech paper TODO: Integrate speaker conditioning vector. Args: in_channels (int): number of input channels. out_channels (int): number of output channels. hidden_channels (int): number of hidden channels params (dict): dictionary for residual convolutional blocks. c srttt||dt|_t|||fi||_tjt||dtt |t||dg|_ dSr) rrr SequentialConv1dReLUrrres_conv_block BatchNorm1dpostnetrrrrr5s    z ResidualConv1dBNEncoder.__init__NcCs@|durd}|||}|||}||||}||Sr)rr/r1rrrrr#Cs  zResidualConv1dBNEncoder.forwardr$r%rrrrr+)s r+csJeZdZdZdddgddgdddd ffd d Zdd dZZS)Encodera~Factory class for Speedy Speech encoder enables different encoder types internally. Args: num_chars (int): number of characters. out_channels (int): number of output channels. in_hidden_channels (int): input and hidden channels. Model keeps the input channels for the intermediate layers. encoder_type (str): encoder layer types. 'transformers' or 'residual_conv_bn'. Default 'residual_conv_bn'. encoder_params (dict): model parameters for specified encoder type. c_in_channels (int): number of channels for conditional input. Note: Default encoder_params to be set in config.json... ```python # for 'relative_position_transformer' encoder_params={ 'hidden_channels_ffn': 128, 'num_heads': 2, "kernel_size": 3, "dropout_p": 0.1, "num_layers": 6, "rel_attn_window_size": 4, "input_length": None }, # for 'residual_conv_bn' encoder_params = { "kernel_size": 4, "dilations": 4 * [1, 2, 4] + [1], "num_conv_blocks": 2, "num_res_blocks": 13 } # for 'fftransformer' encoder_params = { "hidden_channels_ffn": 1024 , "num_heads": 2, "num_layers": 6, "dropout_p": 0.1 } ``` residual_conv_bn)r r4r r5 )r rr r rcst||_||_||_||_||_|dkr$t|||||_ dS|dkr4t |||||_ dS|dkrM||ksBJdt |fi||_ dSt d)Nrelative_position_transformerr3 fftransformerzP[!] must be `in_channels` == `out_channels` when encoder type is 'fftransformer'z [!] unknown encoder type.) rrrrr encoder_type c_in_channelslowerrencoderr+rNotImplementedError)rin_hidden_channelsrr9encoder_paramsr:rrrrys$     zEncoder.__init__NcCs|||}||S)zi Shapes: x: [B, C, T] x_mask: [B, 1, T] g: [B, C, 1] )r<rrrrr#s zEncoder.forward)Nr%rrrrr2Ls/!r2N) torchr"TTS.tts.layers.generic.res_conv_bnr"TTS.tts.layers.generic.transformerr#TTS.tts.layers.glow_tts.transformerrModulerr+r2rrrrs   !#