o i@sDdZddlZddlZddlmZmZddZGdddejjZ dS)z"Tacotron2 encoder related modules.N)pack_padded_sequencepad_packed_sequencecCs2t|tjjrtjj|jtjjddSdS)zInitialize encoder parameters.reluN) isinstancetorchnnConv1dinitxavier_uniform_weightcalculate_gain)mra/home/ubuntu/.local/lib/python3.10/site-packages/espnet/nets/pytorch_backend/tacotron2/encoder.py encoder_inits rcsJeZdZdZ          dfd d ZdddZddZZS)EncoderaEncoder module of Spectrogram prediction network. This is a module of encoder of Spectrogram prediction network in Tacotron2, which described in `Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions`_. This is the encoder which converts either a sequence of characters or acoustic features into the sequence of hidden states. .. _`Natural TTS Synthesis by Conditioning WaveNet on Mel Spectrogram Predictions`: https://arxiv.org/abs/1712.05884 embedTF?rc  stt|||_| |_|dkrtj|||_n|dkr)tjj ||| d|_nt d||dkrtj |_ t j|D]\} | dkrK|dkrK|n|}| rx|j tjtjj|||d|dddd tj|tjtj| g7_ q?|j tjtjj|||d|dddd tjtj| g7_ q?nd |_ |dkr|dkr|n|}tjj||d|d d d |_nd |_|td S) a;Initialize Tacotron2 encoder module. Args: idim (int) Dimension of the inputs. input_layer (str): Input layer type. embed_dim (int, optional) Dimension of character embedding. elayers (int, optional) The number of encoder blstm layers. eunits (int, optional) The number of encoder blstm units. econv_layers (int, optional) The number of encoder conv layers. econv_filts (int, optional) The number of encoder conv filter size. econv_chans (int, optional) The number of encoder conv filter channels. use_batch_norm (bool, optional) Whether to use batch normalization. use_residual (bool, optional) Whether to use residual connection. dropout_rate (float, optional) Dropout rate. linearr) padding_idxzunknown input_layer: rrF)stridepaddingbiasNT) batch_first bidirectional)superr__init__idim use_residualrrLinearr Embedding ValueError ModuleListconvssixmovesrange Sequentialr BatchNorm1dReLUDropoutLSTMblstmapplyr)selfr" input_layer embed_dimelayerseunits econv_layers econv_chans econv_filtsuse_batch_normr# dropout_raterlayerichansiunits __class__rrr!!sh         $ zEncoder.__init__NcCs||dd}|jdur.tjt|jD]}|jr&||j||}q|j||}q|jdur9|ddSt |t j sDt |}t |dd|dd}|j||\}}t|dd\}}||fS)aCalculate forward propagation. Args: xs (Tensor): Batch of the padded sequence. Either character ids (B, Tmax) or acoustic feature (B, Tmax, idim * encoder_reduction_factor). Padded value should be 0. ilens (LongTensor): Batch of lengths of each input batch (B,). Returns: Tensor: Batch of the sequences of encoder states(B, Tmax, eunits). LongTensor: Batch of lengths of each sequence (B,) rrNT)r)r transposer(r)r*r+lenr#r1rrTensortensorrcpuflatten_parametersr)r3xsilensi_hlensrrrforwards      zEncoder.forwardcCs0|d}t|dg}|||ddS)zInference. Args: x (Tensor): The sequeunce of character ids (T,) or acoustic feature (T, idim * encoder_reduction_factor). Returns: Tensor: The sequences of encoder states(T, eunits). r) unsqueezerrEsizerM)r3xrHrIrrr inferences zEncoder.inference) rrrrrrrTFrr)N)__name__ __module__ __qualname____doc__r!rMrQ __classcell__rrr@rrs  ^ r) rUr)rtorch.nn.utils.rnnrrrrModulerrrrrs