~Vji tddlZdgZGddejjZGddejjZdS)N DeepSpeechc ^eZdZdZd dededededdf fd Zd ejdejfd Z xZ S) FullyConnectedzh Args: n_feature: Number of input features n_hidden: Internal hidden unit size.  n_featuren_hiddendropout relu_max_clipreturnNctt|tj||d|_||_||_dS)NT)bias) superr__init__torchnnLinearfcr r )selfrrr r __class__s V/root/voice-cloning/.venv/lib/python3.11/site-packages/torchaudio/models/deepspeech.pyrzFullyConnected.__init__ sM nd##,,...(//)XD/AA* xc<||}tjj|}tjj|d|j}|jr0tjj||j|j}|S)Nr) rrr functionalreluhardtanhr r training)rrs rforwardzFullyConnected.forwardsx GGAJJ H  $ $Q ' ' H  ( (At/A B B < L#++At|T]KKAr)r __name__ __module__ __qualname____doc__intfloatrrTensorr __classcell__rs@rrrs #uUXbf %,rrc deZdZdZ ddedededed d f fd Zd ejd ejfd Z xZ S)ra DeepSpeech architecture introduced in *Deep Speech: Scaling up end-to-end speech recognition* :cite:`hannun2014deep`. Args: n_feature: Number of input features n_hidden: Internal hidden unit size. n_class: Number of output classes (rrn_classr r Nctt|||_t ||||_t ||||_t ||||_tj ||ddd|_ t ||||_ tj |||_dS)NrT) num_layers nonlinearity bidirectional)rrrrrfc1fc2fc3rrRNNbi_rnnfc4rout)rrrr-r rs rrzDeepSpeech.__init__&s j$((***  !)Xw??!(Hg>>!(Hg>>hll8X!RXhllmm !(Hg>>8??8W55rrc2||}||}||}|d}|dd}||\}}|ddddd|jf|dddd|jdfz}||}||}| ddd}tj j |d}|S)z Args: x (torch.Tensor): Tensor of dimension (batch, channel, time, feature). Returns: Tensor: Predictor tensor of dimension (batch, time, class). r/rN)dim)r3r4r5squeeze transposer7rr8r9permuterrr log_softmax)rr_s rrzDeepSpeech.forward6s HHQKK HHQKK HHQKK IIaLL KK1  {{1~~1 aaaOdmO# $qAAAt})>'? ? HHQKK HHQKK IIaA   H  + +A1 + 5 5r)r*r+r,rr(s@rrrs 6666 6  6  666666 %,r)r__all__rModulerrrrrEs{ .UX_,8888888888r