Method init_state

funasr/models/transducer/rnn_decoder.py:313–345

Init state. Args: x: TODO.

(self, x)

        return z_all, ys_in_lens

    def init_state(self, x):
        # to support mutiple encoder asr mode, in single encoder mode,
        # convert torch.Tensor to List of torch.Tensor
        """Init state.
        
            Args:
                x: TODO.
            """
        if self.num_encs == 1:
            x = [x]

        c_list = [self.zero_state(x[0].unsqueeze(0))]
        z_list = [self.zero_state(x[0].unsqueeze(0))]
        for _ in range(1, self.dlayers):
            c_list.append(self.zero_state(x[0].unsqueeze(0)))
            z_list.append(self.zero_state(x[0].unsqueeze(0)))
        # TODO(karita): support strm_index for `asr_mix`
        strm_index = 0
        att_idx = min(strm_index, len(self.att_list) - 1)
        if self.num_encs == 1:
            a = None
            self.att_list[att_idx].reset()  # reset pre-computation of h
        else:
            a = [None] * (self.num_encs + 1)  # atts + han
            for idx in range(self.num_encs + 1):
                # reset pre-computation of h in atts and han
                self.att_list[idx].reset()
        return dict(
            c_prev=c_list[:],
            z_prev=z_list[:],
            a_prev=a,
            workspace=(att_idx, z_list, c_list),
        )

    def score(self, yseq, state, x):
        # to support mutiple encoder asr mode, in single encoder mode,