Method encode

funasr/models/xvector/e2e_sv.py:239–271

Frontend + Encoder. Note that this method is used by asr_inference.py Args: speech: (Batch, Length, ...) speech_lengths: (Batch, )

(
        self, speech: torch.Tensor, speech_lengths: torch.Tensor
    )

        return {"feats": feats, "feats_lengths": feats_lengths}

    def encode(
        self, speech: torch.Tensor, speech_lengths: torch.Tensor
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Frontend + Encoder. Note that this method is used by asr_inference.py
        Args:
            speech: (Batch, Length, ...)
            speech_lengths: (Batch, )
        """
        with autocast(False):
            # 1. Extract feats
            feats, feats_lengths = self._extract_feats(speech, speech_lengths)

            # 2. Data augmentation
            if self.specaug is not None and self.training:
                feats, feats_lengths = self.specaug(feats, feats_lengths)

            # 3. Normalization for feature: e.g. Global-CMVN, Utterance-CMVN
            if self.normalize is not None:
                feats, feats_lengths = self.normalize(feats, feats_lengths)

        # Pre-encoder, e.g. used for raw input data
        if self.preencoder is not None:
            feats, feats_lengths = self.preencoder(feats, feats_lengths)

        # 4. Forward encoder
        # feats: (Batch, Length, Dim) -> (Batch, Channel, Length2, Dim2)
        encoder_out, encoder_out_lens = self.encoder(feats, feats_lengths)

        # Post-encoder, e.g. NLU
        if self.postencoder is not None:
            encoder_out, encoder_out_lens = self.postencoder(encoder_out, encoder_out_lens)

        return encoder_out, encoder_out_lens

    def _extract_feats(
        self, speech: torch.Tensor, speech_lengths: torch.Tensor