Method encode_chunk

funasr/models/scama/model.py:306–346

Encode chunk. Args: speech: Speech audio tensor, shape (batch, time). speech_lengths: Length of each speech sample. cache: State cache dict for streaming inference. **kwargs: Additional keyword arguments.

(
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
        cache: dict = None,
        **kwargs,
    )

        return encoder_out, encoder_out_lens

    def encode_chunk(
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
        cache: dict = None,
        **kwargs,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Encode chunk.
        
            Args:
                speech: Speech audio tensor, shape (batch, time).
                speech_lengths: Length of each speech sample.
                cache: State cache dict for streaming inference.
                **kwargs: Additional keyword arguments.
            """
        if cache is None:
            cache = {}
        """Frontend + Encoder. Note that this method is used by asr_inference.py
        Args:
                speech: (Batch, Length, ...)
                speech_lengths: (Batch, )
                ind: int
        """
        with autocast(False):

            # Data augmentation
            if self.specaug is not None and self.training:
                speech, speech_lengths = self.specaug(speech, speech_lengths)

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

        # Forward encoder
        encoder_out, encoder_out_lens, _ = self.encoder.forward_chunk(
            speech, speech_lengths, cache=cache["encoder"]
        )
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]

        return encoder_out, torch.tensor([encoder_out.size(1)])

    def calc_predictor_chunk(self, encoder_out, encoder_out_lens, cache=None, **kwargs):
        """Calc predictor chunk.