Merge pull request #1825 from modelscope/dev_libt

Shi Xian

2024-06-18 6c467e6f0abfc6d20d0621fbbf67b4dbd81776cc

 funasr/models/transformer/model.py

@@ -10,442 +10,450 @@
from funasr.models.ctc.ctc import CTC
from funasr.models.transformer.utils.add_sos_eos import add_sos_eos
from funasr.metrics.compute_acc import th_accuracy

# from funasr.models.e2e_asr_common import ErrorCalculator
from funasr.train_utils.device_funcs import force_gatherable
from funasr.utils.load_utils import load_audio_and_text_image_video, extract_fbank
from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
from funasr.utils import postprocess_utils
from funasr.utils.datadir_writer import DatadirWriter
from funasr.register import tables


@tables.register("model_classes", "Transformer")
class Transformer(nn.Module):
   """CTC-attention hybrid Encoder-Decoder model"""
    """CTC-attention hybrid Encoder-Decoder model"""

	
   def __init__(
      self,
      frontend: Optional[str] = None,
      frontend_conf: Optional[Dict] = None,
      specaug: Optional[str] = None,
      specaug_conf: Optional[Dict] = None,
      normalize: str = None,
      normalize_conf: Optional[Dict] = None,
      encoder: str = None,
      encoder_conf: Optional[Dict] = None,
      decoder: str = None,
      decoder_conf: Optional[Dict] = None,
      ctc: str = None,
      ctc_conf: Optional[Dict] = None,
      ctc_weight: float = 0.5,
      interctc_weight: float = 0.0,
      input_size: int = 80,
      vocab_size: int = -1,
      ignore_id: int = -1,
      blank_id: int = 0,
      sos: int = 1,
      eos: int = 2,
      lsm_weight: float = 0.0,
      length_normalized_loss: bool = False,
      report_cer: bool = True,
      report_wer: bool = True,
      sym_space: str = "<space>",
      sym_blank: str = "<blank>",
      # extract_feats_in_collect_stats: bool = True,
      share_embedding: bool = False,
      # preencoder: Optional[AbsPreEncoder] = None,
      # postencoder: Optional[AbsPostEncoder] = None,
      **kwargs,
   ):
    def __init__(
        self,
        specaug: str = None,
        specaug_conf: dict = None,
        normalize: str = None,
        normalize_conf: dict = None,
        encoder: str = None,
        encoder_conf: dict = None,
        decoder: str = None,
        decoder_conf: dict = None,
        ctc: str = None,
        ctc_conf: dict = None,
        ctc_weight: float = 0.5,
        interctc_weight: float = 0.0,
        input_size: int = 80,
        vocab_size: int = -1,
        ignore_id: int = -1,
        blank_id: int = 0,
        sos: int = 1,
        eos: int = 2,
        lsm_weight: float = 0.0,
        length_normalized_loss: bool = False,
        report_cer: bool = True,
        report_wer: bool = True,
        sym_space: str = "<space>",
        sym_blank: str = "<blank>",
        # extract_feats_in_collect_stats: bool = True,
        share_embedding: bool = False,
        # preencoder: Optional[AbsPreEncoder] = None,
        # postencoder: Optional[AbsPostEncoder] = None,
        **kwargs,
    ):

      super().__init__()
        super().__init__()

      if frontend is not None:
         frontend_class = tables.frontend_classes.get_class(frontend.lower())
         frontend = frontend_class(**frontend_conf)
      if specaug is not None:
         specaug_class = tables.specaug_classes.get_class(specaug.lower())
         specaug = specaug_class(**specaug_conf)
      if normalize is not None:
         normalize_class = tables.normalize_classes.get_class(normalize.lower())
         normalize = normalize_class(**normalize_conf)
      encoder_class = tables.encoder_classes.get_class(encoder.lower())
      encoder = encoder_class(input_size=input_size, **encoder_conf)
      encoder_output_size = encoder.output_size()
      if decoder is not None:
         decoder_class = tables.decoder_classes.get_class(decoder.lower())
         decoder = decoder_class(
            vocab_size=vocab_size,
            encoder_output_size=encoder_output_size,
            **decoder_conf,
         )
      if ctc_weight > 0.0:
			
         if ctc_conf is None:
            ctc_conf = {}
			
         ctc = CTC(
            odim=vocab_size, encoder_output_size=encoder_output_size, **ctc_conf
         )
	
      self.blank_id = blank_id
      self.sos = sos if sos is not None else vocab_size - 1
      self.eos = eos if eos is not None else vocab_size - 1
      self.vocab_size = vocab_size
      self.ignore_id = ignore_id
      self.ctc_weight = ctc_weight
      self.frontend = frontend
      self.specaug = specaug
      self.normalize = normalize
      self.encoder = encoder
        if specaug is not None:
            specaug_class = tables.specaug_classes.get(specaug)
            specaug = specaug_class(**specaug_conf)
        if normalize is not None:
            normalize_class = tables.normalize_classes.get(normalize)
            normalize = normalize_class(**normalize_conf)
        encoder_class = tables.encoder_classes.get(encoder)
        encoder = encoder_class(input_size=input_size, **encoder_conf)
        encoder_output_size = encoder.output_size()
        if decoder is not None:
            decoder_class = tables.decoder_classes.get(decoder)
            decoder = decoder_class(
                vocab_size=vocab_size,
                encoder_output_size=encoder_output_size,
                **decoder_conf,
            )
        if ctc_weight > 0.0:

      if not hasattr(self.encoder, "interctc_use_conditioning"):
         self.encoder.interctc_use_conditioning = False
      if self.encoder.interctc_use_conditioning:
         self.encoder.conditioning_layer = torch.nn.Linear(
            vocab_size, self.encoder.output_size()
         )
      self.interctc_weight = interctc_weight
            if ctc_conf is None:
                ctc_conf = {}

      # self.error_calculator = None
      if ctc_weight == 1.0:
         self.decoder = None
      else:
         self.decoder = decoder
		
      self.criterion_att = LabelSmoothingLoss(
         size=vocab_size,
         padding_idx=ignore_id,
         smoothing=lsm_weight,
         normalize_length=length_normalized_loss,
      )
      #
      # if report_cer or report_wer:
      #    self.error_calculator = ErrorCalculator(
      #       token_list, sym_space, sym_blank, report_cer, report_wer
      #    )
      #
      if ctc_weight == 0.0:
         self.ctc = None
      else:
         self.ctc = ctc
			
      self.share_embedding = share_embedding
      if self.share_embedding:
         self.decoder.embed = None
		
      self.length_normalized_loss = length_normalized_loss
      self.beam_search = None
	
   def forward(
      self,
      speech: torch.Tensor,
      speech_lengths: torch.Tensor,
      text: torch.Tensor,
      text_lengths: torch.Tensor,
      **kwargs,
   ) -> Tuple[torch.Tensor, Dict[str, torch.Tensor], torch.Tensor]:
      """Encoder + Decoder + Calc loss
      Args:
            speech: (Batch, Length, ...)
            speech_lengths: (Batch, )
            text: (Batch, Length)
            text_lengths: (Batch,)
      """
      # import pdb;
      # pdb.set_trace()
      if len(text_lengths.size()) > 1:
         text_lengths = text_lengths[:, 0]
      if len(speech_lengths.size()) > 1:
         speech_lengths = speech_lengths[:, 0]
		
      batch_size = speech.shape[0]
		
      # 1. Encoder
      encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
      intermediate_outs = None
      if isinstance(encoder_out, tuple):
         intermediate_outs = encoder_out[1]
         encoder_out = encoder_out[0]
		
      loss_att, acc_att, cer_att, wer_att = None, None, None, None
      loss_ctc, cer_ctc = None, None
      stats = dict()
		
      # decoder: CTC branch
      if self.ctc_weight != 0.0:
         loss_ctc, cer_ctc = self._calc_ctc_loss(
            encoder_out, encoder_out_lens, text, text_lengths
         )
			
         # Collect CTC branch stats
         stats["loss_ctc"] = loss_ctc.detach() if loss_ctc is not None else None
         stats["cer_ctc"] = cer_ctc
		
      # Intermediate CTC (optional)
      loss_interctc = 0.0
      if self.interctc_weight != 0.0 and intermediate_outs is not None:
         for layer_idx, intermediate_out in intermediate_outs:
            # we assume intermediate_out has the same length & padding
            # as those of encoder_out
            loss_ic, cer_ic = self._calc_ctc_loss(
               intermediate_out, encoder_out_lens, text, text_lengths
            )
            loss_interctc = loss_interctc + loss_ic
				
            # Collect Intermedaite CTC stats
            stats["loss_interctc_layer{}".format(layer_idx)] = (
               loss_ic.detach() if loss_ic is not None else None
            )
            stats["cer_interctc_layer{}".format(layer_idx)] = cer_ic
			
         loss_interctc = loss_interctc / len(intermediate_outs)
			
         # calculate whole encoder loss
         loss_ctc = (
                       1 - self.interctc_weight
                    ) * loss_ctc + self.interctc_weight * loss_interctc
		
      # decoder: Attention decoder branch
      loss_att, acc_att, cer_att, wer_att = self._calc_att_loss(
         encoder_out, encoder_out_lens, text, text_lengths
      )
		
      # 3. CTC-Att loss definition
      if self.ctc_weight == 0.0:
         loss = loss_att
      elif self.ctc_weight == 1.0:
         loss = loss_ctc
      else:
         loss = self.ctc_weight * loss_ctc + (1 - self.ctc_weight) * loss_att
		
      # Collect Attn branch stats
      stats["loss_att"] = loss_att.detach() if loss_att is not None else None
      stats["acc"] = acc_att
      stats["cer"] = cer_att
      stats["wer"] = wer_att
		
      # Collect total loss stats
      stats["loss"] = torch.clone(loss.detach())
		
      # force_gatherable: to-device and to-tensor if scalar for DataParallel
      if self.length_normalized_loss:
         batch_size = int((text_lengths + 1).sum())
      loss, stats, weight = force_gatherable((loss, stats, batch_size), loss.device)
      return loss, stats, weight
	
            ctc = CTC(odim=vocab_size, encoder_output_size=encoder_output_size, **ctc_conf)

   def encode(
      self, speech: torch.Tensor, speech_lengths: torch.Tensor, **kwargs,
   ) -> Tuple[torch.Tensor, torch.Tensor]:
      """Frontend + Encoder. Note that this method is used by asr_inference.py
      Args:
            speech: (Batch, Length, ...)
            speech_lengths: (Batch, )
            ind: int
      """
      with autocast(False):
        self.blank_id = blank_id
        self.sos = sos if sos is not None else vocab_size - 1
        self.eos = eos if eos is not None else vocab_size - 1
        self.vocab_size = vocab_size
        self.ignore_id = ignore_id
        self.ctc_weight = ctc_weight
        self.specaug = specaug
        self.normalize = normalize
        self.encoder = encoder

         # 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
      # feats: (Batch, Length, Dim)
      # -> encoder_out: (Batch, Length2, Dim2)
      if self.encoder.interctc_use_conditioning:
         encoder_out, encoder_out_lens, _ = self.encoder(
            speech, speech_lengths, ctc=self.ctc
         )
      else:
         encoder_out, encoder_out_lens, _ = self.encoder(speech, speech_lengths)
      intermediate_outs = None
      if isinstance(encoder_out, tuple):
         intermediate_outs = encoder_out[1]
         encoder_out = encoder_out[0]
		
      if intermediate_outs is not None:
         return (encoder_out, intermediate_outs), encoder_out_lens
		
      return encoder_out, encoder_out_lens
	
   def _calc_att_loss(
      self,
      encoder_out: torch.Tensor,
      encoder_out_lens: torch.Tensor,
      ys_pad: torch.Tensor,
      ys_pad_lens: torch.Tensor,
   ):
      ys_in_pad, ys_out_pad = add_sos_eos(ys_pad, self.sos, self.eos, self.ignore_id)
      ys_in_lens = ys_pad_lens + 1
		
      # 1. Forward decoder
      decoder_out, _ = self.decoder(
         encoder_out, encoder_out_lens, ys_in_pad, ys_in_lens
      )
		
      # 2. Compute attention loss
      loss_att = self.criterion_att(decoder_out, ys_out_pad)
      acc_att = th_accuracy(
         decoder_out.view(-1, self.vocab_size),
         ys_out_pad,
         ignore_label=self.ignore_id,
      )
		
      # Compute cer/wer using attention-decoder
      if self.training or self.error_calculator is None:
         cer_att, wer_att = None, None
      else:
         ys_hat = decoder_out.argmax(dim=-1)
         cer_att, wer_att = self.error_calculator(ys_hat.cpu(), ys_pad.cpu())
		
      return loss_att, acc_att, cer_att, wer_att
	
   def _calc_ctc_loss(
      self,
      encoder_out: torch.Tensor,
      encoder_out_lens: torch.Tensor,
      ys_pad: torch.Tensor,
      ys_pad_lens: torch.Tensor,
   ):
      # Calc CTC loss
      loss_ctc = self.ctc(encoder_out, encoder_out_lens, ys_pad, ys_pad_lens)
		
      # Calc CER using CTC
      cer_ctc = None
      if not self.training and self.error_calculator is not None:
         ys_hat = self.ctc.argmax(encoder_out).data
         cer_ctc = self.error_calculator(ys_hat.cpu(), ys_pad.cpu(), is_ctc=True)
      return loss_ctc, cer_ctc
	
   def init_beam_search(self,
                        **kwargs,
                        ):
      from funasr.models.transformer.search import BeamSearch
      from funasr.models.transformer.scorers.ctc import CTCPrefixScorer
      from funasr.models.transformer.scorers.length_bonus import LengthBonus
	
      # 1. Build ASR model
      scorers = {}
		
      if self.ctc != None:
         ctc = CTCPrefixScorer(ctc=self.ctc, eos=self.eos)
         scorers.update(
            ctc=ctc
         )
      token_list = kwargs.get("token_list")
      scorers.update(
         length_bonus=LengthBonus(len(token_list)),
      )
        if not hasattr(self.encoder, "interctc_use_conditioning"):
            self.encoder.interctc_use_conditioning = False
        if self.encoder.interctc_use_conditioning:
            self.encoder.conditioning_layer = torch.nn.Linear(
                vocab_size, self.encoder.output_size()
            )
        self.interctc_weight = interctc_weight

		
      # 3. Build ngram model
      # ngram is not supported now
      ngram = None
      scorers["ngram"] = ngram
		
      weights = dict(
         decoder=1.0 - kwargs.get("decoding_ctc_weight"),
         ctc=kwargs.get("decoding_ctc_weight", 0.0),
         lm=kwargs.get("lm_weight", 0.0),
         ngram=kwargs.get("ngram_weight", 0.0),
         length_bonus=kwargs.get("penalty", 0.0),
      )
      beam_search = BeamSearch(
         beam_size=kwargs.get("beam_size", 2),
         weights=weights,
         scorers=scorers,
         sos=self.sos,
         eos=self.eos,
         vocab_size=len(token_list),
         token_list=token_list,
         pre_beam_score_key=None if self.ctc_weight == 1.0 else "full",
      )
      # beam_search.to(device=kwargs.get("device", "cpu"), dtype=getattr(torch, kwargs.get("dtype", "float32"))).eval()
      # for scorer in scorers.values():
      #    if isinstance(scorer, torch.nn.Module):
      #       scorer.to(device=kwargs.get("device", "cpu"), dtype=getattr(torch, kwargs.get("dtype", "float32"))).eval()
      self.beam_search = beam_search
		
   def generate(self,
             data_in: list,
             data_lengths: list=None,
             key: list=None,
             tokenizer=None,
             **kwargs,
             ):
		
      if kwargs.get("batch_size", 1) > 1:
         raise NotImplementedError("batch decoding is not implemented")
		
      # init beamsearch
      is_use_ctc = kwargs.get("decoding_ctc_weight", 0.0) > 0.00001 and self.ctc != None
      is_use_lm = kwargs.get("lm_weight", 0.0) > 0.00001 and kwargs.get("lm_file", None) is not None
      if self.beam_search is None and (is_use_lm or is_use_ctc):
         logging.info("enable beam_search")
         self.init_beam_search(**kwargs)
         self.nbest = kwargs.get("nbest", 1)
		
      meta_data = {}
      # extract fbank feats
      time1 = time.perf_counter()
      audio_sample_list = load_audio_and_text_image_video(data_in, fs=self.frontend.fs, audio_fs=kwargs.get("fs", 16000))
      time2 = time.perf_counter()
      meta_data["load_data"] = f"{time2 - time1:0.3f}"
      speech, speech_lengths = extract_fbank(audio_sample_list, data_type=kwargs.get("data_type", "sound"), frontend=self.frontend)
      time3 = time.perf_counter()
      meta_data["extract_feat"] = f"{time3 - time2:0.3f}"
      meta_data["batch_data_time"] = speech_lengths.sum().item() * self.frontend.frame_shift * self.frontend.lfr_n / 1000
		
      speech.to(device=kwargs["device"]), speech_lengths.to(device=kwargs["device"])
        # self.error_calculator = None
        if ctc_weight == 1.0:
            self.decoder = None
        else:
            self.decoder = decoder

      # Encoder
      encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
      if isinstance(encoder_out, tuple):
         encoder_out = encoder_out[0]
		
      # c. Passed the encoder result and the beam search
      nbest_hyps = self.beam_search(
         x=encoder_out[0], maxlenratio=kwargs.get("maxlenratio", 0.0), minlenratio=kwargs.get("minlenratio", 0.0)
      )
		
      nbest_hyps = nbest_hyps[: self.nbest]
        self.criterion_att = LabelSmoothingLoss(
            size=vocab_size,
            padding_idx=ignore_id,
            smoothing=lsm_weight,
            normalize_length=length_normalized_loss,
        )
        #
        # if report_cer or report_wer:
        #     self.error_calculator = ErrorCalculator(
        #         token_list, sym_space, sym_blank, report_cer, report_wer
        #     )
        #
        self.error_calculator = None
        if ctc_weight == 0.0:
            self.ctc = None
        else:
            self.ctc = ctc

        self.share_embedding = share_embedding
        if self.share_embedding:
            self.decoder.embed = None

      results = []
      b, n, d = encoder_out.size()
      for i in range(b):
        self.length_normalized_loss = length_normalized_loss
        self.beam_search = None

         for nbest_idx, hyp in enumerate(nbest_hyps):
            ibest_writer = None
            if ibest_writer is None and kwargs.get("output_dir") is not None:
               writer = DatadirWriter(kwargs.get("output_dir"))
               ibest_writer = writer[f"{nbest_idx+1}best_recog"]
            # remove sos/eos and get results
            last_pos = -1
            if isinstance(hyp.yseq, list):
               token_int = hyp.yseq[1:last_pos]
            else:
               token_int = hyp.yseq[1:last_pos].tolist()
					
            # remove blank symbol id, which is assumed to be 0
            token_int = list(filter(lambda x: x != self.eos and x != self.sos and x != self.blank_id, token_int))
				
            # Change integer-ids to tokens
            token = tokenizer.ids2tokens(token_int)
            text = tokenizer.tokens2text(token)
				
            text_postprocessed, _ = postprocess_utils.sentence_postprocess(token)
            result_i = {"key": key[i], "token": token, "text": text, "text_postprocessed": text_postprocessed}
            results.append(result_i)
				
            if ibest_writer is not None:
               ibest_writer["token"][key[i]] = " ".join(token)
               ibest_writer["text"][key[i]] = text
               ibest_writer["text_postprocessed"][key[i]] = text_postprocessed
		
      return results, meta_data
    def forward(
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
        text: torch.Tensor,
        text_lengths: torch.Tensor,
        **kwargs,
    ) -> Tuple[torch.Tensor, Dict[str, torch.Tensor], torch.Tensor]:
        """Encoder + Decoder + Calc loss
        Args:
                speech: (Batch, Length, ...)
                speech_lengths: (Batch, )
                text: (Batch, Length)
                text_lengths: (Batch,)
        """
        if len(text_lengths.size()) > 1:
            text_lengths = text_lengths[:, 0]
        if len(speech_lengths.size()) > 1:
            speech_lengths = speech_lengths[:, 0]

        batch_size = speech.shape[0]

        # 1. Encoder
        encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
        intermediate_outs = None
        if isinstance(encoder_out, tuple):
            intermediate_outs = encoder_out[1]
            encoder_out = encoder_out[0]

        loss_att, acc_att, cer_att, wer_att = None, None, None, None
        loss_ctc, cer_ctc = None, None
        stats = dict()

        # decoder: CTC branch
        if self.ctc_weight != 0.0:
            loss_ctc, cer_ctc = self._calc_ctc_loss(
                encoder_out, encoder_out_lens, text, text_lengths
            )

            # Collect CTC branch stats
            stats["loss_ctc"] = loss_ctc.detach() if loss_ctc is not None else None
            stats["cer_ctc"] = cer_ctc

        # Intermediate CTC (optional)
        loss_interctc = 0.0
        if self.interctc_weight != 0.0 and intermediate_outs is not None:
            for layer_idx, intermediate_out in intermediate_outs:
                # we assume intermediate_out has the same length & padding
                # as those of encoder_out
                loss_ic, cer_ic = self._calc_ctc_loss(
                    intermediate_out, encoder_out_lens, text, text_lengths
                )
                loss_interctc = loss_interctc + loss_ic

                # Collect Intermedaite CTC stats
                stats["loss_interctc_layer{}".format(layer_idx)] = (
                    loss_ic.detach() if loss_ic is not None else None
                )
                stats["cer_interctc_layer{}".format(layer_idx)] = cer_ic

            loss_interctc = loss_interctc / len(intermediate_outs)

            # calculate whole encoder loss
            loss_ctc = (1 - self.interctc_weight) * loss_ctc + self.interctc_weight * loss_interctc

        # decoder: Attention decoder branch
        loss_att, acc_att, cer_att, wer_att = self._calc_att_loss(
            encoder_out, encoder_out_lens, text, text_lengths
        )

        # 3. CTC-Att loss definition
        if self.ctc_weight == 0.0:
            loss = loss_att
        elif self.ctc_weight == 1.0:
            loss = loss_ctc
        else:
            loss = self.ctc_weight * loss_ctc + (1 - self.ctc_weight) * loss_att

        # Collect Attn branch stats
        stats["loss_att"] = loss_att.detach() if loss_att is not None else None
        stats["acc"] = acc_att
        stats["cer"] = cer_att
        stats["wer"] = wer_att

        # Collect total loss stats
        stats["loss"] = torch.clone(loss.detach())

        # force_gatherable: to-device and to-tensor if scalar for DataParallel
        if self.length_normalized_loss:
            batch_size = int((text_lengths + 1).sum())
        loss, stats, weight = force_gatherable((loss, stats, batch_size), loss.device)
        return loss, stats, weight

    def encode(
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
        **kwargs,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """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
        # feats: (Batch, Length, Dim)
        # -> encoder_out: (Batch, Length2, Dim2)
        if self.encoder.interctc_use_conditioning:
            encoder_out, encoder_out_lens, _ = self.encoder(speech, speech_lengths, ctc=self.ctc)
        else:
            encoder_out, encoder_out_lens, _ = self.encoder(speech, speech_lengths)
        intermediate_outs = None
        if isinstance(encoder_out, tuple):
            intermediate_outs = encoder_out[1]
            encoder_out = encoder_out[0]

        if intermediate_outs is not None:
            return (encoder_out, intermediate_outs), encoder_out_lens

        return encoder_out, encoder_out_lens

    def _calc_att_loss(
        self,
        encoder_out: torch.Tensor,
        encoder_out_lens: torch.Tensor,
        ys_pad: torch.Tensor,
        ys_pad_lens: torch.Tensor,
    ):
        ys_in_pad, ys_out_pad = add_sos_eos(ys_pad, self.sos, self.eos, self.ignore_id)
        ys_in_lens = ys_pad_lens + 1

        # 1. Forward decoder
        decoder_out, _ = self.decoder(encoder_out, encoder_out_lens, ys_in_pad, ys_in_lens)

        # 2. Compute attention loss
        loss_att = self.criterion_att(decoder_out, ys_out_pad)
        acc_att = th_accuracy(
            decoder_out.view(-1, self.vocab_size),
            ys_out_pad,
            ignore_label=self.ignore_id,
        )

        # Compute cer/wer using attention-decoder
        if self.training or self.error_calculator is None:
            cer_att, wer_att = None, None
        else:
            ys_hat = decoder_out.argmax(dim=-1)
            cer_att, wer_att = self.error_calculator(ys_hat.cpu(), ys_pad.cpu())

        return loss_att, acc_att, cer_att, wer_att

    def _calc_ctc_loss(
        self,
        encoder_out: torch.Tensor,
        encoder_out_lens: torch.Tensor,
        ys_pad: torch.Tensor,
        ys_pad_lens: torch.Tensor,
    ):
        # Calc CTC loss
        loss_ctc = self.ctc(encoder_out, encoder_out_lens, ys_pad, ys_pad_lens)

        # Calc CER using CTC
        cer_ctc = None
        if not self.training and self.error_calculator is not None:
            ys_hat = self.ctc.argmax(encoder_out).data
            cer_ctc = self.error_calculator(ys_hat.cpu(), ys_pad.cpu(), is_ctc=True)
        return loss_ctc, cer_ctc

    def init_beam_search(
        self,
        **kwargs,
    ):
        from funasr.models.transformer.search import BeamSearch
        from funasr.models.transformer.scorers.ctc import CTCPrefixScorer
        from funasr.models.transformer.scorers.length_bonus import LengthBonus

        # 1. Build ASR model
        scorers = {}

        if self.ctc != None:
            ctc = CTCPrefixScorer(ctc=self.ctc, eos=self.eos)
            scorers.update(ctc=ctc)
        token_list = kwargs.get("token_list")
        scorers.update(
            decoder=self.decoder,
            length_bonus=LengthBonus(len(token_list)),
        )

        # 3. Build ngram model
        # ngram is not supported now
        ngram = None
        scorers["ngram"] = ngram

        weights = dict(
            decoder=1.0 - kwargs.get("decoding_ctc_weight", 0.5),
            ctc=kwargs.get("decoding_ctc_weight", 0.5),
            lm=kwargs.get("lm_weight", 0.0),
            ngram=kwargs.get("ngram_weight", 0.0),
            length_bonus=kwargs.get("penalty", 0.0),
        )
        beam_search = BeamSearch(
            beam_size=kwargs.get("beam_size", 10),
            weights=weights,
            scorers=scorers,
            sos=self.sos,
            eos=self.eos,
            vocab_size=len(token_list),
            token_list=token_list,
            pre_beam_score_key=None if self.ctc_weight == 1.0 else "full",
        )

        self.beam_search = beam_search

    def inference(
        self,
        data_in,
        data_lengths=None,
        key: list = None,
        tokenizer=None,
        frontend=None,
        **kwargs,
    ):

        if kwargs.get("batch_size", 1) > 1:
            raise NotImplementedError("batch decoding is not implemented")

        # init beamsearch
        if self.beam_search is None:
            logging.info("enable beam_search")
            self.init_beam_search(**kwargs)
            self.nbest = kwargs.get("nbest", 1)

        meta_data = {}
        if (
            isinstance(data_in, torch.Tensor) and kwargs.get("data_type", "sound") == "fbank"
        ):  # fbank
            speech, speech_lengths = data_in, data_lengths
            if len(speech.shape) < 3:
                speech = speech[None, :, :]
            if speech_lengths is None:
                speech_lengths = speech.shape[1]
        else:
            # extract fbank feats
            time1 = time.perf_counter()
            audio_sample_list = load_audio_text_image_video(
                data_in,
                fs=frontend.fs,
                audio_fs=kwargs.get("fs", 16000),
                data_type=kwargs.get("data_type", "sound"),
                tokenizer=tokenizer,
            )
            time2 = time.perf_counter()
            meta_data["load_data"] = f"{time2 - time1:0.3f}"
            speech, speech_lengths = extract_fbank(
                audio_sample_list, data_type=kwargs.get("data_type", "sound"), frontend=frontend
            )
            time3 = time.perf_counter()
            meta_data["extract_feat"] = f"{time3 - time2:0.3f}"
            meta_data["batch_data_time"] = (
                speech_lengths.sum().item() * frontend.frame_shift * frontend.lfr_n / 1000
            )

        speech = speech.to(device=kwargs["device"])
        speech_lengths = speech_lengths.to(device=kwargs["device"])
        # Encoder
        encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]

        # c. Passed the encoder result and the beam search
        nbest_hyps = self.beam_search(
            x=encoder_out[0],
            maxlenratio=kwargs.get("maxlenratio", 0.0),
            minlenratio=kwargs.get("minlenratio", 0.0),
        )

        nbest_hyps = nbest_hyps[: self.nbest]

        results = []
        b, n, d = encoder_out.size()
        for i in range(b):

            for nbest_idx, hyp in enumerate(nbest_hyps):
                ibest_writer = None
                if kwargs.get("output_dir") is not None:
                    if not hasattr(self, "writer"):
                        self.writer = DatadirWriter(kwargs.get("output_dir"))
                    ibest_writer = self.writer[f"{nbest_idx + 1}best_recog"]

                # remove sos/eos and get results
                last_pos = -1
                if isinstance(hyp.yseq, list):
                    token_int = hyp.yseq[1:last_pos]
                else:
                    token_int = hyp.yseq[1:last_pos].tolist()

                # remove blank symbol id, which is assumed to be 0
                token_int = list(
                    filter(
                        lambda x: x != self.eos and x != self.sos and x != self.blank_id, token_int
                    )
                )

                # Change integer-ids to tokens
                token = tokenizer.ids2tokens(token_int)
                text = tokenizer.tokens2text(token)

                text_postprocessed, _ = postprocess_utils.sentence_postprocess(token)
                result_i = {"key": key[i], "token": token, "text": text_postprocessed}
                results.append(result_i)

                if ibest_writer is not None:
                    ibest_writer["token"][key[i]] = " ".join(token)
                    ibest_writer["text"][key[i]] = text_postprocessed

        return results, meta_data