Revert "shfit to shift (#2266)" (#2336)

zhifu gao

2024-12-25 3f8294b9d7deaa0cbdb0b2ef6f3802d46ae133a9

 funasr/models/paraformer_streaming/model.py

@@ -41,28 +41,31 @@
    Paraformer: Fast and Accurate Parallel Transformer for Non-autoregressive End-to-End Speech Recognition
    https://arxiv.org/abs/2206.08317
    """
    

    def __init__(
        self,
        *args,
        **kwargs,
    ):
        

        super().__init__(*args, **kwargs)
        
        # import pdb;
        # pdb.set_trace()

        self.sampling_ratio = kwargs.get("sampling_ratio", 0.2)


        self.scama_mask = None
        if hasattr(self.encoder, "overlap_chunk_cls") and self.encoder.overlap_chunk_cls is not None:
            from funasr.models.scama.chunk_utilis import build_scama_mask_for_cross_attention_decoder
            self.build_scama_mask_for_cross_attention_decoder_fn = build_scama_mask_for_cross_attention_decoder
        if (
            hasattr(self.encoder, "overlap_chunk_cls")
            and self.encoder.overlap_chunk_cls is not None
        ):
            from funasr.models.scama.chunk_utilis import (
                build_scama_mask_for_cross_attention_decoder,
            )

            self.build_scama_mask_for_cross_attention_decoder_fn = (
                build_scama_mask_for_cross_attention_decoder
            )
            self.decoder_attention_chunk_type = kwargs.get("decoder_attention_chunk_type", "chunk")


    
    def forward(
        self,
        speech: torch.Tensor,
@@ -78,56 +81,57 @@
                text: (Batch, Length)
                text_lengths: (Batch,)
        """
        # import pdb;
        # pdb.set_trace()
        decoding_ind = kwargs.get("decoding_ind")
        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]
        

        # Encoder
        if hasattr(self.encoder, "overlap_chunk_cls"):
            ind = self.encoder.overlap_chunk_cls.random_choice(self.training, decoding_ind)
            encoder_out, encoder_out_lens = self.encode(speech, speech_lengths, ind=ind)
        else:
            encoder_out, encoder_out_lens = self.encode(speech, speech_lengths)
        

        loss_ctc, cer_ctc = None, None
        loss_pre = None
        stats = dict()
        

        # decoder: CTC branch

        if self.ctc_weight > 0.0:
            if hasattr(self.encoder, "overlap_chunk_cls"):
                encoder_out_ctc, encoder_out_lens_ctc = self.encoder.overlap_chunk_cls.remove_chunk(encoder_out,
                                                                                                    encoder_out_lens,
                                                                                                    chunk_outs=None)
                encoder_out_ctc, encoder_out_lens_ctc = self.encoder.overlap_chunk_cls.remove_chunk(
                    encoder_out, encoder_out_lens, chunk_outs=None
                )
            else:
                encoder_out_ctc, encoder_out_lens_ctc = encoder_out, encoder_out_lens
                

            loss_ctc, cer_ctc = self._calc_ctc_loss(
                encoder_out_ctc, encoder_out_lens_ctc, 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
        

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

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

        # Collect Attn branch stats
        stats["loss_att"] = loss_att.detach() if loss_att is not None else None
        stats["pre_loss_att"] = pre_loss_att.detach() if pre_loss_att is not None else None
@@ -135,17 +139,21 @@
        stats["cer"] = cer_att
        stats["wer"] = wer_att
        stats["loss_pre"] = loss_pre.detach().cpu() if loss_pre is not None else None
        

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

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

    def encode_chunk(
        self, speech: torch.Tensor, speech_lengths: torch.Tensor, cache: dict = None, **kwargs,
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
        cache: dict = None,
        **kwargs,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """Frontend + Encoder. Note that this method is used by asr_inference.py
        Args:
@@ -154,22 +162,24 @@
                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"])
        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_att_predictor_loss(
        self,
        encoder_out: torch.Tensor,
@@ -177,41 +187,49 @@
        ys_pad: torch.Tensor,
        ys_pad_lens: torch.Tensor,
    ):
        encoder_out_mask = (~make_pad_mask(encoder_out_lens, maxlen=encoder_out.size(1))[:, None, :]).to(
            encoder_out.device)
        encoder_out_mask = (
            ~make_pad_mask(encoder_out_lens, maxlen=encoder_out.size(1))[:, None, :]
        ).to(encoder_out.device)
        if self.predictor_bias == 1:
            _, ys_pad = add_sos_eos(ys_pad, self.sos, self.eos, self.ignore_id)
            ys_pad_lens = ys_pad_lens + self.predictor_bias
        mask_chunk_predictor = None
        if self.encoder.overlap_chunk_cls is not None:
            mask_chunk_predictor = self.encoder.overlap_chunk_cls.get_mask_chunk_predictor(None,
                                                                                           device=encoder_out.device,
                                                                                           batch_size=encoder_out.size(
                                                                                               0))
            mask_shfit_chunk = self.encoder.overlap_chunk_cls.get_mask_shfit_chunk(None, device=encoder_out.device,
                                                                                   batch_size=encoder_out.size(0))
            mask_chunk_predictor = self.encoder.overlap_chunk_cls.get_mask_chunk_predictor(
                None, device=encoder_out.device, batch_size=encoder_out.size(0)
            )
            mask_shfit_chunk = self.encoder.overlap_chunk_cls.get_mask_shfit_chunk(
                None, device=encoder_out.device, batch_size=encoder_out.size(0)
            )
            encoder_out = encoder_out * mask_shfit_chunk
        pre_acoustic_embeds, pre_token_length, pre_alphas, _ = self.predictor(encoder_out,
                                                                              ys_pad,
                                                                              encoder_out_mask,
                                                                              ignore_id=self.ignore_id,
                                                                              mask_chunk_predictor=mask_chunk_predictor,
                                                                              target_label_length=ys_pad_lens,
                                                                              )
        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(pre_alphas,
                                                                                             encoder_out_lens)
        
        pre_acoustic_embeds, pre_token_length, pre_alphas, _ = self.predictor(
            encoder_out,
            ys_pad,
            encoder_out_mask,
            ignore_id=self.ignore_id,
            mask_chunk_predictor=mask_chunk_predictor,
            target_label_length=ys_pad_lens,
        )
        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(
            pre_alphas, encoder_out_lens
        )

        scama_mask = None
        if self.encoder.overlap_chunk_cls is not None and self.decoder_attention_chunk_type == 'chunk':
        if (
            self.encoder.overlap_chunk_cls is not None
            and self.decoder_attention_chunk_type == "chunk"
        ):
            encoder_chunk_size = self.encoder.overlap_chunk_cls.chunk_size_pad_shift_cur
            attention_chunk_center_bias = 0
            attention_chunk_size = encoder_chunk_size
            decoder_att_look_back_factor = self.encoder.overlap_chunk_cls.decoder_att_look_back_factor_cur
            mask_shift_att_chunk_decoder = self.encoder.overlap_chunk_cls. \
                get_mask_shift_att_chunk_decoder(None,
                                                 device=encoder_out.device,
                                                 batch_size=encoder_out.size(0)
                                                 )
            decoder_att_look_back_factor = (
                self.encoder.overlap_chunk_cls.decoder_att_look_back_factor_cur
            )
            mask_shift_att_chunk_decoder = (
                self.encoder.overlap_chunk_cls.get_mask_shift_att_chunk_decoder(
                    None, device=encoder_out.device, batch_size=encoder_out.size(0)
                )
            )
            scama_mask = self.build_scama_mask_for_cross_attention_decoder_fn(
                predictor_alignments=predictor_alignments,
                encoder_sequence_length=encoder_out_lens,
@@ -228,31 +246,41 @@
                is_training=self.training,
            )
        elif self.encoder.overlap_chunk_cls is not None:
            encoder_out, encoder_out_lens = self.encoder.overlap_chunk_cls.remove_chunk(encoder_out,
                                                                                        encoder_out_lens,
                                                                                        chunk_outs=None)
            encoder_out, encoder_out_lens = self.encoder.overlap_chunk_cls.remove_chunk(
                encoder_out, encoder_out_lens, chunk_outs=None
            )
        # 0. sampler
        decoder_out_1st = None
        pre_loss_att = None
        if self.sampling_ratio > 0.0:

            if self.use_1st_decoder_loss:
                sematic_embeds, decoder_out_1st, pre_loss_att = \
                    self.sampler_with_grad(encoder_out, encoder_out_lens, ys_pad,
                                           ys_pad_lens, pre_acoustic_embeds, scama_mask)
                sematic_embeds, decoder_out_1st, pre_loss_att = self.sampler_with_grad(
                    encoder_out,
                    encoder_out_lens,
                    ys_pad,
                    ys_pad_lens,
                    pre_acoustic_embeds,
                    scama_mask,
                )
            else:
                sematic_embeds, decoder_out_1st = \
                    self.sampler(encoder_out, encoder_out_lens, ys_pad,
                                 ys_pad_lens, pre_acoustic_embeds, scama_mask)
                sematic_embeds, decoder_out_1st = self.sampler(
                    encoder_out,
                    encoder_out_lens,
                    ys_pad,
                    ys_pad_lens,
                    pre_acoustic_embeds,
                    scama_mask,
                )
        else:
            sematic_embeds = pre_acoustic_embeds
        

        # 1. Forward decoder
        decoder_outs = self.decoder(
            encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, scama_mask
        )
        decoder_out, _ = decoder_outs[0], decoder_outs[1]
        

        if decoder_out_1st is None:
            decoder_out_1st = decoder_out
        # 2. Compute attention loss
@@ -263,19 +291,29 @@
            ignore_label=self.ignore_id,
        )
        loss_pre = self.criterion_pre(ys_pad_lens.type_as(pre_token_length), pre_token_length)
        

        # 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_1st.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, loss_pre, pre_loss_att
    
    def sampler(self, encoder_out, encoder_out_lens, ys_pad, ys_pad_lens, pre_acoustic_embeds, chunk_mask=None):
        
        tgt_mask = (~make_pad_mask(ys_pad_lens, maxlen=ys_pad_lens.max())[:, :, None]).to(ys_pad.device)

    def sampler(
        self,
        encoder_out,
        encoder_out_lens,
        ys_pad,
        ys_pad_lens,
        pre_acoustic_embeds,
        chunk_mask=None,
    ):

        tgt_mask = (~make_pad_mask(ys_pad_lens, maxlen=ys_pad_lens.max())[:, :, None]).to(
            ys_pad.device
        )
        ys_pad_masked = ys_pad * tgt_mask[:, :, 0]
        if self.share_embedding:
            ys_pad_embed = self.decoder.output_layer.weight[ys_pad_masked]
@@ -293,52 +331,65 @@
            input_mask = torch.ones_like(nonpad_positions)
            bsz, seq_len = ys_pad.size()
            for li in range(bsz):
                target_num = (((seq_lens[li] - same_num[li].sum()).float()) * self.sampling_ratio).long()
                target_num = (
                    ((seq_lens[li] - same_num[li].sum()).float()) * self.sampling_ratio
                ).long()
                if target_num > 0:
                    input_mask[li].scatter_(dim=0, index=torch.randperm(seq_lens[li])[:target_num].cuda(), value=0)
                    input_mask[li].scatter_(
                        dim=0, index=torch.randperm(seq_lens[li])[:target_num].cuda(), value=0
                    )
            input_mask = input_mask.eq(1)
            input_mask = input_mask.masked_fill(~nonpad_positions, False)
            input_mask_expand_dim = input_mask.unsqueeze(2).to(pre_acoustic_embeds.device)
        
        sematic_embeds = pre_acoustic_embeds.masked_fill(~input_mask_expand_dim, 0) + ys_pad_embed.masked_fill(
            input_mask_expand_dim, 0)

        sematic_embeds = pre_acoustic_embeds.masked_fill(
            ~input_mask_expand_dim, 0
        ) + ys_pad_embed.masked_fill(input_mask_expand_dim, 0)
        return sematic_embeds * tgt_mask, decoder_out * tgt_mask
    

    def calc_predictor(self, encoder_out, encoder_out_lens):
        
        encoder_out_mask = (~make_pad_mask(encoder_out_lens, maxlen=encoder_out.size(1))[:, None, :]).to(
            encoder_out.device)

        encoder_out_mask = (
            ~make_pad_mask(encoder_out_lens, maxlen=encoder_out.size(1))[:, None, :]
        ).to(encoder_out.device)
        mask_chunk_predictor = None
        if self.encoder.overlap_chunk_cls is not None:
            mask_chunk_predictor = self.encoder.overlap_chunk_cls.get_mask_chunk_predictor(None,
                                                                                           device=encoder_out.device,
                                                                                           batch_size=encoder_out.size(
                                                                                               0))
            mask_shfit_chunk = self.encoder.overlap_chunk_cls.get_mask_shfit_chunk(None, device=encoder_out.device,
                                                                                   batch_size=encoder_out.size(0))
            mask_chunk_predictor = self.encoder.overlap_chunk_cls.get_mask_chunk_predictor(
                None, device=encoder_out.device, batch_size=encoder_out.size(0)
            )
            mask_shfit_chunk = self.encoder.overlap_chunk_cls.get_mask_shfit_chunk(
                None, device=encoder_out.device, batch_size=encoder_out.size(0)
            )
            encoder_out = encoder_out * mask_shfit_chunk
        pre_acoustic_embeds, pre_token_length, pre_alphas, pre_peak_index = self.predictor(encoder_out,
                                                                                           None,
                                                                                           encoder_out_mask,
                                                                                           ignore_id=self.ignore_id,
                                                                                           mask_chunk_predictor=mask_chunk_predictor,
                                                                                           target_label_length=None,
                                                                                           )
        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(pre_alphas,
                                                                                             encoder_out_lens + 1 if self.predictor.tail_threshold > 0.0 else encoder_out_lens)
        
        pre_acoustic_embeds, pre_token_length, pre_alphas, pre_peak_index = self.predictor(
            encoder_out,
            None,
            encoder_out_mask,
            ignore_id=self.ignore_id,
            mask_chunk_predictor=mask_chunk_predictor,
            target_label_length=None,
        )
        predictor_alignments, predictor_alignments_len = self.predictor.gen_frame_alignments(
            pre_alphas,
            encoder_out_lens + 1 if self.predictor.tail_threshold > 0.0 else encoder_out_lens,
        )

        scama_mask = None
        if self.encoder.overlap_chunk_cls is not None and self.decoder_attention_chunk_type == 'chunk':
        if (
            self.encoder.overlap_chunk_cls is not None
            and self.decoder_attention_chunk_type == "chunk"
        ):
            encoder_chunk_size = self.encoder.overlap_chunk_cls.chunk_size_pad_shift_cur
            attention_chunk_center_bias = 0
            attention_chunk_size = encoder_chunk_size
            decoder_att_look_back_factor = self.encoder.overlap_chunk_cls.decoder_att_look_back_factor_cur
            mask_shift_att_chunk_decoder = self.encoder.overlap_chunk_cls. \
                get_mask_shift_att_chunk_decoder(None,
                                                 device=encoder_out.device,
                                                 batch_size=encoder_out.size(0)
                                                 )
            decoder_att_look_back_factor = (
                self.encoder.overlap_chunk_cls.decoder_att_look_back_factor_cur
            )
            mask_shift_att_chunk_decoder = (
                self.encoder.overlap_chunk_cls.get_mask_shift_att_chunk_decoder(
                    None, device=encoder_out.device, batch_size=encoder_out.size(0)
                )
            )
            scama_mask = self.build_scama_mask_for_cross_attention_decoder_fn(
                predictor_alignments=predictor_alignments,
                encoder_sequence_length=encoder_out_lens,
@@ -355,30 +406,32 @@
                is_training=self.training,
            )
        self.scama_mask = scama_mask
        

        return pre_acoustic_embeds, pre_token_length, pre_alphas, pre_peak_index
    

    def calc_predictor_chunk(self, encoder_out, encoder_out_lens, cache=None, **kwargs):
        is_final = kwargs.get("is_final", False)

        return self.predictor.forward_chunk(encoder_out, cache["encoder"], is_final=is_final)
    
    def cal_decoder_with_predictor(self, encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens):

    def cal_decoder_with_predictor(
        self, encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens
    ):
        decoder_outs = self.decoder(
            encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, self.scama_mask
        )
        decoder_out = decoder_outs[0]
        decoder_out = torch.log_softmax(decoder_out, dim=-1)
        return decoder_out, ys_pad_lens
    
    def cal_decoder_with_predictor_chunk(self, encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, cache=None):
        decoder_outs = self.decoder.forward_chunk(
            encoder_out, sematic_embeds, cache["decoder"]
        )

    def cal_decoder_with_predictor_chunk(
        self, encoder_out, encoder_out_lens, sematic_embeds, ys_pad_lens, cache=None
    ):
        decoder_outs = self.decoder.forward_chunk(encoder_out, sematic_embeds, cache["decoder"])
        decoder_out = decoder_outs
        decoder_out = torch.log_softmax(decoder_out, dim=-1)
        return decoder_out, ys_pad_lens
    

    def init_cache(self, cache: dict = {}, **kwargs):
        chunk_size = kwargs.get("chunk_size", [0, 10, 5])
        encoder_chunk_look_back = kwargs.get("encoder_chunk_look_back", 0)
@@ -387,51 +440,67 @@

        enc_output_size = kwargs["encoder_conf"]["output_size"]
        feats_dims = kwargs["frontend_conf"]["n_mels"] * kwargs["frontend_conf"]["lfr_m"]
        cache_encoder = {"start_idx": 0, "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
                    "cif_alphas": torch.zeros((batch_size, 1)), "chunk_size": chunk_size,
                    "encoder_chunk_look_back": encoder_chunk_look_back, "last_chunk": False, "opt": None,
                    "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)),
                    "tail_chunk": False}
        cache_encoder = {
            "start_idx": 0,
            "cif_hidden": torch.zeros((batch_size, 1, enc_output_size)),
            "cif_alphas": torch.zeros((batch_size, 1)),
            "chunk_size": chunk_size,
            "encoder_chunk_look_back": encoder_chunk_look_back,
            "last_chunk": False,
            "opt": None,
            "feats": torch.zeros((batch_size, chunk_size[0] + chunk_size[2], feats_dims)),
            "tail_chunk": False,
        }
        cache["encoder"] = cache_encoder
        
        cache_decoder = {"decode_fsmn": None, "decoder_chunk_look_back": decoder_chunk_look_back, "opt": None,
                    "chunk_size": chunk_size}

        cache_decoder = {
            "decode_fsmn": None,
            "decoder_chunk_look_back": decoder_chunk_look_back,
            "opt": None,
            "chunk_size": chunk_size,
        }
        cache["decoder"] = cache_decoder
        cache["frontend"] = {}
        cache["prev_samples"] = torch.empty(0)
        

        return cache
    
    def generate_chunk(self,
                       speech,
                       speech_lengths=None,
                       key: list = None,
                       tokenizer=None,
                       frontend=None,
                       **kwargs,
                       ):

    def generate_chunk(
        self,
        speech,
        speech_lengths=None,
        key: list = None,
        tokenizer=None,
        frontend=None,
        **kwargs,
    ):
        cache = kwargs.get("cache", {})
        speech = speech.to(device=kwargs["device"])
        speech_lengths = speech_lengths.to(device=kwargs["device"])
        

        # Encoder
        encoder_out, encoder_out_lens = self.encode_chunk(speech, speech_lengths, cache=cache, is_final=kwargs.get("is_final", False))
        encoder_out, encoder_out_lens = self.encode_chunk(
            speech, speech_lengths, cache=cache, is_final=kwargs.get("is_final", False)
        )
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]
        

        # predictor
        predictor_outs = self.calc_predictor_chunk(encoder_out, encoder_out_lens, cache=cache, is_final=kwargs.get("is_final", False))
        pre_acoustic_embeds, pre_token_length, alphas, pre_peak_index = predictor_outs[0], predictor_outs[1], \
                                                                        predictor_outs[2], predictor_outs[3]
        predictor_outs = self.calc_predictor_chunk(
            encoder_out, encoder_out_lens, cache=cache, is_final=kwargs.get("is_final", False)
        )
        pre_acoustic_embeds, pre_token_length, alphas, pre_peak_index = (
            predictor_outs[0],
            predictor_outs[1],
            predictor_outs[2],
            predictor_outs[3],
        )
        pre_token_length = pre_token_length.round().long()
        if torch.max(pre_token_length) < 1:
            return []
        decoder_outs = self.cal_decoder_with_predictor_chunk(encoder_out,
                                                             encoder_out_lens,
                                                             pre_acoustic_embeds,
                                                             pre_token_length,
                                                             cache=cache
                                                             )
        decoder_outs = self.cal_decoder_with_predictor_chunk(
            encoder_out, encoder_out_lens, pre_acoustic_embeds, pre_token_length, cache=cache
        )
        decoder_out, ys_pad_lens = decoder_outs[0], decoder_outs[1]

        results = []
@@ -439,119 +508,144 @@
        if isinstance(key[0], (list, tuple)):
            key = key[0]
        for i in range(b):
            x = encoder_out[i, :encoder_out_lens[i], :]
            am_scores = decoder_out[i, :pre_token_length[i], :]
            x = encoder_out[i, : encoder_out_lens[i], :]
            am_scores = decoder_out[i, : pre_token_length[i], :]
            if self.beam_search is not None:
                nbest_hyps = self.beam_search(
                    x=x, am_scores=am_scores, maxlenratio=kwargs.get("maxlenratio", 0.0),
                    minlenratio=kwargs.get("minlenratio", 0.0)
                    x=x,
                    am_scores=am_scores,
                    maxlenratio=kwargs.get("maxlenratio", 0.0),
                    minlenratio=kwargs.get("minlenratio", 0.0),
                )
                

                nbest_hyps = nbest_hyps[: self.nbest]
            else:
                

                yseq = am_scores.argmax(dim=-1)
                score = am_scores.max(dim=-1)[0]
                score = torch.sum(score, dim=-1)
                # pad with mask tokens to ensure compatibility with sos/eos tokens
                yseq = torch.tensor(
                    [self.sos] + yseq.tolist() + [self.eos], device=yseq.device
                )
                yseq = torch.tensor([self.sos] + yseq.tolist() + [self.eos], device=yseq.device)
                nbest_hyps = [Hypothesis(yseq=yseq, score=score)]
            for nbest_idx, hyp in enumerate(nbest_hyps):
                

                # 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))
                
                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)
                

                result_i = token


                results.extend(result_i)
        

        return results
    
    def inference(self,
                 data_in,
                 data_lengths=None,
                 key: list = None,
                 tokenizer=None,
                 frontend=None,
                 cache: dict={},
                 **kwargs,
                 ):

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

        # 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
        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)
            

        if len(cache) == 0:
            self.init_cache(cache, **kwargs)
        
        

        meta_data = {}
        chunk_size = kwargs.get("chunk_size", [0, 10, 5])
        chunk_stride_samples = int(chunk_size[1] * 960)  # 600ms
        

        time1 = time.perf_counter()
        cfg = {"is_final": kwargs.get("is_final", False)}
        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,
                                                        cache=cfg,
                                                        )
        _is_final = cfg["is_final"] # if data_in is a file or url, set is_final=True
        
        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,
            cache=cfg,
        )
        _is_final = cfg["is_final"]  # if data_in is a file or url, set is_final=True

        time2 = time.perf_counter()
        meta_data["load_data"] = f"{time2 - time1:0.3f}"
        assert len(audio_sample_list) == 1, "batch_size must be set 1"
        

        audio_sample = torch.cat((cache["prev_samples"], audio_sample_list[0]))
        

        n = int(len(audio_sample) // chunk_stride_samples + int(_is_final))
        m = int(len(audio_sample) % chunk_stride_samples * (1-int(_is_final)))
        m = int(len(audio_sample) % chunk_stride_samples * (1 - int(_is_final)))
        tokens = []
        for i in range(n):
            kwargs["is_final"] = _is_final and i == n -1
            audio_sample_i = audio_sample[i*chunk_stride_samples:(i+1)*chunk_stride_samples]

            # extract fbank feats
            speech, speech_lengths = extract_fbank([audio_sample_i], data_type=kwargs.get("data_type", "sound"),
                                                   frontend=frontend, cache=cache["frontend"], is_final=kwargs["is_final"])
            kwargs["is_final"] = _is_final and i == n - 1
            audio_sample_i = audio_sample[i * chunk_stride_samples : (i + 1) * chunk_stride_samples]
            if kwargs["is_final"] and len(audio_sample_i) < 960:
                cache["encoder"]["tail_chunk"] = True
                speech = cache["encoder"]["feats"]
                speech_lengths = torch.tensor([speech.shape[1]], dtype=torch.int64).to(
                    speech.device
                )
            else:
                # extract fbank feats
                speech, speech_lengths = extract_fbank(
                    [audio_sample_i],
                    data_type=kwargs.get("data_type", "sound"),
                    frontend=frontend,
                    cache=cache["frontend"],
                    is_final=kwargs["is_final"],
                )
            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
            
            tokens_i = self.generate_chunk(speech, speech_lengths, key=key, tokenizer=tokenizer, cache=cache, frontend=frontend, **kwargs)
            meta_data["batch_data_time"] = (
                speech_lengths.sum().item() * frontend.frame_shift * frontend.lfr_n / 1000
            )

            tokens_i = self.generate_chunk(
                speech,
                speech_lengths,
                key=key,
                tokenizer=tokenizer,
                cache=cache,
                frontend=frontend,
                **kwargs,
            )
            tokens.extend(tokens_i)
            

        text_postprocessed, _ = postprocess_utils.sentence_postprocess(tokens)
        

        result_i = {"key": key[0], "text": text_postprocessed}
        result = [result_i]
        
        

        cache["prev_samples"] = audio_sample[:-m]
        if _is_final:
            self.init_cache(cache, **kwargs)
        

        if kwargs.get("output_dir"):
            if not hasattr(self, "writer"):
                self.writer = DatadirWriter(kwargs.get("output_dir"))
@@ -561,134 +655,8 @@

        return result, meta_data

    def export(
        self,
        max_seq_len=512,
        **kwargs,
    ):
        self.device = kwargs.get("device")
        is_onnx = kwargs.get("type", "onnx") == "onnx"
        encoder_class = tables.encoder_classes.get(kwargs["encoder"] + "Export")
        self.encoder = encoder_class(self.encoder, onnx=is_onnx)
    
        predictor_class = tables.predictor_classes.get(kwargs["predictor"] + "Export")
        self.predictor = predictor_class(self.predictor, onnx=is_onnx)
        
        if kwargs["decoder"] == "ParaformerSANMDecoder":
            kwargs["decoder"] = "ParaformerSANMDecoderOnline"
        decoder_class = tables.decoder_classes.get(kwargs["decoder"] + "Export")
        self.decoder = decoder_class(self.decoder, onnx=is_onnx)
    
        from funasr.utils.torch_function import MakePadMask
        from funasr.utils.torch_function import sequence_mask
    
        if is_onnx:
            self.make_pad_mask = MakePadMask(max_seq_len, flip=False)
        else:
            self.make_pad_mask = sequence_mask(max_seq_len, flip=False)
    
        self.forward = self._export_forward
    def export(self, **kwargs):
        from .export_meta import export_rebuild_model

        import copy
        import types
        encoder_model = copy.copy(self)
        decoder_model = copy.copy(self)

        # encoder
        encoder_model.forward = types.MethodType(ParaformerStreaming._export_encoder_forward, encoder_model)
        encoder_model.export_dummy_inputs = types.MethodType(ParaformerStreaming.export_encoder_dummy_inputs, encoder_model)
        encoder_model.export_input_names = types.MethodType(ParaformerStreaming.export_encoder_input_names, encoder_model)
        encoder_model.export_output_names = types.MethodType(ParaformerStreaming.export_encoder_output_names, encoder_model)
        encoder_model.export_dynamic_axes = types.MethodType(ParaformerStreaming.export_encoder_dynamic_axes, encoder_model)
        encoder_model.export_name = types.MethodType(ParaformerStreaming.export_encoder_name, encoder_model)
        
        # decoder
        decoder_model.forward = types.MethodType(ParaformerStreaming._export_decoder_forward, decoder_model)
        decoder_model.export_dummy_inputs = types.MethodType(ParaformerStreaming.export_decoder_dummy_inputs, decoder_model)
        decoder_model.export_input_names = types.MethodType(ParaformerStreaming.export_decoder_input_names, decoder_model)
        decoder_model.export_output_names = types.MethodType(ParaformerStreaming.export_decoder_output_names, decoder_model)
        decoder_model.export_dynamic_axes = types.MethodType(ParaformerStreaming.export_decoder_dynamic_axes, decoder_model)
        decoder_model.export_name = types.MethodType(ParaformerStreaming.export_decoder_name, decoder_model)
    
        return encoder_model, decoder_model

    def export_encoder_forward(
        self,
        speech: torch.Tensor,
        speech_lengths: torch.Tensor,
    ):
        # a. To device
        batch = {"speech": speech, "speech_lengths": speech_lengths, "online": True}
        # batch = to_device(batch, device=self.device)
    
        enc, enc_len = self.encoder(**batch)
        mask = self.make_pad_mask(enc_len)[:, None, :]
        alphas, _ = self.predictor.forward_cnn(enc, mask)
    
        return enc, enc_len, alphas

    def export_encoder_dummy_inputs(self):
        speech = torch.randn(2, 30, 560)
        speech_lengths = torch.tensor([6, 30], dtype=torch.int32)
        return (speech, speech_lengths)

    def export_encoder_input_names(self):
        return ['speech', 'speech_lengths']

    def export_encoder_output_names(self):
        return ['enc', 'enc_len', 'alphas']

    def export_encoder_dynamic_axes(self):
        return {
            'speech': {
                0: 'batch_size',
                1: 'feats_length'
            },
            'speech_lengths': {
                0: 'batch_size',
            },
            'enc': {
                0: 'batch_size',
                1: 'feats_length'
            },
            'enc_len': {
                0: 'batch_size',
            },
            'alphas': {
                0: 'batch_size',
                1: 'feats_length'
            },
        }
    
    def export_encoder_name(self):
        return "model.onnx"
    
    def export_decoder_forward(
        self,
        enc: torch.Tensor,
        enc_len: torch.Tensor,
        acoustic_embeds: torch.Tensor,
        acoustic_embeds_len: torch.Tensor,
        *args,
    ):
        decoder_out, out_caches = self.decoder(enc, enc_len, acoustic_embeds, acoustic_embeds_len, *args)
        sample_ids = decoder_out.argmax(dim=-1)
    
        return decoder_out, sample_ids, out_caches

    def export_decoder_dummy_inputs(self):
        dummy_inputs = self.decoder.get_dummy_inputs(enc_size=self.encoder._output_size)
        return dummy_inputs

    def export_decoder_input_names(self):
    
        return self.decoder.get_input_names()

    def export_decoder_output_names(self):
    
        return self.decoder.get_output_names()

    def export_decoder_dynamic_axes(self):
        return self.decoder.get_dynamic_axes()
    def export_decoder_name(self):
        return "decoder.onnx"
        models = export_rebuild_model(model=self, **kwargs)
        return models