Dev gzf exp (#1654)

zhifu gao

2024-04-24 861147c7308b91068ffa02724fdf74ee623a909e

 funasr/models/paraformer_streaming/model.py

@@ -54,14 +54,19 @@
        # 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,
@@ -103,9 +108,9 @@

        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
                
@@ -125,8 +130,11 @@
        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
@@ -145,7 +153,11 @@
        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:
@@ -164,7 +176,9 @@
                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]
        
@@ -177,40 +191,48 @@
        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,
        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)
        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,
@@ -228,22 +250,32 @@
                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
        
@@ -273,9 +305,19 @@
        
        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):
    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)
        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]
@@ -293,51 +335,64 @@
            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,
        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)
        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,
@@ -363,7 +418,9 @@

        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
        )
@@ -371,10 +428,10 @@
        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
@@ -387,22 +444,33 @@

        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,
        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}
            "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,
    def generate_chunk(
        self,
                       speech,
                       speech_lengths=None,
                       key: list = None,
@@ -415,22 +483,27 @@
        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]

@@ -443,8 +516,10 @@
            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]
@@ -454,9 +529,7 @@
                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):
                
@@ -468,8 +541,11 @@
                    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)
@@ -477,12 +553,12 @@
                
                result_i = token


                results.extend(result_i)
        
        return results
    
    def inference(self,
    def inference(
        self,
                 data_in,
                 data_lengths=None,
                 key: list = None,
@@ -494,7 +570,9 @@

        # 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)
@@ -503,14 +581,14 @@
        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,
        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"),
@@ -534,23 +612,39 @@
            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)
                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"])
                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
            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_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:
@@ -567,5 +661,6 @@
    
    def export(self, **kwargs):
        from .export_meta import export_rebuild_model

        models = export_rebuild_model(model=self, **kwargs)
        return models