test

语帆

2024-02-28 39de3adfbc12bc491f6da9eb9ffdc5122a3f623d

 funasr/models/lcbnet/model.py

@@ -21,6 +21,7 @@
from funasr.utils import postprocess_utils
from funasr.utils.datadir_writer import DatadirWriter
from funasr.register import tables

import pdb
@tables.register("model_classes", "LCBNet")
class LCBNet(nn.Module):
@@ -89,8 +90,9 @@
        text_encoder = text_encoder_class(input_size=vocab_size, **text_encoder_conf)
        fusion_encoder_class = tables.encoder_classes.get(fusion_encoder)
        fusion_encoder = fusion_encoder_class(**fusion_encoder_conf)
        bias_predictor_class = tables.encoder_classes.get_class(bias_predictor)
        bias_predictor = bias_predictor_class(bias_predictor_conf)
        bias_predictor_class = tables.encoder_classes.get(bias_predictor)
        bias_predictor = bias_predictor_class(**bias_predictor_conf)


        if decoder is not None:
            decoder_class = tables.decoder_classes.get(decoder)
@@ -272,15 +274,12 @@
                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)
@@ -297,7 +296,6 @@
        
        if intermediate_outs is not None:
            return (encoder_out, intermediate_outs), encoder_out_lens
        
        return encoder_out, encoder_out_lens
    
    def _calc_att_loss(
@@ -413,8 +411,7 @@
            logging.info("enable beam_search")
            self.init_beam_search(**kwargs)
            self.nbest = kwargs.get("nbest", 1)
        pdb.set_trace()
        

        meta_data = {}
        if isinstance(data_in, torch.Tensor) and kwargs.get("data_type", "sound") == "fbank":  # fbank
            speech, speech_lengths = data_in, data_lengths
@@ -425,16 +422,19 @@
        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),
            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}"
            audio_sample_list = sample_list[0]
            ocr_sample_list = sample_list[1]
            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
            frame_shift = 10 
            meta_data["batch_data_time"] = speech_lengths.sum().item() * frame_shift / 1000

        speech = speech.to(device=kwargs["device"])
        speech_lengths = speech_lengths.to(device=kwargs["device"])
@@ -443,6 +443,13 @@
        if isinstance(encoder_out, tuple):
            encoder_out = encoder_out[0]
        
        ocr_list_new = [[x + 1 if x != 0 else x for x in sublist] for sublist in ocr_sample_list]
        ocr = torch.tensor(ocr_list_new)
        ocr_lengths = ocr.new_full([1], dtype=torch.long, fill_value=ocr.size(1))
        ocr, ocr_lens, _ = self.text_encoder(ocr, ocr_lengths)
        fusion_out, _, _, _ = self.fusion_encoder(encoder_out,None, ocr, None)
        encoder_out = encoder_out + fusion_out
        pdb.set_trace()
        # 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)
@@ -450,7 +457,7 @@
        
        nbest_hyps = nbest_hyps[: self.nbest]


        pdb.set_trace()
        results = []
        b, n, d = encoder_out.size()
        for i in range(b):
@@ -472,10 +479,13 @@
                # 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))
                
                pdb.set_trace()
                # Change integer-ids to tokens
                token = tokenizer.ids2tokens(token_int)
                pdb.set_trace()
                text = tokenizer.tokens2text(token)
                
                pdb.set_trace()

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