Merge branch 'main' into dev_smohan

yhliang

2023-05-11 1d1ef01b4e23630a99a3be7e9d1dce9550a793e9

 funasr/datasets/large_datasets/dataset.py

@@ -101,7 +101,7 @@
                if data_type == "kaldi_ark":
                    ark_reader = ReadHelper('ark:{}'.format(data_file))
                    reader_list.append(ark_reader)
                elif data_type == "text" or data_type == "sound":
                elif data_type == "text" or data_type == "sound" or data_type == 'text_hotword':
                    text_reader = open(data_file, "r")
                    reader_list.append(text_reader)
                elif data_type == "none":
@@ -131,6 +131,13 @@
                        sample_dict["sampling_rate"] = sampling_rate
                        if data_name == "speech":
                            sample_dict["key"] = key
                    elif data_type == "text_hotword":
                        text = item
                        segs = text.strip().split()
                        sample_dict[data_name] = segs[1:]
                        if "key" not in sample_dict:
                            sample_dict["key"] = segs[0]
                        sample_dict['hw_tag'] = 1
                    else:
                        text = item
                        segs = text.strip().split()
@@ -167,14 +174,38 @@
    shuffle = conf.get('shuffle', True)
    data_names = conf.get("data_names", "speech,text")
    data_types = conf.get("data_types", "kaldi_ark,text")
    dataset = AudioDataset(scp_lists, data_names, data_types, frontend_conf=frontend_conf, shuffle=shuffle, mode=mode)

    pre_hwfile = conf.get("pre_hwlist", None)
    pre_prob = conf.get("pre_prob", 0)  # unused yet

    hw_config = {"sample_rate": conf.get("sample_rate", 0.6),
                 "double_rate": conf.get("double_rate", 0.1),
                 "hotword_min_length": conf.get("hotword_min_length", 2),
                 "hotword_max_length": conf.get("hotword_max_length", 8),
                 "pre_prob": conf.get("pre_prob", 0.0)}

    if pre_hwfile is not None:
        pre_hwlist = []
        with open(pre_hwfile, 'r') as fin:
            for line in fin.readlines():
                pre_hwlist.append(line.strip())
    else:
        pre_hwlist = None

    dataset = AudioDataset(scp_lists, 
                           data_names, 
                           data_types, 
                           frontend_conf=frontend_conf, 
                           shuffle=shuffle, 
                           mode=mode, 
                           )

    filter_conf = conf.get('filter_conf', {})
    filter_fn = partial(filter, **filter_conf)
    dataset = FilterIterDataPipe(dataset, fn=filter_fn)

    if "text" in data_names:
        vocab = {'vocab': dict, 'seg_dict': seg_dict, 'punc_dict': punc_dict, 'bpe_tokenizer': bpe_tokenizer}
        vocab = {'vocab': dict, 'seg_dict': seg_dict, 'punc_dict': punc_dict, 'bpe_tokenizer': bpe_tokenizer, 'hw_config': hw_config}
        tokenize_fn = partial(tokenize, **vocab)
        dataset = MapperIterDataPipe(dataset, fn=tokenize_fn)