Merge pull request #79 from alibaba-damo-academy/dev_wjm

zhifu gao

2023-02-09 de0ecb446fa429d210397949694d8d9ad6d66112

 funasr/datasets/large_datasets/dataset.py

@@ -13,6 +13,7 @@
from funasr.datasets.large_datasets.datapipes.map import MapperIterDataPipe
from funasr.datasets.large_datasets.utils.filter import filter
from funasr.datasets.large_datasets.utils.padding import padding
from funasr.datasets.large_datasets.utils.clipping import clipping
from funasr.datasets.large_datasets.utils.tokenize import tokenize


@@ -101,6 +102,8 @@
                elif data_type == "text" or data_type == "sound":
                    text_reader = open(data_file, "r")
                    reader_list.append(text_reader)
                elif data_type == "none":
                    continue
                else:
                    raise TypeError("Data type {} is not supported".format(data_type))

@@ -143,7 +146,8 @@
            dict,
            seg_dict,
            conf,
            mode="train"):
            mode="train",
            batch_mode="padding"):
    scp_lists = read_lists(data_list_file)
    shuffle = conf.get('shuffle', True)
    data_names = conf.get("data_names", "speech,text")
@@ -154,9 +158,10 @@
    filter_fn = partial(filter, **filter_conf)
    dataset = FilterIterDataPipe(dataset, fn=filter_fn)

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

    if shuffle:
        buffer_conf = conf.get('shuffle_conf', {})
@@ -180,8 +185,9 @@
                                             batch_size=batch_size,
                                             len_fn=len_fn,
                                             buffer_size=buffer_size,
                                             sort_size=sort_size)
                                             sort_size=sort_size,
                                             batch_mode=batch_mode)

    dataset = MapperIterDataPipe(dataset, fn=padding)
    dataset = MapperIterDataPipe(dataset, fn=padding if batch_mode == "padding" else clipping)

    return dataset