CSTR-Edinburgh/vctk · Non Matching Splits Sizes Error

Problem Description: dataset info mismatch

First, I used huggingface_hub.snapshot_download() to download this repo, and find the dataset file URL https://datashare.is.ed.ac.uk/bitstream/handle/10283/3443/VCTK-Corpus-0.92.zip in vctk.py line 39.
Second, I used wget to download the dataset file into local path CSTR-Edinburgh/vctk/data/VCTK-Corpus-0.92.zip, and modified the vctk.py line 39 to the real path of this zip file.
Third, I used vctk = load_dataset(path=dataset_path, cache_dir=cache_dir, trust_remote_code=True) in another python file to process this dataset, but the error occurred.

Generating train split:   5%|██████████▎                                                                                                                                                                                                      | 4345/88156 [00:53<17:11, 81.22 examples/s]
Traceback (most recent call last):
  File "xxxxxxxxx.py", line 144, in <module>
    vctk = load_dataset(path=dataset_path, cache_dir=cache_dir, trust_remote_code=True)
  File "/opt/conda/envs/py39/lib/python3.9/site-packages/datasets/load.py", line 2614, in load_dataset
    builder_instance.download_and_prepare(
  File "/opt/conda/envs/py39/lib/python3.9/site-packages/datasets/builder.py", line 1027, in download_and_prepare
    self._download_and_prepare(
  File "/opt/conda/envs/py39/lib/python3.9/site-packages/datasets/builder.py", line 1789, in _download_and_prepare
    super()._download_and_prepare(
  File "/opt/conda/envs/py39/lib/python3.9/site-packages/datasets/builder.py", line 1140, in _download_and_prepare
    verify_splits(self.info.splits, split_dict)
  File "/opt/conda/envs/py39/lib/python3.9/site-packages/datasets/utils/info_utils.py", line 101, in verify_splits
    raise NonMatchingSplitsSizesError(str(bad_splits))
datasets.utils.info_utils.NonMatchingSplitsSizesError: [{'expected': SplitInfo(name='train', num_bytes=40103111, num_examples=88156, shard_lengths=None, dataset_name=None), 'recorded': SplitInfo(name='train', num_bytes=2477106, num_examples=4345, shard_lengths=None, dataset_name='vctk')}]

Is there any wrong with my usage or the problem is caused by the zip file?