Datasets:

---

mythicinfinity

/

libriheavy

like 11

Follow

Mythic Infinity 5

There are about 78 rows in the large dataset subset currently experiencing a decode error due to issues with 4 malformed/corrupted librivox source files.

An immediate fix is not planned, and it is recommended to use a workaround if you are hitting this error. Something like this should suffice for many use cases:

class ZeroOnErrorAudio(datasets.Audio):
    def decode_example(
        self, value: dict, token_per_repo_id: Optional[Dict[str, Union[str, bool, None]]] = None
    ) -> dict:
        try:
            return super().decode_example(value, token_per_repo_id=token_per_repo_id)
        except Exception as exc:
            traceback.print_exc()
            print('audio load err', exc)
            print(value)

            # return silence so processing likely still works
            return {"path": value['path'], "array": np.zeros((self.sampling_rate, ), dtype=np.float32), "sampling_rate": self.sampling_rate}

Then

dset = load_dataset('mythicinfinity/libriheavy', 'large')

# You may want to set sampling_rate, etc here
dset = dset.cast_column('audio', ZeroOnErrorAudio())

Exact workarounds will differ by use-case. In my own scripts I catch the error in an outer dataset iteration wrapper and simply return the next sample instead.