testing.py

import os
import datasets
import pandas as pd 

_DESCRIPTION = """\
This dataset is collected from Commonvoice for testing purposes.
"""

_CITATION = "Some citation"

_data_dir = "data"

class ASRTesting(datasets.GeneratorBasedBuilder):
   
    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "path": datasets.Value("string"),
                    "audio": datasets.Audio(sampling_rate=16_000),
                    "transcript": datasets.Value("string"),
                }
            ),
            supervised_keys=None,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        download_dir = dl_manager.download_and_extract(
            {
                "files": os.path.join(_data_dir, "files.zip"),
                "metadata": os.path.join(_data_dir, "metadata.zip")
            }
        )

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "split": datasets.Split.TRAIN,
                    "data_dir": os.path.join(download_dir["files"], "files"),
                    "metapath": os.path.join(download_dir["metadata"], "metadata", "dataset.tsv"),
                    
                },
            )
        ]

    def _generate_examples(self, data_dir, metapath, split):
        metadata = pd.read_csv(metapath, sep="\t")
        for key, row in metadata.iterrows():
            audio_path = os.path.join(data_dir, row["audio"])
            yield key, {
                "audio": audio_path,
                "transcript": row["transcript"],
                "path": audio_path,
            }