audio_textgrid_dataset.py

import os
import datasets

class AudioTextgridDataset(datasets.GeneratorBasedBuilder):
    def _info(self):
        return datasets.DatasetInfo(
            features=datasets.Features({
                "audio": datasets.Audio(),
                "textgrid": datasets.Value("string"),
            })
        )

    def _split_generators(self, dl_manager):
        data_dir = dl_manager.download_and_extract("path/to/dataset_folder")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"data_dir": data_dir},
            ),
        ]

    def _generate_examples(self, data_dir):
        metadata_path = os.path.join(data_dir, "metadata.csv")
        with open(metadata_path, "r") as f:
            for id, line in enumerate(f):
                audio_file, textgrid_file = line.strip().split(",")
                yield id, {
                    "audio": os.path.join(data_dir, "data", audio_file),
                    "textgrid": open(os.path.join(data_dir, "data", textgrid_file)).read(),
                }