mydataset.py

from datasets import DatasetInfo, GeneratorBasedBuilder, Split, SplitGenerator, Value, Audio, Features
import os
import csv

class MyDataset(GeneratorBasedBuilder):
    def _info(self):
        return DatasetInfo(
            features=Features({
                "audio": Audio(sampling_rate=16_000),  # Audio feature
                "text": Value("string")               # Transcription
            }),
        )

    def _split_generators(self, dl_manager):
        audio_dir = os.path.join(self.config.data_dir, "audio")
        metadata_path = os.path.join(self.config.data_dir, "metadata.csv")

        return [
            SplitGenerator(
                name=Split.TRAIN,
                gen_kwargs={
                    "metadata_path": metadata_path,
                    "audio_dir": audio_dir
                },
            )
        ]

    def _generate_examples(self, metadata_path, audio_dir):
        with open(metadata_path, encoding="utf-8") as f:
            reader = csv.DictReader(f)
            for idx, row in enumerate(reader):
                yield idx, {
                    "audio": os.path.join(audio_dir, row["file"]),
                    "text": row["text"]
                }