mypin-voice-dataset.py

import datasets
import os
import random
import json

class MyDataset(datasets.GeneratorBasedBuilder):
    def _info(self):
        return datasets.DatasetInfo(
            description="My dataset with text and audio.",
            features=datasets.Features({
                "sentence": datasets.Value("string"),
                "audio": datasets.Audio(sampling_rate=16_000),  # Defines the audio feature
            }),
            homepage="https://huggingface.co/datasets/my-dataset",
            license="MIT",
        )

    def _split_generators(self, dl_manager):
        # Download and extract the dataset
        # data_dir = dl_manager.download_and_extract("https://github.com/atulksingh011/test-voice-data/raw/main/audios.tar.gz")
        # metadata = dl_manager.download("https://github.com/atulksingh011/test-voice-data/raw/main/metadata.jsonl")
        
        data_dir = dl_manager.download_and_extract("https://raw.githubusercontent.com/atulksingh011/test-voice-data/refs/heads/record-names/audio.tar.gz")
        metadata = dl_manager.download("https://github.com/atulksingh011/test-voice-data/raw/record-names/metadata.jsonl")
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": metadata, 
                    "audio_dir": data_dir,
                    "split": "train"
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepath": metadata, 
                    "audio_dir": data_dir,
                    "split": "eval"
                },
            )
        ]

    def _generate_examples(self, filepath, audio_dir, split):
        # Read and parse the metadata JSONL file
        with open(filepath, "r", encoding="utf-8") as f:
            data = [json.loads(line) for line in f]

        # Shuffle the data for randomness
        random.shuffle(data)

        # Calculate split index for training and evaluation
        split_index = int(len(data) * 0.8)
        if split == "train":
            examples = data[:split_index]
        else:
            examples = data[split_index:]

        # Yield the examples
        for idx, record in enumerate(examples):
            yield idx, {
                "sentence": record["sentence"],
                "audio": os.path.join(audio_dir, record["file_name"]),  # Correct path to the audio file
            }