nanospeech-dataset.py

import json
import os

import datasets

_DESCRIPTION = """
NanoSpeech dataset containing speech samples with transcriptions.
"""

_HOMEPAGE = "https://huggingface.co/datasets/manojkmk/nanospeech-dataset"

class NanoSpeechDatasetConfig(datasets.BuilderConfig):
    """BuilderConfig for NanoSpeech dataset."""

    def __init__(self, **kwargs):
        """BuilderConfig for NanoSpeech dataset.
        
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(NanoSpeechDatasetConfig, self).__init__(**kwargs)


class NanoSpeechDataset(datasets.GeneratorBasedBuilder):
    """NanoSpeech dataset: Collection of audio samples with transcriptions."""

    VERSION = datasets.Version("1.0.0")
    
    BUILDER_CONFIGS = [
        NanoSpeechDatasetConfig(
            name="default",
            version=VERSION,
            description="Default configuration for NanoSpeech dataset",
        ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "audio": datasets.Audio(sampling_rate=24000),
                    "text": datasets.Value("string"),
                    "duration": datasets.Value("float"),
                }
            ),
            supervised_keys=("audio", "text"),
            homepage=_HOMEPAGE,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        # Download files from your dataset repo
        urls = {
            "train": [
                f"https://huggingface.co/datasets/manojkmk/nanospeech-dataset/resolve/main/shard-{i:06d}.tar"
                for i in range(5)  # You have 5 shards as per your verification
            ]
        }
        data_dirs = dl_manager.download_and_extract(urls)
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"filepaths": data_dirs["train"]},
            )
        ]

    def _generate_examples(self, filepaths):
        """Yields examples."""
        id_ = 0
        for filepath in filepaths:
            # The tar file is already downloaded and extracted by the dl_manager
            # Iterate over directories created from extracted tars
            for root, _, files in os.walk(filepath):
                # Group files by their stem (filename without extension)
                file_groups = {}
                for file in files:
                    stem = os.path.splitext(file)[0]
                    ext = os.path.splitext(file)[1]
                    if stem not in file_groups:
                        file_groups[stem] = {}
                    file_groups[stem][ext] = os.path.join(root, file)
                
                # Process each group (mp3 and json file pair)
                for stem, files_dict in file_groups.items():
                    if ".mp3" in files_dict and ".json" in files_dict:
                        audio_path = files_dict[".mp3"]
                        json_path = files_dict[".json"]
                        
                        # Load metadata from json
                        with open(json_path, "r", encoding="utf-8") as f:
                            metadata = json.load(f)
                        
                        yield id_, {
                            "audio": audio_path,
                            "text": metadata["text"],
                            "duration": metadata["duration"],
                        }
                        id_ += 1