| import os |
| import csv |
| import json |
| from datasets import DatasetInfo, BuilderConfig, GeneratorBasedBuilder, Split, SplitGenerator, Features, Value, Audio |
| from datasets.utils.py_utils import size_str |
| from tqdm import tqdm |
|
|
| _BASE_URL = "https://huggingface.co/datasets/Mohammadawad1/my-dataset-test-v2/resolve/main/" |
|
|
| _AUDIO_URL = _BASE_URL + "audio/{lang}/{split}/{split}_audio.tar" |
| _TRANSCRIPT_URL = _BASE_URL + "transcript/{lang}/{split}_metadata.csv" |
| _N_SHARDS_URL = _BASE_URL + "n_shards.json" |
|
|
| class MyDatasetConfig(BuilderConfig): |
| def __init__(self, name, version, language=None, **kwargs): |
| self.language = language |
| super().__init__(name=name, version=version, **kwargs) |
|
|
| class MyDataset(GeneratorBasedBuilder): |
|
|
| BUILDER_CONFIGS = [ |
| MyDatasetConfig(name="ar", version="1.0.0", language="Arabic"), |
| ] |
|
|
| def _info(self): |
| features = Features({ |
| "audio": Audio(sampling_rate=16000), |
| "sentence": Value("string"), |
| "path": Value("string"), |
| |
| }) |
| return DatasetInfo( |
| description="My Arabic Speech Dataset", |
| features=features, |
| supervised_keys=None, |
| ) |
|
|
| def _split_generators(self, dl_manager): |
| lang = self.config.language |
| n_shards_path = dl_manager.download_and_extract(_N_SHARDS_URL) |
| with open(n_shards_path, encoding="utf-8") as f: |
| n_shards = json.load(f) |
|
|
| audio_urls = { |
| split: [ |
| _AUDIO_URL.format(lang=lang, split=split, shard_idx=i) |
| for i in range(n_shards[lang][split]) |
| ] |
| for split in ["train", "validated", "test"] |
| } |
|
|
| archive_paths = dl_manager.download(audio_urls) |
| local_extracted_archive_paths = dl_manager.extract(archive_paths) |
|
|
| meta_urls = { |
| split: _TRANSCRIPT_URL.format(lang=lang, split=split) |
| for split in ["train", "validated", "test"] |
| } |
| meta_paths = dl_manager.download_and_extract(meta_urls) |
|
|
| return [ |
| SplitGenerator( |
| name=Split.TRAIN, |
| gen_kwargs={ |
| "archives": [dl_manager.iter_archive(path) for path in archive_paths["train"]], |
| "local_extracted_archive_paths": local_extracted_archive_paths["train"], |
| "meta_path": meta_paths["train"], |
| }, |
| ), |
| SplitGenerator( |
| name=Split.VALIDATION, |
| gen_kwargs={ |
| "archives": [dl_manager.iter_archive(path) for path in archive_paths["validated"]], |
| "local_extracted_archive_paths": local_extracted_archive_paths["validated"], |
| "meta_path": meta_paths["validated"], |
| }, |
| ), |
| SplitGenerator( |
| name=Split.TEST, |
| gen_kwargs={ |
| "archives": [dl_manager.iter_archive(path) for path in archive_paths["test"]], |
| "local_extracted_archive_paths": local_extracted_archive_paths["test"], |
| "meta_path": meta_paths["test"], |
| }, |
| ), |
| ] |
|
|
| def _generate_examples(self, archives, local_extracted_archive_paths, meta_path): |
| metadata = {} |
| with open(meta_path, encoding="utf-8-sig") as f: |
| reader = csv.DictReader(f) |
| for row in tqdm(reader, desc="Reading metadata"): |
| filename = row["path"] |
| if not filename.endswith(".wav"): |
| filename += ".wav" |
| metadata[filename] = row |
|
|
| for i, archive in enumerate(archives): |
| for path, file in archive: |
| _, filename = os.path.split(path) |
| if filename in metadata: |
| data = dict(metadata[filename]) |
| full_path = os.path.join(local_extracted_archive_paths[i], path) if local_extracted_archive_paths else path |
| data["audio"] = {"path": full_path, "bytes": file.read()} |
| data["path"] = full_path |
| yield full_path, data |
|
|
