|
""" Kathbath Dataset""" |
|
|
|
import csv |
|
import os |
|
|
|
import datasets |
|
from datasets.utils.py_utils import size_str |
|
|
|
from .languages import LANGUAGES |
|
from .release_stats import STATS |
|
|
|
_CITATION = """\ |
|
@misc{https://doi.org/10.48550/arxiv.2208.11761, |
|
doi = {10.48550/ARXIV.2208.11761}, |
|
url = {https://arxiv.org/abs/2208.11761}, |
|
author = {Javed, Tahir and Bhogale, Kaushal Santosh and Raman, Abhigyan and Kunchukuttan, Anoop and Kumar, Pratyush and Khapra, Mitesh M.}, |
|
title = {IndicSUPERB: A Speech Processing Universal Performance Benchmark for Indian languages}, |
|
publisher = {arXiv}, |
|
year = {2022}, |
|
copyright = {arXiv.org perpetual, non-exclusive license} |
|
} |
|
""" |
|
|
|
_HOMEPAGE = "https://ai4bharat.iitm.ac.in/indic-superb/" |
|
|
|
_LICENSE = "https://creativecommons.org/publicdomain/zero/1.0/" |
|
|
|
_DATA_URL = "https://huggingface.co/datasets/ai4bharat/kathbath/resolve/main/data" |
|
|
|
|
|
class KathbathConfig(datasets.BuilderConfig): |
|
"""BuilderConfig for Kathbath.""" |
|
|
|
def __init__(self, name, version, **kwargs): |
|
self.language = kwargs.pop("language", None) |
|
self.release_date = kwargs.pop("release_date", None) |
|
self.num_clips = kwargs.pop("num_clips", None) |
|
self.num_speakers = kwargs.pop("num_speakers", None) |
|
self.total_hr = kwargs.pop("total_hr", None) |
|
self.size_bytes = kwargs.pop("size_bytes", None) |
|
self.size_human = size_str(self.size_bytes) |
|
description = ( |
|
f"Kathbath speech to text dataset in {self.language} released on {self.release_date}. " |
|
f"The dataset comprises {self.total_hr} hours of transcribed speech data" |
|
) |
|
super(KathbathConfig, self).__init__( |
|
name=name, |
|
version=datasets.Version(version), |
|
description=description, |
|
**kwargs, |
|
) |
|
|
|
|
|
class Kathbath(datasets.GeneratorBasedBuilder): |
|
DEFAULT_CONFIG_NAME = "_all_" |
|
|
|
BUILDER_CONFIGS = [ |
|
KathbathConfig( |
|
name=lang, |
|
version=STATS["version"], |
|
language=LANGUAGES[lang], |
|
release_date=STATS["date"], |
|
|
|
|
|
total_hr=float(lang_stats["totalHrs"]) if lang_stats["totalHrs"] else None, |
|
size_bytes=int(lang_stats["size"]) if lang_stats["size"] else None, |
|
) |
|
for lang, lang_stats in STATS["locales"].items() |
|
] |
|
|
|
def _info(self): |
|
total_languages = len(STATS["locales"]) |
|
total_hours = self.config.total_hr |
|
description = ( |
|
"LibriVox-Indonesia is a speech dataset generated from LibriVox with only languages from Indonesia." |
|
f"The dataset currently consists of {total_hours} hours of speech " |
|
f"in {total_languages} languages, but more voices and languages are always added." |
|
) |
|
features = datasets.Features( |
|
{ |
|
"path": datasets.Value("string"), |
|
"language": datasets.Value("string"), |
|
"speaker": datasets.Value("string"), |
|
"sentence": datasets.Value("string"), |
|
"audio": datasets.features.Audio(sampling_rate=16000) |
|
} |
|
) |
|
|
|
return datasets.DatasetInfo( |
|
description=description, |
|
features=features, |
|
supervised_keys=None, |
|
homepage=_HOMEPAGE, |
|
license=_LICENSE, |
|
citation=_CITATION, |
|
version=self.config.version, |
|
) |
|
|
|
def _split_generators(self, dl_manager): |
|
"""Returns SplitGenerators.""" |
|
dl_manager.download_config.ignore_url_params = True |
|
audio_path = {} |
|
local_extracted_archive = {} |
|
metadata_path = {} |
|
split_type = {"train": datasets.Split.TRAIN, "valid": datasets.Split.VALIDATION, "test_unknown": datasets.Split.TEST, "test_known": datasets.Split.TEST} |
|
for split in split_type: |
|
|
|
if split == 'train': |
|
audio_paths = [ |
|
f"{_DATA_URL}/audio_{split}.tar.partaa", |
|
f"{_DATA_URL}/audio_{split}.tar.partab", |
|
f"{_DATA_URL}/audio_{split}.tar.partac", |
|
] |
|
else: |
|
audio_paths = [f"{_DATA_URL}/audio_{split}.tar"] |
|
audio_path[split] = dl_manager.download(audio_paths) |
|
local_extracted_archive[split] = dl_manager.extract(audio_path[split]) if not dl_manager.is_streaming else None |
|
metadata_path[split] = dl_manager.download(f"{_DATA_URL}/metadata_{split}.tsv") |
|
path_to_clips = "kb_data_clean_m4a" |
|
|
|
return [ |
|
datasets.SplitGenerator( |
|
name=split_type[split], |
|
gen_kwargs={ |
|
"local_extracted_archive": local_extracted_archive[split], |
|
"audio_files": dl_manager.iter_archive(audio_path[split]), |
|
"metadata_path": metadata_path[split], |
|
"path_to_clips": path_to_clips, |
|
}, |
|
) for split in split_type |
|
] |
|
|
|
def _generate_examples( |
|
self, |
|
local_extracted_archive, |
|
audio_files, |
|
metadata_path, |
|
path_to_clips, |
|
): |
|
"""Yields examples.""" |
|
data_fields = list(self._info().features.keys()) |
|
metadata = {} |
|
with open(metadata_path, "r", encoding="utf-8") as f: |
|
reader = csv.DictReader(f, delimiter="\t") |
|
for row in reader: |
|
if self.config.name == "_all_" or self.config.name == row["language"]: |
|
row["path"] = os.path.join(path_to_clips, row["path"]) |
|
|
|
for field in data_fields: |
|
if field not in row: |
|
row[field] = "" |
|
metadata[row["path"]] = row |
|
id_ = 0 |
|
for path, f in audio_files: |
|
if path in metadata: |
|
result = dict(metadata[path]) |
|
|
|
path = os.path.join(local_extracted_archive, path) if local_extracted_archive else path |
|
result["audio"] = {"path": path, "bytes": f.read()} |
|
result["path"] = path |
|
yield id_, result |
|
id_ += 1 |
|
|
