yodas.py

import os
from collections import OrderedDict
from pathlib import Path
import datasets
import os
from .meta import lang2shard_cnt


class YodasConfig(datasets.BuilderConfig):
    """BuilderConfig for Yodas."""

    def __init__(self, lang, version, **kwargs):
        self.language = lang
        self.base_data_path = f"data/{lang}"

        description = (
            f"Youtube speech to text dataset in {self.language}."
        )
        super(YodasConfig, self).__init__(
            name=lang,
            version=datasets.Version(version),
            description=description,
            **kwargs,
        )


DEFAULT_CONFIG_NAME = "all"
LANGS = list(lang2shard_cnt.keys())
VERSION = "1.0.0"

class Yodas(datasets.GeneratorBasedBuilder):
    """Yodas dataset."""

    BUILDER_CONFIGS = [
        YodasConfig(lang, version=VERSION) for lang in LANGS
    ]

    VERSION = datasets.Version("1.0.0")

    def _info(self):
        return datasets.DatasetInfo(
            description="Yodas",
            features=datasets.Features(
                OrderedDict(
                    [
                        ("id", datasets.Value("string")),
                        ("utt_id", datasets.Value("string")),
                        ("audio", datasets.Audio(sampling_rate=16_000)),
                        ("text", datasets.Value("string")),
                    ]
                )
            ),
            supervised_keys=None,
            homepage="",  # TODO
            citation="",  # TODO
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        # TODO

        total_cnt = lang2shard_cnt[self.config.name]

        idx_lst = [f"{i:08d}" for i in range(total_cnt)]
        audio_tar_files = dl_manager.download([f"{self.config.base_data_path}/audio/{i:08d}.tar.gz" for i in range(total_cnt)])
        text_files = dl_manager.download([f"{self.config.base_data_path}/text/{i:08d}.txt" for i in range(total_cnt)])
        #duration_files = dl_manager.download([f"{self.config.base_data_path}/duration/{i:08d}.txt" for i in range(total_cnt)])

        if dl_manager.is_streaming:
            audio_archives = [dl_manager.iter_archive(audio_tar_file) for audio_tar_file in audio_tar_files]
            text_archives = [dl_manager.extract(text_file) for text_file in text_files]

        else:
            print("extracting audio ...")
            extracted_audio_archives = dl_manager.extract(audio_tar_files)

            audio_archives = []
            text_archives = []
            for idx, audio_tar_file, extracted_dir, text_file in zip(idx_lst, audio_tar_files, extracted_audio_archives, text_files):
                audio_archives.append(str(extracted_dir)+'/'+idx)
                text_archives.append(text_file)


        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "is_streaming": dl_manager.is_streaming,
                    "audio_archives": audio_archives,
                    'text_archives': text_archives,
                },
            ),
        ]

    def _generate_examples(self, is_streaming, audio_archives, text_archives):
        """Yields examples."""

        id_ = 0

        if is_streaming:
            for tar_file, text_file in zip(audio_archives, text_archives):

                utt2text = {}

                with open(text_file) as f:
                    for id_, row in enumerate(f):
                        row = row.strip().split(maxsplit=1)
                        utt2text[row[0]] = row[1]

                for path, audio_f in tar_file:

                    path = Path(path)
                    utt_id = path.stem

                    if utt_id in utt2text:

                        result = {
                            'id': id_,
                            'utt_id': utt_id,
                            'audio': {"path": None, "bytes": audio_f.read()},
                            'text': utt2text[utt_id]
                        }

                        yield id_, result
                        id_ += 1
        else:
            for extracted_dir, text_file in zip(audio_archives, text_archives):

                utt2text = {}
                extract_root_dir = Path(extracted_dir).parent
                extracted_dir = list(extract_root_dir.glob('./*'))[0]

                with open(text_file) as f:
                    for _, row in enumerate(f):
                        row = row.strip().split(maxsplit=1)
                        utt2text[row[0]] = row[1]

                for audio_file in list(Path(extracted_dir).glob('*')):

                    utt_id = audio_file.stem
                    if utt_id in utt2text:

                        result = {
                            'id': id_,
                            'utt_id': utt_id,
                            'audio': {"path": str(audio_file.absolute()), "bytes": open(audio_file, 'rb').read()},
                            'text': utt2text[utt_id]
                        }

                        yield id_, result
                        id_ += 1