gigaspeech2_metadata.py

import os
from collections import OrderedDict
from pathlib import Path
import datasets
import glob
import os

from datasets.arrow_dataset import config


class Gigaspeech2Config(datasets.BuilderConfig):
    """BuilderConfig for Yodas."""

    def __init__(self, config_name, version, **kwargs):
        config_split = config_name.split("-")
        self.language = config_split[0]
        print(config_split)
        self.split = config_split[1]
        self.index = config_split[2] if len(config_split)>=3 else None
        self.base_data_path = f"data/{self.language}"

        description = (
            f"Youtube speech to text dataset in {self.language}."
        )
        super(Gigaspeech2Config, self).__init__(
            name=config_name,
            version=datasets.Version(version),
            description=description,
            **kwargs,
        )


DEFAULT_CONFIG_NAME = "all"
LANGS = ['id', 'th', 'vi']
LANGS_SPLITS={
    "id":2,
    "th":2,
    "vi":2
}
VERSION = "1.0.1"
CONFIG_NAME=[]
for lang in LANGS:
    CONFIG_NAME.append(f"{lang}-test")
    CONFIG_NAME.append(f"{lang}-dev")
    CONFIG_NAME.append(f"{lang}-train_raw")
    CONFIG_NAME.append(f"{lang}-train_refine")
    for i in range(LANGS_SPLITS[lang]):
        CONFIG_NAME.append(f"{lang}-train_raw-{i}")
        CONFIG_NAME.append(f"{lang}-train_refine-{i}")

class Gigaspeech2(datasets.GeneratorBasedBuilder):
    """Yodas dataset."""

    BUILDER_CONFIGS = [
        Gigaspeech2Config(config_name, version=VERSION) for config_name in CONFIG_NAME
    ]

    VERSION = datasets.Version("1.0.1")

    def _info(self):
        return datasets.DatasetInfo(
            description="Gigaspeech2",
            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
        lang = self.config.language
        split = self.config.split
        index = self.config.index
        if index is not None:
            assert int(index)<LANGS_SPLITS[lang], f"split {self.config.name} not found"

        text_file = dl_manager.download([f"{self.config.base_data_path}/{split}.tsv"])
        if split == "test":
            audio_tar_files = dl_manager.download([f"{self.config.base_data_path}/{split}.tar.gz"])
        elif split == "dev":
            audio_tar_files = dl_manager.download([f"{self.config.base_data_path}/{split}.tar.gz"])
        elif index is not None:
            audio_tar_files = dl_manager.download([f"{self.config.base_data_path}/train/{index}.tar.gz"])
        else:
            audio_tar_files = dl_manager.download([f"{self.config.base_data_path}/train/{i}.tar.gz" for i in range(LANGS_SPLITS[lang])])

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

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

            audio_archives = []
            for extracted_dir in extracted_audio_archives:
                audio_archives.extend(glob.glob("{}/**/*.wav".format(str(extracted_dir)), recursive=True))

        if split=="test":
            dataset_split = datasets.Split.TEST
        elif split=="dev":
            dataset_split = datasets.Split.VALIDATION
        else:
            dataset_split = datasets.Split.TRAIN

        return [
            datasets.SplitGenerator(
                name=dataset_split,
                gen_kwargs={
                    "is_streaming": dl_manager.is_streaming,
                    "audio_archives": audio_archives,
                    'tsv_file': text_file[0],
                },
            ),
        ]

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

        id_ = 0

        with open(tsv_file, 'r') as f:
            lines = f.read().splitlines()
            lines = [line.split("\t") for line in lines]
            id2label={line[0]:line[1] for line in lines}

        if is_streaming:
            for tar_file in audio_archives:
                for path, audio_f in tar_file:
                    path = Path(path)
                    utt_id = path.stem
                    if utt_id not in id2label:
                        continue
                    result = {
                        'id': id_,
                        'utt_id': utt_id,
                        'audio': {"path": None, "bytes": audio_f.read()},
                        'text': id2label[utt_id],
                    }

                    yield id_, result
                    id_ += 1
        else:
            for audio_file in audio_archives:
                utt_id = Path(audio_file).stem
                if utt_id not in id2label:
                    continue
                result = {
                    'id': id_,
                    'utt_id': utt_id,
                    'audio': {"path": audio_file, "bytes": open(audio_file, 'rb').read()},
                    'text': id2label[utt_id]
                }

                yield id_, result
                id_ += 1