dusha_emotion_audio.py

import csv

import datasets

_DESCRIPTION = """\
Dusha is a bi-modal corpus suitable for speech emotion recognition (SER) tasks. 
The dataset consists of audio recordings with Russian speech and their emotional labels. 
The corpus contains approximately 350 hours of data. Four basic emotions that usually appear in a dialog with
 a virtual assistant were selected: Happiness (Positive), Sadness, Anger and Neutral emotion.
"""

_HOMEPAGE = "https://github.com/salute-developers/golos/tree/master/dusha#dusha-dataset"

_DATA_URL_TRAIN = "https://huggingface.co/datasets/KELONMYOSA/dusha_emotion_audio/resolve/main/data/train.tar.gz"
_DATA_URL_TEST = "https://huggingface.co/datasets/KELONMYOSA/dusha_emotion_audio/resolve/main/data/test.tar.gz"
_METADATA_URL_TRAIN = "https://huggingface.co/datasets/KELONMYOSA/dusha_emotion_audio/resolve/main/data/train.csv"
_METADATA_URL_TEST = "https://huggingface.co/datasets/KELONMYOSA/dusha_emotion_audio/resolve/main/data/test.csv"


class Dusha(datasets.GeneratorBasedBuilder):
    DEFAULT_WRITER_BATCH_SIZE = 256

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "file": datasets.Value("string"),
                    "audio": datasets.Audio(sampling_rate=16_000),
                    "label": datasets.ClassLabel(num_classes=5, names=['neutral', 'positive', 'sad', 'angry', 'other']),
                }
            ),
            supervised_keys=None,
            homepage=_HOMEPAGE,
        )

    def _split_generators(self, dl_manager):
        metadata_train = dl_manager.download(_METADATA_URL_TRAIN)
        metadata_test = dl_manager.download(_METADATA_URL_TEST)
        archive_train = dl_manager.download(_DATA_URL_TRAIN)
        archive_test = dl_manager.download(_DATA_URL_TEST)
        return [
        datasets.SplitGenerator(
            name=datasets.Split.TRAIN,
            gen_kwargs={
                "audio_files": dl_manager.iter_archive(archive_train),
                "metadata": metadata_train},
        ),
        datasets.SplitGenerator(
            name=datasets.Split.TEST,
            gen_kwargs={
                "audio_files": dl_manager.iter_archive(archive_test),
                "metadata": metadata_test},
        )
        ]

    def _generate_examples(self, audio_files, metadata):
        examples = dict()

        with open(metadata, encoding="utf-8") as f:
            csv_reader = csv.reader(f, delimiter=",")
            next(csv_reader)
            for row in csv_reader:
                audio_path, label = row
                examples[audio_path] = {
                    "file": audio_path,
                    "label": label,
                }

        key = 0
        for path, f in audio_files:
            if path in examples:
                audio = {"path": path, "bytes": f.read()}
                yield key, {**examples[path], "audio": audio}
                key += 1