pianos.py

import os
import random
import datasets
from datasets.tasks import ImageClassification, AudioClassification


_NAMES = [
    "PearlRiver",
    "YoungChang",
    "Steinway-T",
    "Hsinghai",
    "Kawai",
    "Steinway",
    "Kawai-G",
    "Yamaha",
]

_DBNAME = os.path.basename(__file__).split(".")[0]

_HOMEPAGE = f"https://www.modelscope.cn/datasets/ccmusic/{_DBNAME}"

_DOMAIN = f"https://www.modelscope.cn/api/v1/datasets/ccmusic/{_DBNAME}/repo?Revision=master&FilePath=data"

_CITATION = """\
@dataset{zhaorui_liu_2021_5676893,
  author       = {Monan Zhou, Shenyang Xu, Zhaorui Liu, Zhaowen Wang, Feng Yu, Wei Li and Zijin Li},
  title        = {CCMusic: an Open and Diverse Database for Chinese and General Music Information Retrieval Research},
  month        = {mar},
  year         = {2024},
  publisher    = {HuggingFace},
  version      = {1.2},
  url          = {https://huggingface.co/ccmusic-database}
}
"""

_DESCRIPTION = """\
Piano-Sound-Quality is a dataset of piano sound. It consists of 8 kinds of pianos including PearlRiver, YoungChang, Steinway-T, Hsinghai, Kawai, Steinway, Kawai-G, Yamaha(recorded by Shaohua Ji with SONY PCM-D100). Data was annotated by students from the China Conservatory of Music (CCMUSIC) in Beijing and collected by Monan Zhou.
"""

_PITCHES = {
    "009": "A2",
    "010": "A2#/B2b",
    "011": "B2",
    "100": "C1",
    "101": "C1#/D1b",
    "102": "D1",
    "103": "D1#/E1b",
    "104": "E1",
    "105": "F1",
    "106": "F1#/G1b",
    "107": "G1",
    "108": "G1#/A1b",
    "109": "A1",
    "110": "A1#/B1b",
    "111": "B1",
    "200": "C",
    "201": "C#/Db",
    "202": "D",
    "203": "D#/Eb",
    "204": "E",
    "205": "F",
    "206": "F#/Gb",
    "207": "G",
    "208": "G#/Ab",
    "209": "A",
    "210": "A#/Bb",
    "211": "B",
    "300": "c",
    "301": "c#/db",
    "302": "d",
    "303": "d#/eb",
    "304": "e",
    "305": "f",
    "306": "f#/gb",
    "307": "g",
    "308": "g#/ab",
    "309": "a",
    "310": "a#/bb",
    "311": "b",
    "400": "c1",
    "401": "c1#/d1b",
    "402": "d1",
    "403": "d1#/e1b",
    "404": "e1",
    "405": "f1",
    "406": "f1#/g1b",
    "407": "g1",
    "408": "g1#/a1b",
    "409": "a1",
    "410": "a1#/b1b",
    "411": "b1",
    "500": "c2",
    "501": "c2#/d2b",
    "502": "d2",
    "503": "d2#/e2b",
    "504": "e2",
    "505": "f2",
    "506": "f2#/g2b",
    "507": "g2",
    "508": "g2#/a2b",
    "509": "a2",
    "510": "a2#/b2b",
    "511": "b2",
    "600": "c3",
    "601": "c3#/d3b",
    "602": "d3",
    "603": "d3#/e3b",
    "604": "e3",
    "605": "f3",
    "606": "f3#/g3b",
    "607": "g3",
    "608": "g3#/a3b",
    "609": "a3",
    "610": "a3#/b3b",
    "611": "b3",
    "700": "c4",
    "701": "c4#/d4b",
    "702": "d4",
    "703": "d4#/e4b",
    "704": "e4",
    "705": "f4",
    "706": "f4#/g4b",
    "707": "g4",
    "708": "g4#/a4b",
    "709": "a4",
    "710": "a4#/b4b",
    "711": "b4",
    "800": "c5",
}

_URLS = {
    "audio": f"{_DOMAIN}/audio.zip",
    "mel": f"{_DOMAIN}/mel.zip",
    "eval": f"{_DOMAIN}/eval.zip",
}


class pianos_Config(datasets.BuilderConfig):
    def __init__(self, features, supervised_keys, task_templates, **kwargs):
        super(pianos_Config, self).__init__(version=datasets.Version("0.1.2"), **kwargs)
        self.features = features
        self.supervised_keys = supervised_keys
        self.task_templates = task_templates


class pianos(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("0.1.2")
    BUILDER_CONFIGS = [
        pianos_Config(
            name="eval",
            features=datasets.Features(
                {
                    "mel": datasets.Image(),
                    "label": datasets.features.ClassLabel(names=_NAMES),
                    "pitch": datasets.features.ClassLabel(
                        names=list(_PITCHES.values())
                    ),
                }
            ),
            supervised_keys=("mel", "label"),
            task_templates=[
                ImageClassification(
                    task="image-classification",
                    image_column="mel",
                    label_column="label",
                )
            ],
        ),
        pianos_Config(
            name="default",
            features=datasets.Features(
                {
                    "audio": datasets.Audio(sampling_rate=22050),
                    "mel": datasets.Image(),
                    "label": datasets.features.ClassLabel(names=_NAMES),
                    "pitch": datasets.features.ClassLabel(
                        names=list(_PITCHES.values())
                    ),
                }
            ),
            supervised_keys=("audio", "label"),
            task_templates=[
                AudioClassification(
                    task="audio-classification",
                    audio_column="audio",
                    label_column="label",
                )
            ],
        ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=self.config.features,
            homepage=_HOMEPAGE,
            license="mit",
            citation=_CITATION,
            supervised_keys=self.config.supervised_keys,
            task_templates=self.config.task_templates,
        )

    def _split_generators(self, dl_manager):
        dataset = []
        if self.config.name == "eval":
            data_files = dl_manager.download_and_extract(_URLS["eval"])
            for path in dl_manager.iter_files([data_files]):
                fname = os.path.basename(path)
                if fname.endswith(".jpg"):
                    dataset.append(
                        {
                            "mel": path,
                            "label": os.path.basename(os.path.dirname(path)),
                            "pitch": _PITCHES[fname.split("_")[0]],
                        }
                    )
        else:
            subset = {}
            audio_files = dl_manager.download_and_extract(_URLS["audio"])
            for path in dl_manager.iter_files([audio_files]):
                fname = os.path.basename(path)
                if fname.endswith(".wav"):
                    subset[fname.split(".")[0]] = {
                        "audio": path,
                        "label": os.path.basename(os.path.dirname(path)),
                        "pitch": _PITCHES[fname[1:4]],
                    }

            mel_files = dl_manager.download_and_extract(_URLS["mel"])
            for path in dl_manager.iter_files([mel_files]):
                fname = os.path.basename(path)
                if fname.endswith(".jpg"):
                    subset[fname.split(".")[0]]["mel"] = path

            dataset = list(subset.values())

        random.shuffle(dataset)
        count = len(dataset)
        p80 = int(0.8 * count)
        p90 = int(0.9 * count)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN, gen_kwargs={"files": dataset[:p80]}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION, gen_kwargs={"files": dataset[p80:p90]}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST, gen_kwargs={"files": dataset[p90:]}
            ),
        ]

    def _generate_examples(self, files):
        if self.config.name == "eval":
            for i, path in enumerate(files):
                yield i, {
                    "mel": path["mel"],
                    "label": path["label"],
                    "pitch": path["pitch"],
                }

        else:
            for i, path in enumerate(files):
                yield i, {
                    "audio": path["audio"],
                    "mel": path["mel"],
                    "label": path["label"],
                    "pitch": path["pitch"],
                }