dataset/label2id.py

LABEL_TO_ID = {
    "intro": 0,
    "verse": 1,
    "chorus": 2,
    "bridge": 3,
    "inst": 4,
    "outro": 5,
    "silence": 6,
    "intchorus": 7,
    "prechorus": 8,
    "gtrbreak": 9,
    "solo": 10,
    "quietchorus": 11,
    "bre": 12,
    "break": 13,
    "introverse": 14,
    "mainriff": 15,
    "chorushalf": 16,
    "instintro": 17,
    "gtr": 18,
    "vocaloutro": 19,
    "verse_slow": 20,
    "fadein": 21,
    "saxobeat": 22,
    "transition": 23,
    "verse1a": 24,
    "build": 25,
    "pre-chorus": 26,
    "outroa": 27,
    "bigoutro": 28,
    "fast": 29,
    "instrumentalverse": 30,
    "section": 31,
    "choruspart": 32,
    "instbridge": 33,
    "guitar": 34,
    "instrumental": 35,
    "breakdown": 36,
    "rhythmlessintro": 37,
    "intropt": 38,
    "interlude": 39,
    "postchorus": 40,
    "postverse": 41,
    "opening": 42,
    "altchorus": 43,
    "stutter": 44,
    "oddriff": 45,
    "synth": 46,
    "preverse": 47,
    "quiet": 48,
    "raps": 49,
    "verseinst": 50,
    "instchorus": 51,
    "chorus_instrumental": 52,
    "slowverse": 53,
    "slow": 54,
    "worstthingever": 55,
    "transition2a": 56,
    "miniverse": 57,
    "refrain": 58,
    "introchorus": 59,
    "drumroll": 60,
    "guitarsolo": 61,
    "versepart": 62,
    "chorusinst": 63,
    "ending": 64,
    "no-vocal-intro": 65,
    "no-vocal-interlude": 66,
    "no-vocal-outro": 67,
    "NO_LABEL": 68,  # Only referring to cases without labels, this portion of labels will be ignored during the loss calculation process.
}

ID_TO_LABEL = {v: k for k, v in LABEL_TO_ID.items()}

# Reserve 64 embedding positions for dataset identifiers in the model.
DATASET_LABEL_TO_DATASET_ID = {
    "SongForm-HX-7Class": 0,  # Categories after rule mapping for HarmonixSet
    "SongForm-HX-Widen": 1,  # Original HarmonixSet
    "SongForm-Private-Raw": 2,
    "SongForm-Private": 3,
    "SongForm-HX-Gemini-Relabeled": 4,  # Rule-mapped HarmonixSet corrected by Gemini
    "SongForm-HX-8Class": 5,  # Rule-mapped (pre-chorus retained)
    "SongForm-Hook": 6,
    "SongForm-Gem": 7,
    "SongForm-Gem-Only-Label": 8,  # Use only segments with labels in SongForm-Gem
}

DATASET_ID_TO_DATASET_LABEL = {v: k for k, v in DATASET_LABEL_TO_DATASET_ID.items()}

DATASET_ID_ALLOWED_LABEL_IDS = {
    0: [0, 1, 2, 3, 4, 5, 6],
    1: [
        0,
        1,
        2,
        3,
        4,
        5,
        6,
        7,
        8,
        9,
        10,
        11,
        12,
        13,
        14,
        15,
        16,
        17,
        18,
        19,
        20,
        21,
        22,
        23,
        24,
        25,
        27,
        28,
        29,
        30,
        31,
        32,
        33,
        34,
        35,
        36,
        37,
        38,
        40,
        41,
        42,
        43,
        44,
        45,
        46,
        47,
        48,
        49,
        50,
        51,
        52,
        53,
        54,
        55,
        56,
        57,
        58,
        59,
        60,
        61,
        62,
        63,
    ],
    2: [0, 1, 2, 3, 26, 39, 64, 65, 66, 67],
    3: [0, 1, 2, 3, 4, 5, 6, 26, 39, 64, 65, 66, 67],
    4: [0, 1, 2, 3, 4, 5, 6, 26],
    5: [0, 1, 2, 3, 4, 5, 6, 26],
    6: [0, 1, 2, 3, 4, 5, 6, 26],
    7: [0, 1, 2, 3, 4, 5, 6, 26],
    8: [0, 1, 2, 3, 4, 5, 6, 26],
}