configs/exp_hifisinger.py

_base_ = [
    "./_base_/archs/hifi_svc.py",
]

speaker_mapping = {'DELETED0': 0, 'opencpop': 1, 'DELETED2': 2, 'DELETED3': 3, 'M4Singer-Alto-7': 4, 'M4Singer-Alto-1': 5, 'M4Singer-Alto-5': 6, 'M4Singer-Tenor-5': 7, 'M4Singer-Alto-2': 8, 'M4Singer-Tenor-7': 9, 'M4Singer-Tenor-4': 10, 'M4Singer-Alto-6': 11, 'M4Singer-Soprano-3': 12, 'M4Singer-Bass-1': 13, 'M4Singer-Bass-3': 14, 'M4Singer-Tenor-2': 15, 'M4Singer-Alto-3': 16, 'M4Singer-Tenor-6': 17, 'M4Singer-Bass-2': 18, 'M4Singer-Alto-4': 19, 'M4Singer-Soprano-2': 20, 'M4Singer-Soprano-1': 21, 'M4Singer-Alto-2#forever': 22, 'M4Singer-Tenor-3': 23, 'M4Singer-Tenor-1': 24, 'M4Singer-Tenor-1#always': 25}

model = dict(
    type="HiFiSVC",
    speaker_encoder=dict(
        input_size=len(speaker_mapping),
    ),
)

preprocessing = dict(
    text_features_extractor=dict(
        type="ContentVec",
    ),
    pitch_extractor=dict(
        type="ParselMouthPitchExtractor",
        keep_zeros=False,
        f0_min=40.0,
        f0_max=1600.0,
    ),
    energy_extractor=dict(
        type="RMSEnergyExtractor",
    ),
    augmentations=[
        dict(
            type="RandomPitchShifting",
            key_shifts=[-5., 5.],
            probability=1.5,
        ),
        dict(
            type="RandomTimeStretching",
            factors=[0.8, 1.2],
            probability=0.75,
        )
    ],
)