Masked Spectrogram Modeling using Masked Autoencoders for Learning General-purpose Audio Representation