TERA: Self-Supervised Learning of Transformer Encoder Representation for Speech | IEEE Journals & Magazine | IEEE Xplore