SpidR is a self-supervised speech representation model that efficiently learns strong representations for spoken language modeling. It is trained on unlabelled speech using a masked prediction ...