Do speech foundation models really learn words?

Robin Huo, Ewan Dunbar

Self-supervised speech foundation models are now used in a wide array of downstream applications, including traditional speech recognition and as the basis for tokens in speech-awa…