XTREME-S covers four task families: speech recognition, classification, speech-to-text translation and retrieval. Covering 102
languages from 10+ language families, 3 different domains and 4
task families, XTREME-S aims to simplify multilingual speech
representation evaluation, as well as catalyze research in “universal” speech representation learning.
