Research collection
Audio & Speech
Research on Audio & Speech: explanations, evidence, limitations and connections between papers.
3 results
DiffuPlex lets spoken-dialog models predict ahead and revise drafts
DiffuPlex predicts future conversation frames to reduce model calls while preserving tested interaction behavior.
A text-based method makes audio answers traceable to named sounds
Listen-to-Reason answers audio attribute questions from readable clues, outperforming compared audio models on one test.
YuE2 makes a readable score part of full-song generation
YuE2 generates a readable music score before audio; expert listeners preferred planned songs in a matched test.