1 paper
Peijun Yang, Zhan Jin, Xiaoyi Qin +5
Audio-visual target speaker extraction should recover the person indicated by the video, rather than repeatedly return the acoustically dominant voice. We introduce REAL-2MIX, a Ma…