Yi Gao(WeChat Work, WXG, Tencent Corp.), Ian Liu(Engineer), Jimeng Zheng(Tencent Corp.), Cheng Luo(Tencent Corp.) and Bin Li(Tencent Corp.)
As stereophonic audio devices, such as smart speakers and cellphones, evolve to be daily essentials, stereophonic acoustic echo cancellation becomes more important for voice and audio applications. The cross-correlation between the far-end channels and the associated ambiguity in the estimated echo path transfer functions lead the misalignment and instability issues with conventional stereophonic acoustic echo cancellers (SAEC). In this paper, we propose a novel SAEC algorithm, which can better model the acoustic echo path between each loudspeaker and microphone. Specifically, filter adaptations are modeled independently by applying pre-whitening in solving the misalignment problem. Improvement in echo suppression capability is evaluated in terms of echo return loss enhancement(ERLE) and wakeup word detection accuracy.