반향환경에서의 DUET에 기반한 암묵 음원 분리

In this paper, we present a two-stage algorithm based on the DUET(Degenerate unmixing and estimation technique) for improving the performance of blind source separation in reverberant environments. In the first stage, we propose to use frequency-wise attenuation and delay parameters which were irrelevant to frequency in conventional DUET. In the next stage, we propose a mask re-estimation algorithm based on the precedence effect. Experimental results successfully demonstrate better performance of the described method in reverberant environments.