이중 채널 음질 향상을 위한 Blocking Matrix 기반의 잡음 마스크 추정 방법

In this paper, a time-frequency (T-F) soft mask estimation method is proposed based on a blocking matrix (BM) for dual-microphone speech enhancement. For a given T-F bin, the proposed mask is estimated according to the BM magnitude reduction value and phase difference between dual-microphone speech signals. A speech enhancement system employing the proposed method is shown to outperform not only beamforming techniques including delay-and-sum beamformer (DSB) and generalized sidelobe canceller (GSC), but also phase error based soft mask (PBF) in regards to segmental signal-to-noise ratio enhancement (SSNRE).