Time varying forgetting factor for the noise estimation in multi-channel noise reduction

We introduce a time varying forgetting factor for the noise estimation in multi-channel noise reduction. In conventional multi-channel noise reduction system, the noise statistics are estimated during noise-only periods and kept fixed during speech-present periods. For deciding whether the current frame is speech-present period or not, a voice activity detector (VAD) is usually used. Instead of this conventional scheme that needs an explicit VAD, we adopt a time varying forgetting factor which is parameterized from the normalized cross correlation (NCC). The simulation results show that multi-channel noise reduction using the proposed method yields better performance in several typical objective measures.