A noise spectral estimation method based on VAD and recursive averaging using new adaptive parameters for non-stationary noise environments

A noise spectral estimation method, which is used in spectral suppression noise cancellers, is proposed for highly non-stationary noise environments. Speech and non-speech frames are detected by using the entropy-based voice activity detector (VAD). An adaptive normalization parameter and a variable threshold are newly introduced for the VAD. They are very useful for rapid change in the noise spectrum and power. Furthermore, a recursive averaging method is applied to estimating the noise spectrum in the non-speech frames. In this method, an adaptive smoothing parameter is proposed, based on speech presence probability. Simulations are carried out by using many kinds of noises, including white, babble, car, pink, factory and tank, which are changed from one to the other. The segmental SNR is improved by 2:0 ∼ 3:8dB, and noise spectral estimation error is improved by 3:2 ∼ 4:7dB for the white noise and the babble noise, which are changed from one to the other.