A second generation wavelet-based adaptive noise estimation method for speech enhancement

A second-generation wavelet based implementation of two adaptive noise estimation algorithms, which do not require explicit use of voice activity detector or signal statistics learning process, is introduced. The fast algorithm utilises a smoothing parameter based on estimation of the wavelet subbands signal-to-noise ratio of the signal. The second algorithm is based on tracking the minimum variance of subband noisy speech signal. A new robust noise-tracking algorithm, which combines a quantile-based noise estimation technique with a modified version of the above smoothing approach, is then introduced and its performance is evaluated and compared to the above two noise estimation methods, using various speech signals contaminated by different levels and types of noise.