Estimating Speech Spectral Amplitude Based on the Nakagami Approximation
Danhui Xie, Weibin Zhang · IEEE Signal Processing Letters · 2014
In this letter, we propose to simplify the estimation of speech spectral amplitude by using the Nakagami distribution to approximate the Rician distribution, a technique widely used in wireless communication. Based on the complex Gaussian assumptions, the a posteriori density of the clean speech spectral amplitude given the noisy speech spectrum follows a Rician distribution. Most state-of-art speech spectral amplitude estimators are derived based on the Rician distribution and are therefore complicated. We propose to simplify these estimators based on the Nakagami approximation. Six popular estimators are derived. Our results are remarkably simpler, compared with their counterparts based on the Rician distribution. In addition, the united form of our results sheds light on the relation of these estimators. Finally, experimental results demonstrate that the new estimators are close approximations of the original ones.