An Evolutionary Approach to Speech Quality Estimation
Adil Raja, Raja Muhammad Atif Azad, Colin Flanagan, Conor Ryan · 2007
In this paper we have employed a genetic programming (GP) based symbolic regression approach to estimate the speech quality as a function of impairments due to IP network and low bitrate coding. A main advantage of GP is that it can produce human-readable results in the form of analytical expressions. Moreover, GP is capable of weeding out irrelevant parameters and concentrating on the most salient ones. These features of GP make our results superior to the past research based on Artificial Neural Networks (ANNs) by Sun and Ifeachor, Mohamed et. al. and on lookup tables by Hoene et. al. We have used PESQ as a reference for evolutionary modeling. The results of proposed models show a high correlation with PESQ. Moreover, our models are suitable for real-time and non-intrusive estimation of VoIP quality.