A comparison of parametrically different objective speech quality measures using correlation analysis with subjective quality results

In this study, 500 parametrically different versions of 6 "basic" objective speech quality measures are defined. Each objective measure is applied to 1,056 distorted speech samples, and a statistical correlation is done with subjective quality measures on the same data base. Using a nonparametric pair-wise rank statistics, the best values for the various parameters are investigated. The results of this study clearly show that a number of the parametric forms of objective speech quality measures currently in popular use are non-optimal.