A new objective quality measure for speech-coding systems based on the estimation of their nonlinear properties

The problem of estimating the quality of coded speech-signals by means of objective quality-measures is addressed. A class of objective measures is introduced which is based on spectral distances. It is shown that the separate treatment of the linear and the nonlinear distortions of the coder yields good estimations of the subjective quality. A composite measure is proposed which is the result of the evaluation of the relationship between subjective analyses and the single objective measures. As the test signal, a speech-model process instead of natural speech is used