A new speech quality assessment method based on human perception and its application

We present a new objective method that is based on human perception to estimate the quality of a speech codec. A new perception based spectrum distortion (PBSD) parameter is deduced for this purpose and a parabola curve is adopted to map this parameter into subjective MOS score. Experiments prove that the mean estimation error of this new speech quality assessment method is about 0.1 and maximum estimation variance less than 0.3. The authors then apply this new PBSD distance measure to designing a new speech codec instead of the conventional MSE and improve the subjective quality of decoded speech.