A bark coherence function for perceived speech quality estimation

A new methodology for perceptual quality measure is presented. The new method defines the bark coherence function (BCF) as a new cognition module. False prediction errors are occasionally observed in previously developed perceptual measures when they are applied to the end-to-end speech quality measurement of communication systems. Those errors are mainly caused by the linear distortion of the analogue interface of the system being evaluated. The BCF itself normalizes those effects of linear filtering, so that it is ideal for the speech quality assessment of mobile communication systems. In addition, the proposed scheme does not require the local as well as global scaling, so that it is robust to the difference between the original and received speech levels. To evaluate the performance of the new perceptual model, the regression analysis was performed with CDMA digital cellular, CDMA personal communication service (PCS) and speech codec’s. The correlation coefficients computed using the BCF showed noticeable improvements over the PSQM that is recommended by ITU-T. Robustness of the BCF to various conditions was also tested.