Speech quality measure for voIP using wavelet based bark coherence function

The Bark Coherence Function (BCF) [1] defines a coherence function with loudness speech as a new cognition module, robust to linear distortions due to the analog interface of digital mobile system. Preliminary experiments have shown the superiority of BCF over current measures. In this paper, a new BCF suitable for VoIP is developed. The new BCF is based on the wavelet series expansion that provides good frequency resolution while keeping good time locality. The proposed Wavelet based Bark Coherence Function (WBCF) is robust to variable delay often observed in internet telephony such as VoIP. We also show that the refinement of time synchronization after signal decomposition can improve the performance of the WBCF. The regression analysis was performed with VoIP speech data. The correlation coefficients and the standard error of estimates computed using the WBCF showed noticeable improvement over the PSQM that is recommended by ITU-T.