Objective Speech Quality Measurement of IP Telephony using artificial speech signals

摘要

IP telephony is one of the most promising applications over IP networks. Since such services are implemented using best-effort networks, it is important to monitor the quality of service and take necessary actions, as well as to design it properly in advance. To do this, it is indispensable to develop a method for estimating subjective quality accurately and effectively without subjective assessment, which is time-consuming and expensive. Therefore, a method that estimates the subjective quality based on physical characteristics of transmitted speech, so called "objective speech quality measurement," is desirable. We investigated the applicability of the artificial speech signals, which have the average characteristics of real speech, to the objective speech quality measurement in order to guarantee the reproducibility of the objective measurement. The experimental results show that the artificial voice defined by ITU-T Recommendation P.50 can be applied to the objective evaluation of speech with packet-loss degradation if the speech activity factor is properly controlled. Our study also reveals that the speech quality varies even if the speech activity factor is the same, depending on the pattern of talk spurts.