Modelling Speech Quality for NB and WB SILK Codec for VoIP Applications

Abstract

In the last decade, VoIP telephony has gained a tremendous popularity. Skype is one of the most successful and popular VoIP services which has inspired a new generation of VoIP and multimedia users. SILK speech codec is the latest development by Skype and has been integrated into the current version of Skype and is expected to be incorporated into new and emerging mobile devices such as iphone and soft phones. One of the major challenges in every VoIP service is to find an easily accessible objective quality model to predict/measure the perceived speech quality or the degree of user satisfaction. In this paper, we present a regression-based model to quantify the speech quality of the wideband (WB) and narrowband (NB) SILK codec for VoIP applications. The developed model uses the network level parameter (i.e., packet loss) and the application level parameter (i.e., send bit rate) to predict the perceived voice quality in terms of the Mean Opinion Score (MOS). Subjective tests were also carried out to validate the model and good accuracy was achieved (97% for wideband and 91% for narrowband). The developed model can be easily implemented in soft phones or mobile devices to predict voice quality for SILK codec in VoIP applications and can also be used for real-time adaptation and control of VoIP applications to further explore the adaptive feature of the SILK codec in future mobile devices or softphones.