Objective non-intrusive conversational voip quality prediction using data mining methods

Abstract

Nowadays, there is a growth in the number of applications running on the Internet involving real-time transmission of speech and audio streams. Among these applications, Voice over Internet Protocol (VoIP) has become a widespread application based on the Internet Protocol (IP). However, its quality- of-service (QoS) is not robust to network impairments and codecs. It is hard to determine conversational voice quality within real-time network by using ITU-T standards, PESQ and E-model. In this research, three data mining methods: Regression-based, Decision tree and Neural network were used to create the prediction models. The datasets were generated from the combination of PESQ and E-model. The statistical error analysis was conducted to compare accuracy of each model. The results show that the Neural network model proves to be the most suitable prediction model for VoIP quality of service.