VoIP call quality assesment based on RPROP neural networks

Abstract

The modelling of the network effects on the quality of speech in the Voice over IP networks is the main focus of this paper. The main purpose of the ideas presented here is to achieve high-precision estimation of the speech quality in the environments where the classical approaches of speech quality determination fail. To achieve this high precision a modular neural network model is used to map the effects of a packet loss on the speech quality based on the PESQ reference. To incorporate the temporal effects E-model is partially utilized as well. This way a universal tool capable of harnessing the information about the speech quality for stress testing and monitoring of the local infrastructure has been developed enabling the telephony infrastructure administrators to evaluate the performance and stability of the systems in their hands. Moreover, a high-performance simulation environment has been developed as well to ensure sufficient amount of measurement data for the statistical analysis.