A non-intrusive method to assess voice quality over internet

Voice over IP (VoIP) is an emerging service and has gained popularity in recent years. The most challenging task for a service provider is to provide and monitor better voice quality. The nature of quality of service (QoS) in VoIP networks is very variable and hence it is important to be able to measure the QoS offered by the system in real time with a low computational cost. In this paper, we discuss our new model to monitor and predict voice quality non-intrusively using Random Neural Networks (RNN). Our model is based on using network delay, delay variation (jitter), packet loss and different voice codecs as input. The RNNs are also known as G-Networks in which signals are either positive or negative. The advantage of using the RNN model is that it is closer to human perception and has a simple mathematical model compared to other neural networks. We study the RNN model with 4, 5, 6, 7 and 8 hidden layers of neurons. We use ITU-T objective speech quality measurement algorithm PESQ to validate our results. Our model shows a high degree of accuracy.