Objective Evaluation Model Of VoIP Speech Quality Based On Regression Analysis

Voice over IP (VoIP) has been rapidly developed in recent years. The firststep of VoIP is encoding the speech signal, and then transmitting the data packetsover Internet protocol (IP) network. Compared with the traditional publicswitched telephone network (PSTN), VoIP not only has lower cost, but alsosupports multimedia services. However, VoIP employs an unreliable userdatagram protocol (UDP) protocol for transmission, the speech quality can’t beguaranteed. How to improve the reliability and validity of VoIP technologybecomes an urgent issue to solve. Therefore, this promotes us to propose aneffective VoIP speech quality evaluation model.The existing VoIP speech quality evaluation models can be divided into twocategories: subjective models and objective models. Subjective models candirectly reflect people’s perception of speech quality, but needs a large quantity oftime and money. Moreover, since it is difficult to be realized, its application islimited. Perceptual evaluation of speech quality (PESQ) is an intrusive objectivemodel. It has the highest correlation with the subjective model. However, PESQneeds to input the original speech and degraded speech, and it ignores someimpact factors, such as codec, packet loss, delay and jitter. P.563is anon-intrusive model, and it doesn’t need to input the original speech.Nevertheless, P.563has very low correlation with the subjective model. E-Modelis a parametric model, which reflects the impact factors of speech quality.However,it is computationally intensive, and it does not consider the impact ofjitter.To overcome the aforementioned shortcomings of these existing methods, inthis paper, we propose an objective and non-intrusive model which can evaluatethe end–to-end speech quality in real-time network. In Chapter3, we present aspeech quality evaluation model that is based on codec and packet loss. Thismodel combines adaptive multi-rate narrowband (AMR-NB) and Gilbert Modelto generate codec degradation model and packet loss degradation model, thenuses P.563to evaluate the speech quality. Finally, we utilize the regressionanalysis to obtain a codec and packet loss based speech quality evaluation model.In practice, jitter is an important impact factor that can affect the speechquality in VoIP. In Chapter4, we present a speech quality evaluation model that isbased on codec and jitter. We first build a jitter simulation platform to generatedegradation model. Then we use P.563to evaluate the speech quality. Finally, by combining the regression analysis, we get a codec and jitter based speech qualityevaluation model. Simulation results demonstrate the effectiveness of theproposed approaches.