The Research On Speech Quality Assessment In Commnication System

With the development of communication technology, there are a variety of voice services provided by modern communication networks. The speech communications have became an indispensable part of modern life. In order to provide high quality voice communication service, it is needs to evaluate the performance of communication networks and speech codec. There are many metrics to evaluate the communication system. Among of them, the speech quality is one of the most important metric to assess the performance of speech communication system. So how to accurate and effective evaluate the speech quality becomes an important topic.The method of speech quality assessment is divided into subjective assessment and objective assessment. At present, the subjective assessment is a reliable method for speech quality evaluation. However, it is time-consuming and expensive. In addition, it is not suitable for real-time communication. The objective quality assessment is more convenient, which predicts the speech quality by computer. Thus, objective quality assessment replaces the subjective method, and it has become the main assessment method.This dissertation focuses on the researches of QoE-based speech quality evaluation models. Firstly, the ITU-T P.862PESQ algorithm is studied in detail, and the improved of the PESQ algorithm is proposed. Then, in order to more effective evaluate the speech quality, a new perceptual evaluation of speech quality (NPESQ) model is proposed. After that, a new VoIP simulation platform is established; the speech quality of VoIP networks is evaluated by using the NPEQ model. Finally, multiple regression speech quality assessment models are proposed.The research work can be composed of following four parts:Firstly, the performance of PESQ algorithm is studied in-depth, and the improved of the PESQ algorithm is proposed. The PESQ algorithm is an objective speech quality assessment method for end-to-end speech quality assessment of communication system and speech codecs. It is approved as ITU-T recommendation P.862. However, PESQ is proposed under English environment. The effectiveness of PESQ is varying according to the different language. There exists study of PESQ under English environment, but no study under Chinese environment. So study the effectiveness of PESQ under Chinese environment is necessary. It is need to choose the speech codecs which are the most popular and useful codecs in3G communication systems, including Enhanced Variable Rate Codec (EVRC), EVRC-B and Adaptive Multi-Rate (AMR) codec. In order to improve the accuracy of PESQ for speech codec, the improved of the PESQ algorithm is proposed. The experiment results shown that improved PESQ result can better evaluate speech qualitySecondly, In order to more accurate assess the speech quality of communication, a new perceptual evaluation of speech quality (NPESQ) is proposed. The NPESQ model use the ERB-scle and Moore loudness transform to get the objective evaluation result. The calculation process of the NPESQ algorithm is given. And the new algorithm is verified by doing experiments. Accuracy, consistency and linearity are chosen as performance metric. Through comparing the result of subjective test and objective tests, it is conclude that the NPESQ model maintains the best consistency and linearity with subjective test result.Thirdly, a new VoIP simulation platform is established. Then it is use the new speech quality assessment model (NPESQ) to evualate the speech quality of VoIP networks. The new VoIP platform is different form the other’s. The first advantage is that the latest network simulation WANem softwore is chosen to simulate the real network environment. The second advantage, the script program is written to realize batch processing. In orer to guarantee the accurate of the experiment, it is needs to repeate the experiment many times, such as repeat several hundreds times or even tens of thousands times. It is too time comsuing to do it by manual. So the script program is written to realize batch processing. It can realize the program by automatically many times. Then it analyzes the factors that affect the speech quality of VoIP networks, and use the NPESQ algorithm to assess the speech quality of VoIP.Fourthly, multiple regression speech quality assessment models for VoIP networks are proposed. The multiple regression analysis is a statistical method, which to determine the relationship between dependent variable and multiple variation factors. The QoS parameters play a fundamental impact on the result of QoE. It is use the method of multiple regression analysis to present new mapping model from QoS parameters to QoE for VoIP service. In the VoIP network, there are lots of QoS parameters, three main network impairments QoS parameters is considered, i.e. end-to-end packet loss, time delay and packet reordering. New regression model is gived to evaluate QoE value. According the new regression model, it is can calculate that the speech quality faster and more convenient.