The Design And Implementation Of Automatic Evaluation System For Speech Quality

With the continuous development of wireless network technology and the increasing popularity of network, speech quality directly affects the users’choice. Therefore, according to the communication network quality of service requirements, to better quantitative analysis and evaluation of network voice quality of service, the establishment of a set of objective speech quality evaluation criteria is the key issure that must be condidered in network construction, maintenance and optimization.The voice quality evaluation standard has gone through three stages that from the bit error rate, based on the wireless indicator RxQual, the human ear subjective speech quality assessment to the current MOS value based on the PESQ algorithm software voice quality evaluation.At present, the MOS test generally tested by a professional test equipment, use of the drive test software to implement the MOS test, this method reduces the error of the human ear to listen to the subjective judgment, but there’s still a huge consumption to arrange for a large number of test engineers in the back and forth on the main road,this method can not be effective evaluation of the MOS value for each cell, and cat not carry out analysis of the MOS value in an ordinary voice users to voice complaints can only be passive, time-consuming, laborious post test solution. Then, as the optimization engineer of the network you need a system that can be assessment MOS value of every call in the network reasonablely, and quickly find the root causes of complaints when dealing with voice complaints, even to achieve early warning of an important customer and the key sites in the MOS, this paper is based on this goal, Expand the theoretical research, as well as the combination of the actual project development, application, implementation.This paper explored how to use the spectrum analysis and neural network algorithm, the mass solution of the MOS value of voice users in a wireless communication network and the scientific calculations. It combines domestic and international voice quality assessment as well as neural network, a set of output-based MOS evaluation method, combined with the Changde mobile research and development projects. This thesis describes the voice quality, feature extraction, neural network research use, as well as research and development process of this project. In view of the actual situation of the mobile network, this paper in the implementation of MOS based evaluation method, also collecting the signaling data from network, combined with the mobile network signaling communication process and user associated with the call signalling analysis function, provides a comprehensive network of quality assessment method.Through the speech feature extraction and neural network technology research and practical application of this project, there is a initial realization of the neural network to compute mos scores for speech quality, and has been successfully applied to the project, but an accurate calculation of mos scores degree still need to be improved.After the research and the actual project application, the neural network technique in speech quality assessment can play a very good effect, If the MOS prediction accuracy is increased to more than95%, the system can be popularized in large area, by useing this system,it can greatly improve network voice quality optimization efficiency, and reduce the cost of network speech quality assessment in the network quality evaluation.