The Research And Implementation Of An Improved Objective Speech Quality Assessment Algorithm

Nowadays, speech communications technologies are involved in every aspects of, and speech quality evaluation is an effective measure to evaluate communication system functions. It can be divided into subjective evaluation and objective evaluation. Subjective evaluation is based on human perception, which reflects true feelings of humans. But when it appears a large amount of data, there exist disadvantages of big measurement errors. Objective evaluation is to use the machine to determine speech quality good or bad, which is short-time, feasible and flexible. Currently, the PESQ algorithm is recommended by ITU, mainly used in the testing of communication networks and coding system. The algorithm is based on modeling of human hearing and perceived characteristics, using method in program to objectively assess speech system, and eventually coupling with subjective evaluation.On the basis of lots of references, the development of speech quality evaluation is described, and some algorithms before PESQ are deeply studied. After analysis of PESQ algorithm on P.862standard, the delay calibration and auditory model in the PESQ algorithm are improved. The improved method promotes calculation efficiency of speech location and subject coupling with object evaluation. And it reduces the complexity of the algorithm by the spectrum region transformation, which improve the efficiency.Now Malden and Psytechnics are the two main companies in making speech quality evaluation system products, which is characterized by full-featured, easy looking, but high cost and relatively complex. For the above, the improved PESQ algorithm is used as a core to design an integral product of speech evaluation system with the comparison prototype of DSLA product made by Malden company, which is low cost, real-time and portable. The designed system is implemented in the platform of virtual instrument, and a large number of speech data are tested. Experiment result show that it could display the related speech parameters in assessment process. And the performances of the measured speech equipments are analyzed by transfer function, electricity level information and frequency response.