The Research On Objective Measures Of Speech Intelligibility

With the rapid development of the information age, communicationtechnology has made great progress, especially wireless communication,satellite communication network, personally mobile communication and Internetspeech communication, etc. The research of speech signal process has beenwidely applied in various fields. It’s urgently need speech objective measureswith flexible and unlimited. Meanwhile, new ideas need to be continuouslyproposed and improved on the basis of previous study and research. Previousdevelopment to some objective measures is in frequency domain but neglectedto select signal in time domain.Due to the different speech has different contribution to speechintelligibility, for example, the vocal has a larger effect to speech intelligibilitythan voiceless and silent. It’s consistent to the daily experience. The research inthis paper firstly based on the traditional objective measures such as SNRsegand fwSNRseg. A measure with high intelligibility proposed through combiningspeech onset detection. Secondly, another measure put forward based on thetraditional NCM and combined linear combination and regression analysis forfinding new weights. At first, this paper introduced speech and speech signal, and development,classification and performance comparison of speech objective measures. Andthen simply presented speech quality and speech intelligibility, summarized therelated measures, including subjective and objective measures.In this paper, speech onset was selected in time domain, and did AI analysis.This paper described AI, syllable and syllable segmentation, and find that thesyllable onset is generally best preserved in continuous speech while the nucleusvowel may be reduced or altered to fit the speaking rate and adjacent syllables.The syllable coda may be lost entirely. Because of it, the research of speechonset was caused. After speech onset detection methods were introduced indetail, the objective measure and traditional method were combined with theimproved method is put forward, through MATLAB simulation and experimentprove that it did raise the speech intelligibility.The previous study proposed a lot of improvement and evaluation based onSTI. Mostly concentrated in the development of the band important function, thefunction was used in the speech with fluctuation masking. Understand the SingleBIF may not be suitable for estimating the speech intelligibility embedded in thefluctuation masking. The goal is to find a new weight. Combine the linearcombination and the regression analysis, NCM can be divided into20criticalbands and did STI analysis of each frequency band. At last, modified NCM canbe received through the different weighted coefficient. The experiment of MATLAB shows that the improved measure has a good performance inpredicting speech intelligibility.