Automatic Personality Estimate,Recognition And Application Research Based On Chinese Prosodic Features

Personality traits occupy an important position in terms of interpersonal relationships,career planning.Under the current backgrounds of Internet,mobile,personality prediction are generally based on text questionnaires or handwritten text form.But such time-consuming ways have lengthy predicting step while the effect is susceptible to subjective factors,and the measurement of data are difficult to obtain.To solve this problem,researches based on speech features to perceive and recognize personality automatically have been done.But these studies are focused on English,French,German and other western languages.Chinese are distinguishing with other languages in the pronunciation and the acoustic characteristics,and Chinese people are different from Westerners in personality.Since the speech corpus for Chinese personality p rediction is lacking,and scoring classification model accuracy is low and the kinds of analyze characteristics are lacking too,a study has been developed as follows:(1)Collecting the corpus of Chinese speech.Gathering 1936 clean speech segment from 78 Chinese people and the "Big Five" personality inventory self-assessment as the initial dates.Then extracting the maximum,minimum,average,and standard deviation as statistical characteristic through pitch,sound intensity,the first three formants as well as time voice and non-voice time of prosodic features.Thus the personality prediction experiment of Chinese speech co rpus can be structured by those characteristic values and the scores of the "Big Five" personality quantity table.(2)Building Chinese personality recognition model.Experiment of multi-disaggregated statistical learning results shows: first,the third formant has no significant difference on the personality recognition with the first and the second formant which are traditional ways.Second,better recognition accuracy can be found while pitch,intensity,the first three formants and speaking rate combinations are used as classification parameters in character recognition model.What's more,in character recognition modeling,establishing male model and female m odel based on gender has six percent higher accuracy rate than a mixed model with both male and female.(3)Completing personality recognition photetype system.On the basis of previously obtained experimental model,a character recognition system prototyp e based on Chinese speech prosody can be designed.The personality assessment mode which is easy to operate.And can be promoted and popularized in a number of applications of the internet and the mobile environment to provide users with additional social services in an accurate and efficient way.