| Driving fatigue has serious damage to the traffic safety. Based on the existing study of speech and fatigue study, this paper do further work. It designs the scheme to study fatigue by speech, selects multiple features as fatigue-related parameters, then extracts the parameters and study their rules with fatigue degree.Firstly, this paper starts with designing the scheme to study fatigue by speech. By getting the fatigue degree and the corresponding speech signal, the rule of speech with fatigue change is studied. The mathematical relationship between them is set up, that is, the fatigue can be determined by speech. For fatigue degree of calibration, the fatigue judging table is introduced, which is designed based on the existing authority fatigue questionnaire table to distinguish fatigue degree.Then speech features are selected and extracted to study fatigue, from the time domain, frequency domain and the cepstrum domain. Short average energy, short average range, short average zero-cross rate and short autocorrelation function are selected in the time domain, voiced and unvoiced sounds are selected in time-frequency domain by discrete wavelet multi-scale decomposition, and pitch period is selected in the cepstrum domain to study fatigue. Detailed theory design and the extraction experiments of various parameters complete. And the speech pretreatments are made to guarantee the accuracy before the parameters’extraction.At last, with the level of the fatigue scale of fatigue judging table, all parameters are extracted. Through the calculation and analysis, the rules of each speech fatigue parameters with fatigue are studied. Then it takes comprehensive consideration of all the fatigue-related speech parameters, using the linear multivariate regression analysis method to analyze the relation among fatigue and relative parameters. Finally regression equation is established and it passes the verification, so as to prove it is feasible through speech signal to detect fatigue study. |
PDF Full Text Request