Research Of Fatigue Detection Based On Speech Analysis

Fatigue,which is a kind of human body's natural response and self-regulation for protection,is a complex physiological and mental phenomena.Fatigue can be divided into two kinds: physiological fatigue and mental fatigue.It occurs when human body's physiological or mental strength reaches to a certain stage.With the increase of fatigue,the accident caused by mistakes will increase.On the contrary,the athletic ability and work efficiency will decrease.Obviously,it is vital important to study the fatigue's effects on different crowds' physical health,safety in production and work(e.g.safe operation and drive)and so on.Fatigue can be detected from physiological parameters,but this method relies on the instrument and can not be carried out in real time.Detecting fatigue from speech has the advantages of simple operation,real time and so on.In recent years,there has been a preliminary study on fatigue detection based on speech analysis.However,the research on the mental fatigue,especially the combination of physical and mental fatigue,is still relatively small.In our paper,we propose a fatigue detection system to recognize the level of both physical and mental fatigue,which mainly includes:1.We construct a fatigue corpus which include both sports fatigue(SPF)and learning fatigue(LEF)to overcome the lack of corpus.The classification of fatigue degree was carried out by subjective fatigue scale.Then,objective relationships between physiological parameters and fatigue are employed to illustrate the validity of the constructed corpus.2.Selection of effective speech feature parameters.Firstly,according to the influence of human fatigue on the organs of speech,the speech feature parameters which may affect the recognition rate are selected.By comparing recognition rates of different combination of feature parameters,the most effective combination for identifying different levels of fatigue is obtained.3.A single-gender&single-character fatigue detection system based on voice is proposed.In this paper,the features of speech signals,which are sound samples,areextracted.Traditional machine learning method SVM and the integrated method Ada Boost classifier are employed for single-gender&single-character fatigue training and recognition,respectively.According to experimental results,we compare the pros and cons of traditional machine learning method and integrated methods.4.We construct and optimize the cross-gender&single-character fatigue detection system and single-gender fatigue detection system based on speech.Owning to the distribution diversity on different genders,most of the researches on speech fatigue detection are focused on the study of single-gender&single-character fatigue detection.In this paper,SVM is utilized to identify the fatigue degree between different genders and different kinds of fatigue.The TrAdaBoost algorithm is first used to optimize the fatigue performance of cross-gender&single-character fatigue detection and single-gender fatigue detection system.Based on the method we propose,fatigue detection based on speech is carried out and compared with the physiological parameters.The experimental results show that the proposed method can effectively detect the fatigue.What's more,the study of continuous fatigue is carried out by simplifying the degree of fatigue.How to carry out effective and accurate analysis of continuous fatigue detection needs to be further studied.