Detecting Sprots Fatigue States Based On Speech Features

Sports fatigue detection is very important to scientific sports training. Accurate sports fatigue detection makes sports training more effective. Without scientific sports fatigue evaluation system, it’s easy to cause excessive sports of athletes. That would lead to fatigue accumulation and fatigue damage if excessive sports happen regularly. The experimenta l scheme of fatigue detection based on speech is designed in this paper. Researching several speech signal feature’s variation with sports fatigue. Analyzing and selecting speech feature. Then classifying speech feature with SVM. The main work of this paper is as follows:Firstly, the program of fatigue detection based on speech is designed. Researching the relationship between sports fatigue and speech signal by capturing 30 subjects’ speech signal in four fatigue levels(before exercise, a little tired, tired, very tired). Then implement ing sports fatigue detection using speech signal. The sports fatigue level is determined by RPE 6-20 scale. We also designed the COP(Center of Pressure) data acquisition program in this paper.Secondly, the features of sports fatigue speech signal are extracted in the paper. The features including short-time average energy, short time average zero-crossing rate, average speech rate, pitch and the particular feature of sports fatigue speech signal(duration of breathing segment). Then analysising the changing regularity of each feature along with the change of fatigue level. And carried out a statistical hypothesis testing based feature analysising. The significant level of the feature’s difference between each two sports fatigue level were calculated, and the analysising results on the significant level of the features’ s difference of each two sports fatigue are relatively wel. The changing regularity of the parameters of balance capacity(COP data) are analyzed.Finally, support vector machine(SVM) is applied to sports fatigue detection. First, we introducing the SVM classification algorithm. Then using SVM to detecting the fatigue samples. In order to ensure the reliability of the detection accuracy rate, 10-fold crossvalidation was introduced in the detection. Finally, the accuracy rate of fatigue detection in several different feature combination were carried out. Among which, the highest feature combination’s accuracy rate is 61.1%. Since the difference of features between “a little tired” and “tired” these two fatigue level is not significant, we remove the fatigue level of “tired”. Then the highest feature combination’s accuracy rate of the three sports fatigue level is 78.4%. So this paper arrives at a conclusion that the detection of fatigue based on speech signal is feasible.