Research On Fatigue Detection Method Of Urban Rail Train Attendant Based On Voice Features

With the rapid development of the urban rail transportation,the workload of driver has gradually changed from physical load to mental load,while too high or too low mental load will cause the driver's fatigue,which will affect transportation safety.Considering current urban train driver needs to implement the tasks of pointing and calling,so it is possible to detect the driver's fatigue by voice.Studying the characteristics of different speech features between awake and fatigue,then using speech features to detect driver's fatigue,and providing theoretical and technical support for the human-machine adaptive interaction of subway vehicles through reasonable classification of driver's fatigue statue.The specific research contents are as follows:1.In order to improve the accuracy of speech feature calculation,a set of speech preprocessing methods are implemented for feature extraction,which including speech denoise,pre-emphasis,windowing and framing.Through the analysis of fatigue effect speech production,the fatigue-sensitive speech features are selected and established.2.The workload of the urban rail train driver is mental load,so this thesis use a 2-back task to induce mental fatigue,and the effectiveness of this task on fatigue induction is verified.3.Speech features are calculated by different methods.Using the rate changes of features' average to analyze the fatigue-sensitive speech features which are not affected by phonemes and the fatigue sensitivity of the aspirated consonants.Using the fatigue-sensitive factor to analyze the different vowels and aspirated consonants fatigue-sensitive level,which provide analysis framework for fatigue classification model.4.Based on the support vector machine algorithm,a fatigue detection model is built.On the basis of fatigue classification with overall sensitive features of vowels and consonants,further classification experiments are carried out with phonemes and the fatigue sensitivity levels of different phonemes are determined,which provides theoretical support for text selection based on the detection of fatigue by speech features.