Research Of Key Words Recognition In Aircraft

With the development of aviation technology,the ratio of aircraft accidents due to the failure of the aircraft itself or the failure of the control system is decreasing.Man-made factors of flight accidents have accounted for 80% of the total accident,which fully shows that human factors have become one of the most important factor leading to the aviation accidents.Analyzing the workload of the pilot is one important way to analyze human factors.In order to measure the pilots’ workload,we need to identify and analyze the keywords in the pilots’ speech content which are related to the actions on the checklist,and then determine the workload of the pilots.Although there are many mature tools for keyword recognition in domestic and overseas,we will still face many problems that professional terms cannot identify,which eventually leading to time-consuming and poor recognition results due to the lack of relevant technical terms of the training template.Therefore,it is necessary to study the keyword recognition technology in the cockpit.The main contents of this paper include the following aspects:Firstly,speech enhancement is researched based on the analysis of speech and noise in the cockpit.Multi-taper spectrum and improved NMF are used in the de-noise steps,both of them have excellent performance,and the experiments are shown that the multi-taper spectrum algorithm is more suitable in aircraft.Secondly,the endpoint detection and feature extraction are studied.For endpoint detection,the classical double threshold detection algorithm is used to judge whether the speech is in silence or not,marking the start point and finish point.For feature extraction,MFCC feature is extracted for each keywords and speech words,which lay base for the recognition.Thirdly,the keywords recognition is researched on the improved DTW.The DTW is improved by the sliding windows in the match,the threshold estimation,and the adaptive module match which can recognize the keywords of pilots effectively in the experiments.Forth,on the basis of the above work,a speech keyword recognition system framework for aircraft cockpit is constructed and systematically implemented the acquisition,enhancement,endpoint detection,feature extraction and keyword recognition of speech signal.The experimental results show that the system can effectively identify key words of pilots.