Keywords Recognition Of Pilot Voice Based On Template Matching

The pilot’s time pressure evaluation based on unit function assignment is an important index of work load evaluation.The acquisition of flight crew function distribution based on time series requires the use of pilot’s voice information.However,the complex noise environment in the cockpit and the pilot’s way of speaking varies according to personal habits and mission pressure,which makes the acquisition of pilot voice information very difficult.This paper studies the method of pilot’s speech recognition in strong noise environment,and aims at providing more detailed pilot voice information for unit function recognition.The acquisition of pilot’s voice information plays a key role in the identification of flight crew functions.Speech endpoint detection is an important pre-processing of speech recognition.Aiming at the problem that it is difficult to judge the endpoints of the pilot’s speech in the strong noisy environment of the cockpit,a method of two-step noise reduction and improvement of the endpoint detection condition for the pilot voice data is proposed.The ratio of the speech’s short-term energy to the spectral entropy of the sub-band is taken as the parameter of the double-threshold decision to further improve the accuracy of the endpoint detection method.Under the condition of lacking language models and training samples,we propose a pilot’s speech keywords recognition method based on adaptive matching threshold,which uses dynamic time warping(DTW)technique to average multiple voice samples of a keyword.The keyword reference template is obtained and matched with the input speech,and the matching threshold is calculated using the statistical characteristics of matching distance.The experimental results show that the two-step noise reduction of the pilot’s speech data can further improve the resolution of the endpoint detection parameters to speech and noise.The results of keyword recognition on the voice sample database show that the pilot voice recognition method based on the adaptive matching threshold can obtain the pilot voice information accurately.The proposed approach can provide reference for the minimum flight crew airworthiness compliance verification,provide information about people for the modeling of the human-machine-environment system,and improve the modeling of the human.