Research On Automatic Recognition Of The Pilot Response Speech

CVR is an electronic device which is used to record the sound in cockpit. The recorded cockpit information acts as an important data source of airborne aircraft In the regular flying process, the cockpit information not only includes the dialogue among deputy pilots, crews as well as the communication contents with ground control center, but also records lots of noise information. These kinds of noise can give great interference and pollution to cockpit sound, which makes its SNR lower and unstable. Meanwhile, the serious colloquial language as well as its dynamic SNR can be the difficulties for language processing.1. In this paper, we develop to label specifications to meet the needs, including tagging information that the speech recognition system and the speaking way system need.2. This paper constructs a speaking style recognition system. The better experimental results are obtained, and the average accuracy rate is 91.62% when use gaussian radial basis function of SVM. This paper applies one type of the paralinguistics information to the front end processing module of the pilot automatic answering voice recognition system, and detect out formal answer voice sets for the back-end speech recognition module.3. This speech recognition system focus on the commanding voice of the relevant information in the flight, however there are also a lot of cockpit voice independent the dialogue, this article adopts a Filler model with the keyword-based detection method. This paper considers the differences of the channel and the adaptive technology is applied to recognition system of the acoustic model in the channel. Recall rate at 50.45% of the cases, the accuracy rate of 81.56%,and the result of the response voice recognition performance F value reaches 62.79%.4. In terms of robustness, this paper attempts to the wavelet transform and spectral subtraction for speech signal enhancement pilot response denoise. However, since the cockpit voice environment is complex, the performance is declining, But the application method of model compensation, F value of system performance increases 1.5%. Automatic speech recognition system of response of pilot is combined with speaking style recognition system and automatic robust speech recognition system. At the end of the this article, it integrates speech recognition system with robust speech recognition system and completes the construction of the pilot response automatic speech recognition system.