Key Technology Of Speech Recognition Based On Civil Aviation Air-ground Communication

It has always been a critical issue to ensure the safety of aircraft flight via the correct recognition of voice commands during civil aviation air-ground communication.In recent years,accidents caused by air-ground communication occurs sometimes,which is a serious threat to the safety of civil aviation.Therefore,it is an important issue to reduce the risk and ensure the correct transmission of air-ground communication instructionsunder current conditions.In recent years,speech recognition has attracted many attentions in the field of pattern recognition and is widely used in many field.According to the application scenarios and the characteristics of civil aviation air-ground communication,speech recognition and Depth Neural Network(DNN)are combined to deal with the problem of air-ground communication.This thesis compares four different methods by using the actual air-ground communication recordings to deal with noise reduction problems.The experimental results show that the improved spectrum subtraction is more suitable for reducing noise effectively.For speech recognition,a corpus is established according to the inherent sentence structure and the regular patterns about pronunciation of civil aviation air-ground communication.Both GMM-HMM and DNN-HMMacoustic model are separately trained by using this corpus.Firstly,according to the GMM-HMM principle and its training methods,monophone and triphone acoustic models are built separately.In addition,DNN-HMM acoustic model is successfully built by using the improved GMM-HMM triphone model as labels.The experimental results show that DNN-HMM model is better than GMM-HMM model in the phoneme recognition for air-ground communication speech recognition.