Airborne Speech Signal Detection And Processing Technology

When a pilot is carrying out a flight mission, he not only needs to fly carefully, but also needs topay attention to the plane panel and know about the flight parameters. So the pilot must highly focuson it and work a lot. Apply speech detection and processing technology to the airborne avionicssystem and built an airborne speech system. The system can tell the pilots the flight parameters andhelp them do operating control.It absolutely reduces pilots`work load greatly.Therefore, the thesiscarrys out the research of airborne speech detection and processing technology to realize airbornespeech system control and synthetic output under strong noise environment.Firstly, the thesis analyzes the mechanism and basic characteristics of speech signals, andresearches the mathematical model of speech signal and its expressive methods. Secondly based onresearching short-time energy and short-time zero-crossing rate of endpoint detection algorithm, itsuggests that the algorithm in the airborne noise environment will invalid because of backgroundnoise’s excessive average energy. The thesis puts forward that the result of endpoint detection can beimproved by using the spectral subtraction to reduce the average energy of noisesegments.Furthermore, to solve the problem that the recognition rate and communication accuracy ofspeech system is considered to be decreased under strong noise environment, a speech enhancementmethod is presented which is based on adaptive noise estimation. In the new algorithm, the speechsignal is divided into speech segments and non-speech segments by endpoint detection, and the noiseamplitude spectrums of the two kinds segments are estimated adaptively and respectively. Inorder toimprove the speech recognition accuracy and communction quality, this algorithm also improves thespectral subtraction principle formulas according to the research of hypothesis in spectral subtractionwhich is not common. At last, the airborne speech system is completed by programming which canrealize speech control, speech pretreatment, information index, speech recognition and synthesis.The experiment results show that the improved double threshold endpoint detection algorithmwhich is in airborne strong noise environment can get a better detective effect of endpoint. Speechenhancement algorithm of adaptive noise estimation can eliminate airborne background noise easily,and improve the noise speech recognition rate of the airborne speech recognition system. Throughflight simulation environment testing, in this thesis, the airborne speech system can make the pilot’svoice command get quick and accurate response, as a result that it can simplify the flight operationand improve the efficiency of the flight.