Study On Sound Source Localization Algorithm Based On Binaural Auditory And Naive Bayes Theory

As a comprehensive research filed that involves many disciplines such as auditory physiology,auditory psychology and artificial intelligence,sound source localization technology has widely application value in speech recognition,separation and enhancement,artificial intelligence and military filed.The localization of sound source based on auditory features is one of the research hotspots,while the artificial head is an important part of the simulated auditory system,which has important research value for sound source localization and virtual hearing.An artificial head test bench based on binaural sound source localization is designed.Based on the test bench,the measurement of Head-related Transfer Function(HRTF),auditory feature extraction and binaural sound source localization algorithm are studied.(1)According to the requirements of binaural sound source localization,a sound localization system based on binaural auditory is designed.The system includes artificial head,data collection and mechanical movement part,to achieve binaural sound collection,preservation and waveform display for sound source localization providing effective data protection.(2)The measurement principle of the Head-related Transfer Function,the selection of the measurement signal,the measurement procedure and the calculation method are studied,and the Head-related Transfer Functions of the artificial head are obtained.Based on the measured Head-related Transfer Function,the Interaural Time Difference and(ITD)the Interaural Level Difference(ILD)are studied.The Head-related Transfer Function is the comprehensive filtering of the physiological structure of the human head to the sound reached to the ears,which covers a wide range of comprehensives information,but the Head-related Transfer Function id specific,is a combination of multiple personality parameters.So the acquisition of the Head-related Transfer Function mainly depends on the experimental measurement.(3)Aiming at the different frequency band of the Interaural Time Difference and the Interaural Level Difference,and combined with the band-pass filtering function of the human ear cochlear basal membrane,a method of extracting the auditory features of the Interaural Time Difference and the Interaural Level Difference is proposed.Then,the frequency bands that better performed in the Interaural Time Difference and the Interaural Level Difference,and these cues are used for following position.Studies have shown that the frequency band of the Interaural Time Difference cues are 300Hz-1kHz and the frequency band for the Interaural Level Difference are 4kHz-5kHz.(4)In the view of the confusion of front and back orientation of the Interaural Time Difference and the complex relationship between the Interaural Level Difference and azimuth angle,the ITD and ILD of the sub-band are combined into the comprehensive auditory feature as the characteristic parameter of the sound source.Considering the calculation accuracy of the ITD and ILD,sound source localization may be seen a probabilistic problem,so a probabilistic sound source matching algorithm based on Bayesian theory is proposed.Verification shows that this method can effectively avoid the confusion of front and back orientation and improve the sound localization accuracy.(5)Since the speech signal is not continuous,there is the noise segment between speeches,which may affect the localization performance.An endpoint detection method using MFCC coefficients is proposed,which has a good noise robustness,can effectively extract the speech segment,and can help to improve the localization performance.Finally,the virtual sound signal generated by the Head-related Transfer Function and the speeches measured in the real scene are used to test the sound source localization performance.The results show that this method has a good positioning accuracy and noise robustness.