Research On Sound Source Localization Algorithm Based On Microphone Array

Sound source localization using microphone arrays has been widely applied in life and production in recent years,and has received extensive attention and research in various fields such as video conference system,smart home and noise detection systems.However,the noise and reverberation problems in the real environment,the real-time problem of the algorithm and the number of microphones still affect the effect of the sound source localization algorithm,especially in the reverberation environment,how to use a small number of microphones for accurate and fast sound source localization.It is the hotspot and difficulty of current research.Based on the above problems,the paper firstly designs an improved speech endpoint detection algorithm,which can extract the speech signal more accurately from the received signal,and then design an improved cepstrum filtering delay estimation technique to estimate the time delay of the speech signal.Finally,the sound source localization is performed by using a cross-shaped microphone array combined with delay estimation.The specific work of the thesis is reflected in the following three points:1、The spectral entropy endpoint detection is insensitive to signal-to-noise ratio and is less affected by signal energy.However,when the noise is close to the spectral distribution of speech signal,the estimation performance decreases rapidly.To this end,a spectral weight-based spectral entropy endpoint detection algorithm is designed.In the algorithm,by introducing a spectrum weighting technique,the spectrum of the received signal is whitened to different degrees to achieve the effect of changing the spectral entropy of the noise and the speech signal.The noise signal and the speech signal can be distinguished by detecting the spectral entropy value.The experimental simulations show that the spectral weighted spectral entropy endpoint detection has better estimation performance than the spectral entropy method.Under white noise,both have better detection performance;under other types of noise,When the SNR is 0dB-5dB,the correct rate is increased by 10%-50%;when the SNR is above 5dB,the detection accuracy is increased about 30%.2、The cepstrum pre-filtering delay estimation technique can effectively suppress the reverberation component in the received signal.However,since the minimum phasecomponent is eliminated,the all-pass component signal is whitened on the amplitude spectrum,and the signal amplitude spectrum is not effectively utilized.The effective information is such that the estimated peak is reduced.In this paper,an improved time delay estimation technique is designed.Based on the all-pass signal,the signal amplitude spectrum is added and the improved phase-weighted generalized cross-correlation function method is used to estimate the delay.Finally,simulations are carried out in three different reverberation time environments,which further proves that the proposed algorithm has better estimation performance than the generalized cross-correlation delay estimation algorithm.3、The one-dimensional linear microphone array has the advantages of simple structure,easy mathematical modeling,and a small number of required microphones,but the positioning blurring problem occurs during the positioning process,and the correct position cannot be determined.For this thesis,a cross-shaped array with four microphones is designed.The delay of the two pairs of microphones is used to eliminate the positioning fuzzy problem.The angle of arrival estimation is combined with the DOA estimation model and time delay.Finally,a total of 60 sets of sound source localization experiments are designed in the real environment to verify the algorithm,which further proves that the proposed algorithm has more accurate estimation performance.