Research On Source Localization Based On Classification Of Cross-correlation Function With Microphone Array

As a research hotspot in the field of speech signal processing,sound source localization technology is widely used in intelligent robot,human-computer interaction,intelligent speech recognition system and,etc.The traditional localization method need make use of the spatial information included by the received signals,but not rely on any prior information obtained in advance.In some cases,such as in a conference room or car,the location of the source is limited to a predetermined area,so that representative samples in certain predetermined areas can be measured in advance to establish a data-driven model.The measured signal is associated with the corresponding sound source position.The advantage of this kind of method is that it does not need to accurately model the acoustic environment and has better robustness to reverberation and noise.Therefore,combined with microphone array signal processing technology,this paper studies the localization algorithm based on cross-correlation function classification,and improves the location accuracy of the algorithm in the harsh environment of strong reverberation.The main work of this paper is as follows:1.The different models of speech signal are introduced,and the preprocessing method of speech signal,near-far field model and topology of microphone array are studied emphatically.2.This paper briefly introduces several typical machine learning and classification algorithms for sound source location,including deep neural network,support vector machine and K-means clustering algorithm.Theshortcomings of these algorithms are briefly explained.3.The generalized cross-correlation function weighted by phase transformation has a high degree of similarity between the features extracted from different positions,the accuracy is reduced.In order to improve the localization performance of the algorithm,this paper is based on the sound source localization algorithm of cross-correlation function classification.Two improved algorithms are used:(1)the smoothing filter is introduced,(2)the normalized phase-smooth coherent transform combined with weighted generalized cross-correlation function is used for feature extraction.There are obvious differences between the features extracted by the improved algorithm in different positions.The simulation results show that the two improved algorithms have good performance in strong reverberation environment,and the sound source localization platform in the actual environment is designed and implemented,which verifies the real localization performance of the improved algorithm.4.The importance judgment of each attribute feature in the above algorithm is consistent for classification decision,but there is a certain correlation between different attributes and categories in practical application.In this paper,on the basis of the above research contents,Fisher discriminant criterion function is introduced,and the feature weighting algorithm and Naive Bayes classifier are combined.The classification weight of each feature attribute is established by using Fisher discriminant criterion function.The effectiveness of the algorithm is verified by simulation experiments and practical experiments.