Binaural character extraction and clustering of head related transfer function
Hongmei Hu · Journal of Circuits and Systems · 2007
Head related transfer function(HRTF) plays a basal role in 3-D audio systems implementation and virtual auditory space simulation.In view of the similarity and difference of HRTF in different direction,a clustering scheme to obtain the representatives of HRTF for approximation is presented by extracting binaural directional characteristics with the use of improved K-means clustering method in iteratively adjusting the initial clustering center.The characteristics are obtained by making principal component analysis(PCA) on binaural HRTF.Compared to previous clustering method,the proposed clustering results show more proximity to the human localization characteristic and reduce the HRTF reconstruction error.Particular analysis is made on the HRTF length,the cluster size and the comparability of the clustering results for different database.The results are verified by simulation.