基于偏最小二乘回归的与头相关传递函数的个人化

胡红梅; 周琳; 马浩; 吴镇扬

doi:10.3724/SP.J.1146.2007.00146

基于偏最小二乘回归的与头相关传递函数的个人化

doi: 10.3724/SP.J.1146.2007.00146

胡红梅^{①②;周琳},
周琳,
马浩,
吴镇扬

基金项目:

国家973计划项目(2002CB312102)资助课题

计量
- 文章访问数: 3193
- HTML全文浏览量: 92
- PDF下载量: 1188
- 被引次数: 0
出版历程
- 收稿日期: 2007-01-24
- 修回日期: 2007-07-19
- 刊出日期: 2008-01-19

Head-Related Transfer Function Personalization Based on Partial Least Square Regression

Hu Hong-mei^{①②;周琳},
Zhou Lin,
Ma Hao,
Wu Zhen-yang

摘要

摘要: 该文提出了一种基于偏最小二乘回归(Partial Least Squares Regression, PLSR)的与头相关传递函数(Head Related Transfer Function, HRTF)的个人化方法。通过对HRTF进行预处理和主元分析，并对人体参数进行筛选，只要相对简单的人体参数测量，就可利用PLSR得到特定人的HRTF。客观误差分析和主观声音定位测试结果表明估计的HRTF与实际测量的HRTF之间不仅均方误差较小，而且感知区别不大；同时由PLSR估计的个人化HRTF在水平面上的主观测试定位准确率明显优于非个人化HRTF，也优于由最小二乘回归(Least Squares Regression, LSR)估计的个人化HRTF。
- 偏最小二乘回归; 最小二乘回归; 与头相关传递函数; 声音定位测试
Abstract: A method is proposed to obtain individual Head Related Transfer Function (HRTF) based on the Partial Least Square Regression (PLSR), in which anthropometric parameters are selected by correlation analysis and the original HRTF data are preprocessed before principal component analysis. Then objective simulation experiment and subjective sound localization experiment are implemented to evaluate the method. The experiments show that the estimated HRTF not only has small mean square error, but also has little perception difference from the measured one；the localization performance of the individual HRTF estimated by PLSR presented in this article is more accurate than that of the non-individual HRTF obtained from the database and also more accurate than that estimated via Least Square Regression (LSR).

HTML全文

参考文献(1)

Blauert J P. Spatial Hearing. Revised edition. Cambridge,MA: MIT, 1997: 37-176.[2]Wenzel E M and Doris J K, et al.. Localization usingnonindividualized head-related transfer functions[J].J. Acoust.Soc. Am.1993, 94(1):111-123[3]Wu Z Y, Chen J S, and Richard A R, et al.. Measurementand estimation of free field to eardrum transfer function forvirtual acoustics space research. ACTA Acustica, 1996, 21(1):645-652.[4]Zotkin D N and Hwang J, et al.. HRTF personalization usinganthropometric measurements. Proc. IEEE WASPAA, NewPaltz, NY, USA, 2003: 157-160.[5]Middlebrooks J C. Virtual localization improved by scalingnonindividualized external-ear transfer functions in frequencyJ[J].Acoust. Soc. Am.1999, 106(3):1493-1510[6]Otani M and Ise S. Numerical calculation of the head-relatedtransfer functions by using the boundary element method[J].Proc. WESTPRAC 2000, Kumamoto, Japan.2000, VII:305-308[7]Nishino T and Nakai Y, et al.. Estimating head relatedtransfer function using multiple regression analysis. IEICETrans. A, 2001, J84-A: 260-268.[8]Hu H M.[J].Zhou L, and Wu Z Y, et al.. Head related transferfunction personalization based on multiple regression analysis.CIS06, Guangzhou, China.2006,:-[9]Algazi V R and Duda R O, et al.. The CIPIC HRTF database.IEEE ASSP WASPAA, 2001: 99-102.[10]Kistler D J and Wightman F L. A model of head-relatedtransfer functions based on principal components analysisand minimum-phase reconstruction[J].J. Acoust. Soc. Am.1992, 91(3):1637-1647[11]Wu Z Y and Chan H Y, et al.. A time domain binaural modelbased on spatial feature extraction for the head-relatedtransfer function[J].J. Acoust. Soc. Am.1997, 102(4):2211-2218[12]Wu Z Y and Wang W B. A spatial hearing model based onspatial feature extraction and artificial neural network. AcatAcustica, 1999, 24(6): 645-652.[13]Wang H W. Partial Least Square RegressionMethod andApplication. Beijing: National defense Industry Press, 1999:150-171.

施引文献

资源附件(0)

访问统计