Modeling personalized head-related impulse response using support vector regression

A new customization approach based on support vector regression （SVR） is proposed to obtain individual headrelated impulse response （HRIR） without complex measurement and special equipment. Principal component analysis （PCA） is first applied to obtain a few principal components and corresponding weight vectors correlated with individual anthropometric parameters. Then the weight vectors act as output of the nonlinear regression model. Some measured anthropometric parameters are selected as input of the model according to the correlation coefficients between the parameters and the weight vectors. After the regression model is learned from the training data, the individual HRIR can be predicted based on the measured anthropometric parameters. Compared with a back-propagation neural network （BPNN） for nonlinear regression, better generalization and prediction performance for small training samples can be obtained using the proposed PCA-SVR algorithm.

INTERPOLATION OF HEAD-RELATED TRANSFER FUNCTIONS USING SPHERICAL FOURIER EXPANSION

A new interpolation algorithm for Head-Related Transfer Function （HRTF） is proposed to realize 3D sound reproduction via headphones in arbitrary spatial direction. HRTFs are modeled as a weighted sum of spherical harmonics on a spherical surface. Truncated Singular Value Decomposition （SVD） is adopted to calculate the weights of the model. The truncation number is chosen according to Frobenius norm ratio and the partial condition number. Compared with other interpolated methods, our proposed approach not only is continuous but exploits global information of available directions. The HRTF from any desired direction can be and interpolated results demonstrate that our obtained more accurately and robustly. Reconstructed proposed algorithm acquired better performance.

Multiple Scattering Between Adjacent Sound Sources in Head-Related Transfer Function Measurement System

To accelerate head-related transfer functions(HRTFs)measurement,two or more independent sound sources are usually employed in the measurement system.However,the multiple scattering between adjacent sound sources may influence the accuracy of measurement.On the other hand,the directivity of sound source could induce measurement error.Therefore,a model consisting of two spherical sound sources with approximate omni-directivity and a rigid-spherical head is proposed to evaluate the errors in HRTF measurement caused by multiple scattering between sources.An example of analysis using multipole re-expansion indicates that the error of ipsilateral HRTFs are within the bound of±1.0 dB below a frequency of 20 kHz,provided that the sound source radius does not exceed 0.025 m,the source distance relative to head center is not less than 0.5 m,and the angular interval between two adjacent sources is not less than 20 degrees.Similar conclusions under different conditions can also be analyzed and discussed by using this calculation method.Furthermore,the results are verified by measurements of HRTFs for a rigid sphere and a KEMAR artificial head.

Head-related transfer function database and its analyses

Based on the measurements from 52 Chinese subjects (26 males and 26 females), a high-spatial-resolution head-related transfer function (HRTF) database with corre- sponding anthropometric parameters is established. By using the database, cues relating to sound source localization, including interaural time difference (ITD), interaural level difference (ILD), and spectral features introduced by pinna, are analyzed. Moreover, the statistical relationship between ITD and anthropometric parameters is estimated. It is proved that the mean values of maximum ITD for male and female are significantly different, so are those for Chinese and western sub- jects. The difference in ITD is due to the difference in individual anthropometric parameters. It is further proved that the spectral features introduced by pinna strongly depend on individual; while at high frequencies (f≥ 5.5 kHz), HRTFs are left-right asymmetric. This work is instructive and helpful for the research on bin- aural hearing and applications on virtual auditory in future.

A head-related transfer function model for fast synthesizing multiple virtual sound sources

A head-related transfer function （HRTF） model for fast and real-time synthesizing multiple virtual sound sources is proposed. A head-related impulse response （HRIR, time- domain version of HRTF） is first decomposed by a two-level wavelet packet and then represented by a model composed of subband filters and reconstruction filters. The coefficients of the subband filters are the zero interpolation of the wavelet coefficients of the HRIR. The coefficients of the reconstruction filters can be calculated from the wavelet function. The model is simplified by applying a threshold method to reduce the wavelet coefficients. The calculated results indicate that for a model with 30 wavelet coefficients, the error of reconstructed HRIR is about 1%. And the result of a psychoacoustic test shows that a model with 35 wavelet coefficients is perceptually indistinguishable from the original HRIR. When multiple virtual sound sources are synthesized simultaneously, the computational cost of the proposed model is much less than the traditional HRTF filters.

On the low frequency characteristics of head-related transfer function

A method to correct the measured head-related transfer functions （HRTFs） at low frequency was proposed. By analyzing the HRTFs from the spherical head model at low frequency, it is proved that below the frequency of 400 Hz, magnitude of HRTF is nearly constant and the phase is a linear function of frequency both for the far and near field. Therefore, if the HRTFs above 400 Hz are accurately measured by experiment, it is able to correct the HRTFs at low frequency by the theoretical model. The results of calculation and subjective experiment show that the feasibility of the proposed method.

Design and validation on a multiple sound source fast-measurement system of near-field head-related transfer functions

Near-field head-related transfer functions （HRTFs） are essential to scientific re- searches of binaural hearing and practical applications of virtual auditory display. High ef- ficiency, accuracy and repeatability are required in a near-field HRTF measurement. Hence, there is no reference which intents on solving the measuring difficulties of near-field HRTF for human subjects. In present work, an efficient near-field HRTF measurement system based on computer control is designed and implemented, and a fast calibration method for the system is proposed to first solve the measurement of near-field HRTF for human subjects. The efficiency of measurement is enhanced by a comprehensive design on the acoustic, electronic and mechanical parts of the system. And the accuracy and repeatability of the measurement are greatly im- proved by carefully calibrating the positions of sound source, subject and binaural microphones. This system is suitable for near-field HRTF measurement at various source distances within 1.0 m, for both human subject and artificial head. The time costs of HRTF measurement at a single sound source distance and full directions has been reduced to less than 20 minutes. The measurement results indicate that the accuracy of the system satisfies the actual requirements. The system is applicable to scientific research and can be used to establish an individualized near-field HRTF database for human subjects.

Experimental analysis of considering the sound pressure distribution pattern at the ear canal entrance as an unrevealed head-related localization clue

By analyzing the differences between binaural recording and real listening, it was deduced that there were some unrevealed auditory localization clues, and the sound pressure distribution pattern at the entrance of ear canal was probably a clue. It was proved through the listening test that the unrevealed auditory localization clues really exist with the reduction to absurdity. And the effective frequency bands of the unrevealed localization clues were in- duced and summed. The result of finite element based simulations showed that the pressure distribution at the entrance of ear canal was non-uniform, and the pattern was related to the direction of sound source. And it was proved that the sound pressure distribution pattern at the entrance of the ear canal carried the sound source direction information and could be used as an unrevealed localization clue. The frequency bands in which the sound pressure distribution patterns had significant differences between front and back sound source directions were roughly matched with the effective frequency bands of unrevealed localization clues obtained from the listening tests. To some extent, it supports the pattern could be a kind of unrevealed auditory hypothesis that the sound pressure distribution localization clues.