Research On Interpolation Technique Of Head-Related Transfer Function

The Head-related Transfer Function(HRTF)is one of the main techniques of binaural three-dimensional audio.The HRTF of a specific spatial position is convolved with the audio signal to generate a three-dimensional sound corresponding to the spatial position.The HRTF changes with the spatial position.The current HRTF can mostly obtain discrete values by measurement.The measured HRTF has a relatively sparse distribution of measurement locations in space due to measurement equipment,time cost,etc.In practical applications,the demand for spatial distribution of HRTF by threedimensional sound cannot be satisfied.Interpolating from a relatively sparsely spatially distributed HRTF database to obtain a spatially dense and accurate HRTF data set is a powerful way to solve the above problems.To this end,this paper carried out related research on HRTF interpolation,the main work is as follows:In the known Head-related Impulse Response(HRIR)interpolation method,basing on the spatial position relationship to calculate the HRIR delay is difficult to express the nonlinear relationship between the spatial orientation and the HRIR delay;The aligned HRIR interpolation method does not guarantee full alignment of the HRIR delay,resulting in a decrease in HRIR interpolation accuracy.Aiming at this problem,this paper uses the nonlinear fitting ability of Radial Basis Function Neural Network(RBF neural network),to design the HRIR delay prediction model,and establishes the mapping relationship between spatial orientation and HRIR delay.Based on this,a HRIR interpolation algorithm based on radial basis function neural network is proposed.Compared with the traditional spatial tetrahedron-based HRIR interpolation method,the average signal-to-Deviation ratio(SDR)of the HRIR obtained by the method is increased by 5dB.The traditional weight-based linear combination interpolation method has fixed weights at each frequency.By analyzing and correlating the HRTF data of different spatial orientations,it is found that the HRTF values at each frequency are inconsistent in space.The fixed weights may result in HRTF interpolation errors.To solve this problem,this paper proposes an HRTF interpolation model based on spectral component correlation analysis.The spectral features of the interpolated HRTF are estimated using the spectral component correlation between known HRTFs to be interpolated point neighbors,thereby achieving a more accurate HRTF interpolation effect.Compared with the traditional linear interpolation method,the average SDR of the HRTF obtained by this method is increased by 1.8dB.The objective experimental results show that,compared with the traditional method based on spatial position weight,the radial basis neural network can effectively predict the HRIR delay information of different spatial orientations;compared with the traditional weight-based linear combination interpolation method,Spectral component correlation analysis can effectively improve the interpolation accuracy of HRTF.