| With the rapid development of science and technology,many science fiction technologies that can only appear on the screen have gradually become reality,such as virtual reality and augmented reality technology.In order to serve the public well,these technologies need multidisciplinary development.For instance,in order to produce a shocking presence,virtual reality not only requires to render images in real time,but also the three-dimensional surround sound is essential.If the video's rendering rate is very low,people will obviously feel that the image is not smooth,then the audience will feel dizzy and have a bad experience.Meanwhile,if we enjoy the virtual reality which has abundunt vedios but in absence of sounds,the reality of experience will also be greatly reduced because it is obviously inconsistent with people's living habits.In real life,not only do we observe an object through our eyes,but we also perceive the vary of surrounding space in real time through our ears.This is what the Chinese ancients often say “Observant and alert,very alert''.We want to simulate people's real life as realistically as possible,and give the audience most real and comfortable experience.In first,the basic theory of head-related transfer function(HRTF)is introduced,and various methods of modeling head related transfer function are analyzed in terms of their advantages and disadvantages.Finally,according to relevant papers,head-related transfer function is decomposed into the spherical harmonic domain,then HRTF at an arbitrary position can be represented by a set of weighted spherical harmonic functions.Due to the high complexity of current modeling algorithms,playback is time consuming,the aim of our research is how to quickly playback multiple sound sources in space in virtual reality and achieve the consistency of virtual object's position in the audience's visual and hearing.Because the HRTF is mainly affected by the diffraction of human head at low frequencies,the spectrum is relatively smooth in this frequency band,however,due to the irregular physiological structure of human ear's pinna,the incident sound wave will encounter multiple reflections,the high frequency of HRTF will have a strong fluctuation.Therefore,based on this physical characteristic of head related transfer function,an efficient method combines with the HRTF spectrum characteristic is firstly proposed for modeling HRTF in spherical harmonica domain,and it can achieve the aim of efficient modeling HRTF.In terms of the variability of HRTF's spectrum,the amplitude spectrum of HRTF is fitted and the fitting error at any frequency point is calculated,then the mean and variance of fitting errors are clustered by using a data-driven optimization approach.According to the clustering result,an optimal transition frequency is determined.Based on the optimal transition frequency,the HRTF is separated to model at low frequency and high frequency,and the theoretical basis of our mixed-order algorithm for modeling HRTF is explored.Finally,our algorithm's efficiency is preliminary evaluated by the objective experiments.Comparing with previous algorithms of modeling HRTF in the spherical harmonic domain,we find our mixed-order reduces the storage space by approximately 20%,and improves the computational efficiency by approximately 28%.In order to validate the objective experiments,then two open HRTF database are adopted to design a set of detailed listening experiments,they are the FABAIN database and MIT database.The subjective test is requested to give the similarity score by listening to sounds that are filtered by the measured and reconstructed HRTF,respectively.We calculate the results of subjective test and compare with the results of objective experiments,we find the experimental results of subjective and objective experiments are similar,which also verifies the validity of our efficient method for modeling HRTF algorithm in spherical harmonica domain. |
PDF Full Text Request