Research Of Virtual Source Reconstruction Based On Perturbation Characteristics Of The Listener To Sound Field

With the rapid development of multimedia film and television technology,three-dimensional audio technology has been initially applied in movies,games,music and other fields.The 3D audio system can reconstruct a virtual sound and create sound scenes at any position in 3D space,realizing immersive and realistic three-dimensional sound effects.The accurate estimate and reconstruction of the virtual sound in three-dimensional space is the key to realizing realistic three-dimensional space perception experience.The existing method for performing 3D audio reconstruction using less speakers is mainly amplitude panning method VBAP,based on a simple geometric model,a virtual sound is reconstructed by vector synthesis,the number of speakers required is relatively small(such as 22.2 channel systems).However,VBAP simply treats the listener as a geometric point and does not consider the perturbation of the binaural auditory characteristics of the listener in the sound field,which results in a large deviation between the spatial position of the virtual sound perceived by the listener and that reconstructed by VBAP.In view of the above problems,based on the CIPIC HRTF database,this paper analyzes the perturbation characteristics of the listener to the sound field,proposes a virtual sound reconstruction model based on the sound field perturbation characteristics of the listener,and conducts relevant experimental verification.(1)We first describes the basic principles of VBAP reconstruction of virtual sound,and analyzes the limitations of the VBAP method in this paper.Then,a virtual sound reconstruction algorithm based on the perturbation characteristics of the listener's sound field is presented.Based on the consistency between the binaural signals synthesized by the multi-speaker signal and the binaural signals produced by the single sound source,a virtual sound image reconstruction model is made.According to the reconstruction model,the frequency domain subband modulation parameters of the loudspeaker signal are calculated,the virtual sound is reconstructed,and the binaural signals of the virtual sound are obtained.The objective experimental verification shows that,compared with the VBAP method,the binaural signal of the reconstructed virtual sound image is closer to the original single source binaural signal.The average spectral distance of the binaural signal is declined by 16.21%.Subjective listening test CMOS score was improved by 0.58 points,indicating that the accuracy of this method for reconstruction of virtual sound images has been improved.(2)The increase in the number of sub-band divisions brings about an increase in the amount of modulation parameter data.In order to explore in which sub-band division mode is more efficient,based on the auditory perception characteristics of the human ear,four different sub-band division methods are selected in this paper.The frequency domain modulation parameters of the loudspeaker signal are calculated,and the binaural signals of the virtual sound image are reconstructed according to the modulation parameters.The experimental verification shows that,as the number of sub-band divisions increases,the subjective test CMOS value gradually increases,and tends to be flat,and the data amount of the modulation parameter increases.Under the 32-subband division method,CMOS score is about 0.60,this method can effectively reduce the data amount of the frequency domain modulation coefficient of the loudspeaker signal while ensuring the accuracy of the reconstruction of the virtual sound.