Research On Binaural 3D Virtual Sound Generation And Localization

As a typical example of the integration and innovation of a new generation of information technology,virtual reality technology has achieved great success,which not only brings great visual impact to the audience,but also makes the 3d immersive sound field experience be expected by the public.Virtual reality technology hopes to play three-dimensional virtual sound through headphones,so the research on the generation and localization of binaural 3D virtual sound based on headphones has received extensive attention from scholars.Binaural 3D virtual sound is mainly realized by head-related Transfer Function(HRTF),which can generate binaural 3D virtual sound at the corresponding location by convolving the HRTF at the known spatial location with the monophonic audio signal.However,due to the reasons of measuring equipment and time cost,the measured positions of the HRTF are relatively sparse,and the generated 3D virtual sound will have the problem of acoustic image positioning distortion.On the other hand,HRTF is a set of anthropometric parameters for sound comprehensive filtering,with highly personalized characteristics,the use of non-personalized HRTF will appear in the head positioning or confusion.The generation of binaural 3D virtual sound from HRTF database with relatively sparse spatial distribution and small number of samples through interpolation and personalization research is a favorable way to solve the above problems.For this reason,the following works are mainly carried out:(1)Perceptually consistent binaural signal generation based on 3D virtual sound localization.Aiming at the problem of the distortion of binaural 3D virtual sound localization caused by the insufficient spatial resolution of the collected data in the traditional method using head-related transfer function(HRTF)to construct binaural 3D virtual sound signals.In this paper,the distortion law of binaural perception between the position of any sound source in the threedimensional space and the position of the existing HRTF data to synthesize three-dimensional virtual sound is studied.A head-related transfer function interpolation method based on binaural perceptual consistency is proposed,which improves the signal-to-distortion ratio(SDR)by 1.72 d B.Subjective listening tests also show that the proposed method achieves a MOS score of 3.89 on the generated 3D virtual acoustic binaural perceptual localization;(2)Personalized generation of binaural 3D virtual acoustic signals based on HRTF.The traditional method uses general HRTF to generate 3D virtual sound and does not consider the problem that human characteristic parameters such as two ears,head,and torso between subjects will affect the subject’s distortion of3 D virtual sound localization.This paper studies the mapping law between human body characteristic parameters and synthetic binaural 3D virtual sound signals.A deep learning-based head-related transfer function personalization method is proposed,which reduces the root mean square error(RMSE)by 1.13 d B,and subjective tests show that the 3D virtual sound generated by our method reduces the aliasing rate by 6.25%.(3)Binaural 3D virtual sound generation and localization system.Aiming at the problem that the current HRTF acquisition process requires complex experimental environment and it is difficult to accurately obtain the user’s anthropometric parameters,an anthropometric parameter acquisition system is designed for the intelligent acquisition of subjects’ anthropometric parameters,and on this basis,implements a binaural 3D virtual sound generation and localization system,which can be widely used in News,scientific research,education,and entertainment industries,have certain application value.This paper analyzes the correlation mechanism and mapping law of binaural3 D virtual sound localization distortion,and obtains HRTF with high spatial resolution and user personalization through interpolation and personalization of the head related transfer function.A virtual sound generation and localization system,which uses personalized HRTF and mono audio signals to render binaural3 D virtual sound with more accurate sound image localization and greater immersion.The system can be widely used in news,scientific research,education and entertainment industries,and has certain application value.