Research Of Dynamic Spatial Parameters Quantization In 3D Audio Coding Based On Frequency Subband

With the spectacular boom of 3D video,the three-dimensional audio has attracted a wide attention.The three-dimensional audio system should reconstruct three-dimensional sound effect with three free degrees of horizontal direction,vertical direction and distance,and the sound image could be reconstructed at any spatial position,realizing a full range of immersive and surrounded audio scene.With the process of three-dimensional audio standardization at home and abroad,three-dimensional audio has become a hotspot at the forefront of multimedia.The 3D audio is classified into channel-based 3D audio and object-based 3D audio,and object-based 3D audio allows the users to adjust background,volume and so on,and meets personalized requirements of interaction.AVS proposed that 3D audio coding system should support reconstruction of 64 audio objects.However the side information bitrate would increase with the increasing number of objects sharply,which is far more than bitrate of downmix signal.In order to handle this issue,a parameter extraction between three-dimensional objects employed and a dynamic spatial parameters quantization in 3D audio coding based on virtual sound of frequency subband by using location of a group of three objects is proposed.(1)Spatial parameters quantization based on spatial location constraint of 3D audio Objects.Existing spatial audio object coding extracted a set of parameters for each audio object,and quantization codebook is non-uniform quantization that is confined to the statistical character of data distribution.Because it reduced redundancy only considering data distribution,bitrate of spatial parameters increased linearly with the increasing number of objects.Aiming at this issue,this approach designs global vector quantization codebook in three dimensional space based on human ears' spatial orientation sensitive features of spatial audio source.The existing method to compute a set of spatial parameters for coupled audio objects is changed into extracting coupled spatial parameters for a tirple of audio objects.By the region surrounded by the three objects as a constraint,a dynamic local vector quantization codebook could be generated.This approach could achieve a much lower bitrate of parameters with equal reconstruction and quality of audio objects.The tests show that the average bitrate for spatial parameters of proposal is reduced by 32%,compared with MPEG SAOC.(2)Dynamic partition method for frequency subband based on spatial parameters.In order to balance bitrate and audio quality,subband partition is necessary and human could only hear one virtual sound in each partition.Exist subband partition is based on frequency awareness of cochlea,which is sensitive to low frequency and not very sensitive to high frequency.So the partition of low frequency is fine and that of high frequency is loose,which doesn't take influence of spatial information into consideration.Spatial parameters calculated from the standard test sequences is full of universality,and is clustered,trained and selected to build a library of standard subband partition.In this library a suitable subband partition is elected for pending frequency spectrum of objects,and then spatial paremeters are computed in the elected subband.Tests show that with the same bitrate of spatial parameters,this approach could enhance the audio quality of reconstructed objects.