Research On Spatial Parameter Perceptual Coding In 3D Audio Based On Virtual Sound

Three-dimensional audio system can reproduce sound effects in arbitrary spatial position in three-dimensional space, which attracts widespread interests in recent years. Both MPEG standard working group and China's audio and video national standards organization AVS started 3D audio standardization work.3D audio technology has become a frontier research in audio field.Because it is suitable for multiple users, the speaker-based 3D audio system has been more widely concerned than headphone-based 3D audio system. Compared with traditional 2.1 stereo system and single-layer surround sound system such as 5.1 and 7.1, the existing 3D audio multi-channel system has typical characteristics as well as unique challenges. At first, general 3D audio multi-channel system often includes lots of channels, the corresponding amount of signal data bring enormous pressure to the storage and transmission for channel data. It is urgent for efficient compression of three-dimensional audio multi-channel signal. Second, the key difference between 3D audio and traditional audio lies in the free spatial position of sound image. The accurate localization of spatial image is the key to realize realistic 3D spatial perception. Third,3D audio object coding can provide a flexible interactive listening experience, but with the increase of audio objects, the amount of objects data proportionally grows, which needs efficient spatial audio object coding technology.In response to the above-mentioned needs and challenges, three-dimensional audio spatial parameter perception coding technology based on virtual sound image is studied, including virtual sound image estimation and reconstruction, perceptual coding of three-dimensional audio spatial parameter and three-dimensional audio object coding. It is expected to realize efficient storage and real-time transmission of three-dimensional audio signal, and ensure that the three-dimensional audio perceptual lossless reconstruction and good 3D listening experience for listeners.(1) Virtual sound image estimation and reconstruction based on physical properties of sound fieldBased on the spherical wave propagation theory of point sound source, a new method was proposed to solve the problem of non-conservation of energy between the synthesized sound field with multiple speakers and single sound source in existing virtual sound image estimation and reconstruction method. Based on the total energy conservation of the sound pressure and particle velocity at the listening point, a constraint model was established to estimate and reconstruct the virtual sound. Compared with the existing methods, the average energy distortion at two ears of human head with the proposed method was reduced by 28.68%, as well as obviously improved subjective spatial sound quality.(2) Channel-pair-based spatial parameters coding based on spatial azmimuthal perception characteristics of human auditory systemThe coding bits for spatial parameters can not be effectively utilized in existing methods, which would cause either perceptual distortion or perceptual redundancy in quantization. We found and proved the function mapping relationship between JND (Just Noticeable Difference) of spatial parameters and azimuths. The JND characteristics of spatial parameters was introduced into the perceptual quantization codebook design of spatial parameters, which can effectively remove the perceptual redundancies of spatial parameters and realize efficient compression coding. Experimental results on standard multichannel test sequences showed that the average coding rate of spatial parameters with proposed method can be reduced by 18.76% compared with MPEG in the case of comparable subjective spatial sound quality.(3) Spatial parameters quantization based on local spatial constraintsAiming at the problem that the inefficient compression of spatial parameters in existing spatial audio object coding methods, based on human ear's perceptual characteristics for sound in 3D space, two-dimensional spatial parameters extraction method between objects was proposed. By using the spatial region enclosed by 3D audio objects as the constraint condition, the local vector quantization codebook can be generated dynamically, which can greatly reduce the coding rate of spatial parameters. The test results showed that the average coding rate of spatial parameters between objects was reduced by 29.46% compared with the method of MPEG.