Abstract: In some certain interactive scenes recently, object-based audio coding can provide more flexibility for     receivers in personalized rendering to create particular mixture of objects. However, if these objects are coding separately, the total bitrate of all objects goes up sharply with the increasing number of objects. Spatial audio object    coding (SAOC) proposed by MPEG mixes all objects to one mixture signal and some side information for each object. However, when there are more than 32 objects to encode, the bitrate of side information will be high with the increase of object number and can be much more than half of the total bitrate. Aiming at this problem, this paper proposes the spatial parameters quantization based on spatial location constraint of 3D audio objects. On the theoretical basis of summing localization, the location constraint between virtual sound source and original 3D audio objects is utilized to find out spatial constraint region, and then the local spatial quantization codebook is figured out for extracted spatial parameters in quantization. Objective and subjective experimental results indicate that, with roughly equivalent perceptual quality and spatial location, the side information bitrate of the proposed approach is 30% lower than that of SAOC on average.

Key words: multi-object coding, spatial parameters, spatial location constraint, dynamic quantization

摘要： 近年来，在某些交互场景中，面向对象空间音频编码能够允许用户更加灵活地对特定对象进行个性化的渲染和重组。然而，如果对象音频分别编码会导致整体码率随着对象数目的增加而同步大幅增加。MPEG组织提出的面向对象音频编码（spatial audio object coding，SAOC）可以将所有对象下混成一个单独的混合信号，同时为每个对象提取少量的边信息。但是，当其对超过32个对象音频进行编码时，边信息的码率会随着对象数目的增加而增大，甚至会远大于下混对象的码率。为了解决这个问题，提出一种在面向对象编码中基于空间位置约束的空间参数动态量化方法，以叠加定位原理为理论基础，利用虚拟声源的空间位置与产生该声源的音频对象的空间位置之间的约束关系，确定出空间约束区域以及局部空间量化码本，并针对提取出的虚拟声源的空间方位，进行空间参数的量化编码。最后给出主观实验和客观实验，表明在音质和空间方位大致相当的情况下，边信息的码率比SAOC方法降低约30%。

关键词: 多对象音频编码, 空间参数, 空间位置约束, 动态量化