Abstract: The complexity of reproducing binaural 3D audio using head-related transfer function (HRTF) is too high to achieve the real-time generating of 3D audio scene. This paper designs a framing convolution optimization algorithm based on over-lapping reservation and implements a 3D audio real-time generating system, which reduces the processing time of generating 3D audio significantly and realizes the real-time generating and playback of 3D audio. This paper also uses reverberation simulation module to simulate acoustic environments in order to enhance spatial sense of the generated audio and improve the accuracy of sound localization in reproduction system. The experimental results demonstrate that the proposed system can generate real-time 3D audio, and subjective tests using the CMOS standard suggest that the spatial localization of the proposed method is better than compared methods.

Key words: 3D audio, head-related transfer function, reverberation, real-time generating

摘要： 由于使用头相关传递函数重建双耳三维音频的运算复杂度很高，导致无法实现音频的实时三维重建。针对此问题，设计了基于重叠保留法的分帧卷积优化算法，并实现了三维音频实时生成系统，显著降低了三维音频的生成时间，实现了三维音频的实时生成与播放。同时增加混响模拟模块模拟声学环境，提高了生成音频的空间感，使听音者对声音定位更加准确。实验结果表明，该系统能零延时生成三维音频，利用CMOS标准进行主观测试显示其空间定位感明显优于对比方法。

关键词: 三维音频, 头相关传递函数, 混响, 实时生成