Comparison of HRTF Interpolation Algorithms based on Neural Network

Head-related transfer functions (HRTFs) are used in immersive audio rendering applications. In many cases, these functions have to be calculated or measured in many relative positions between head and source requiring a large amount of time and significant computational resources. Therefore, the generation of HRTFs becomes crucial and interpolation procedures can solve this problem. An extensive analysis of an interpolation method based on neural network is presented taking into consideration the state of the art and the use of a real dataset filtered by a refinement procedure. The neural network interpolation approach is compared with conventional nearest-neighbor interpolation methods. The investigation based on differences between interpolated and measured HRTFs shows promising results of the proposed methodology that are in line with traditional interpolation techniques.