Regresar

Optimum ANN architecture for HRIR interpolation

Abstract:

This chapter presents an optimal configuration for spatial interpolation of Head Related Impulse Responses (HRIRs) by using artificial neural networks (ANNs). The proposed method is capable to reconstruct the desired functions by providing only the directional information (azimuth and elevation angles). In order to cover the whole reception auditory space, without increasing the network complexity, a set consisting by multiple networks was adopted, where each network is responsible for a specific area. The three main factors that influence the model accuracy are investigated: the network architecture, the covered area size and the initial HRIR time delays and the optimal setup is presented. Error evaluation is performed in terms of time and frequency domains, considering wide spectrum and octave bands. The computational effort for the optimum setup is shown to be 50% smaller than traditional interpolation methods. Moreover, since all errors investigated reached very low levels, the proposed system is suitable for auralization systems.