A compact representation of the head-related transfer function inspired by the wavelet transform on the sphere

The head-related transfer function (HRTF) characterizes the propagation of sound from its source to the listener's ears. It is commonly used in the research of spatial sound and its applications. The HRTF takes different values depending on the source's position relative to the listener and its frequency. Storing its values directly results in considerably large data sets. Previous research to encode the HRTF using harmonic functions cannot handle local features efficiently. This paper advances a new analysis method to represent raw HRTF data using local functions of the azimuth and elevation angles. The proposal treats features of different scales separately, in a manner similar to the wavelet transform. This allows us to compress HRTF data considerably while accurately preserving its features at all scales. The proposal yields better accuracy than harmonic analysis methods for the finer spatial patterns; this holds even when our method uses more aggressive compression.