Retrieval of "primal signals" of surrounding spherical microphone array recordings

Project thesis

The IEM owns a spherical microphone array to record the directional sound radiation of instruments. It consists of 64 microphones arranged on a spherical surface. With a uniform arrangement, it is possible to determine the sound radiation for a comprehensive set of directions and also to calculate signals for any direction in between by interpolation. Since there are 64 mono-channels, the required amount of data that needs to be stored is quite large. The idea of the project work is to join the information of the 64 signals to form a single signal. Simple filters could be used to recover the 64 microphone signals from the single signal. As the aggregation of a single "primal signal" turns out to be a very delicate subject that requires some effort and the design of suitable algorithms, the focus in this work is put on this matter. To collect the information from the 64 signals, a frequency selective averaging of the short-time signal energy is performed. However this method discards all phase information. To generate a valid and suitable phase, several methods which use the existing phase information have been investigated. The phase estimate by Griffin and Lim algorithm, which does not use the existing phase information, leads to the best results. The sound of the aggregated signal depends strongly on the selected averaging process, in which instead of the spectral energies other exponents of the spectral magnitudes are employed.