The measurement of the radiation balloon of a loudspeaker involves the acquisition of 2664 responses when acquired with 5º resolution in Theta and Phi angles, each response with magnitude and phase at a high number of frequencies which depends on the measurement spectral resolution. This large amount of information causes, many times, that its analysis is limited to certain frequencies and to certain planes (horizontal and vertical polar plots or isobars). In order to help to investigate radiation balloons, unsupervised machine learning data analysis tools have been applied to automatically group the loudspeaker responses that conforms a full balloon measurement according to their similarity, in order to extract meaningful patterns. Similar algorithms have also been applied to reduce the number of involved frequencies, keeping the same radiation information.
