By using the 3D intensimetry algorithm, a sensor probe has a significant advantage at low frequencies, allowing for a compact array cluster that is far smaller than a wavelength. A super compact 3D sound camera is designed to estimate the direction of arrival by calculating the three-dimensional intensity vectors based on the measured pressure data. An array of flush mounted MEMS microphones is configured over the spherical surface of a commercial omnidirectional camera with a diameter of 38 mm. Implementing the operation over a wide frequency range requires the scattering caused by the microphone holder and the irregularity of the array's directional response to be minimized. Although the spherical scattering reduces the effective upper-bound frequency by two thirds, a truncated stellated-octahedral array using five microphones can significantly reduce the spatial bias error at high frequencies. The test results in an anechoic chamber show an average localization error of 3.2 degrees for human voices. The results in a reverberant room with T-30 = 0.66 s reveal an average bearing angle error of 6.5 degrees when a source is positioned within two times the critical distance in the interior space. In such a live room, it is demonstrated that the speakers wearing the face masks can be localized in real-time.