4.1 Selecting the number of microphones

Problem

Read this section when mounting microphones on your robot.

Solution

Theoretically, arbitrary sound sources can be separated by 1 more microphone than the number of sound sources. In actuality, the performance of a system will depend on the layout and operating environment of individual microphones. It is rare to know the number of sound sources a priori, so the optimal number of microphones must be determined by trial and error. It is therefore necessary to select the minimum number of microphones required for sound source localization, sound source separation, separated sound recognition. Theoretically, the optimal layout for recording from sound sources in all directions is an equally-spaced layout on the arc of a concentric circle. We presume here that the head of a robot is a perfect sphere; if not, it might result in less than optimal sound source localization and separation in the direction at which the head shape shows a discontinuous reflection. It is better to set a concentric circle in the head, at which continuous reflection occurs.

Discussion

In our three-speaker simultaneous utterance recognition demonstration, three sound sources are separated with eight microphones. The number of separable sound sources is thus below half the theoretical value, 7. The number of microphones may be increased to, for example, 16 to improve performance and separate a larger number of sound sources. However, over-close microphone intervals between microphones have little effect on performance, while increasing calculation costs.