Humanoid Active Audition System Improved by the Cover Acoustics

Perception system for humanoid should be active, e.g., by moving its body or controlling parameters of sensors such as cameras or microphones, to perceive environments better. This paper focuses on active audition, whose main problem is to suppress internal sounds made by humanoid movements. Otherwise, such sounds would deteriorate the performance of auditory processing. Our 4-degree-of-freedom (DOF) humanoid, called SIG, has a cover to enclose internal sounds from the outside. SIG has a pair of left and right microphones to collect internal sounds and another pair to collect external sounds originating from the outside of SIG. A simple strategy of choosing a subband of external sounds if sounds from internal microphones in the same subband is weaker than those from external microphones sometimes fails in internal sound cancellation due to resonance within the cover. In this paper, we report the design of internal sound cancellation system to enhance external sounds. First, the acoustic characteristic of the humanoid cover is measured to make a model of each motor movement. Then, an adaptive lter is designed based on the model by taking movement commands into accounts. Experiments show that this cancellation system enhances external sounds and SIG can track and localize sound sources during its movement.