Abstract
An intelligent sound sensing system is proposed. The system can sense a target signal with better signal-to-noise ratio automatically. The target sound is picked up with disturbing ambient noise sounds by multiple microphones. The multiple outputs of microphones are processed by an adaptive linear filter.
In this paper, a new learning algorithm of an adaptive filter is described. For adaptation, this method needs only a simple cue signal. The cue signal is used for identification of the target signal. Any signal which correlate with the power level of the target signal can be used as the cue signal. As the results, it is not necessary to know transfer functions between sound sources and microphones and/or signal waveforms of sound sources for training. Therefore, spatial arrangement of multiple microphones and relative locations of the target signal source and noise sources can be chosen arbitrarily.
The efficiency of this method is demonstrated by numerical simulations of the self-adapting multiple microphone system.
