Selective Active Noise Control System for Open Windows using Sound Classification

Environment noise is of significant concern especially for people living in urban homes. There has been a recent trend in development of active noise control (ANC) systems for mitigation of noises entering homes through open windows. ANC systems capture sounds entering the windows and renders anti-noise signals to cancel the noise inside the room. Averaged fixed-filter approach should be preferred to avoid the use of error microphones because of practicality and ease of implementation in real scenario. However, noise entering windows can be from different sources (bike, bus, train, construction machinery, natural sounds etc.) including both pleasant and unpleasant sounds. Therefore, use of such an average fixed-filter for all types of noises may result in sub-optimal noise reduction performance. In this paper, a selective ANC system is presented, which characterizes the incoming sounds based on temporal and spectral audio features (energy, entropy, spectral roll-off, MFCC, periodicity etc.) and accordingly select control filters from a set of pre-tuned filters based on estimated class of incoming sounds. Both ANC system and sound classification are trained using different types of sounds. Results on real recorded noise samples shows that selective ANC results in better noise reduction than the averaged fixed-filter approach.