Acoustic feature extraction for robust event recognition on cleaning robot platform

If a cleaning robot can understand auditory scene, it can do a variety of roles including that of surveillance for home/office security. However, ego noise generated by the embedded motors can pose a serious problem against accurately performing auditory scene analysis. We propose an appropriate feature set using log-scale filter bank that effectively performs acoustic event recognition robust against ego noise and in home environment. Its effectiveness is demonstrated against the most prominent conventional feature set.