Since great progress has been made in the fieldof speech recognition, recognition ofenvironmental sounds has become an importantissue for researchers to solve. There are manyresearches trying to recognizing severalenvironmental sounds in recent years. Chu et al.[1] used an MP dictionary to decompose thesound and combined this feature with MFCC.They achieve over-90% accuracy rate for somesounds, such as the sound inside a car. Also,McDermott et al. [2] demodulated the sound inseveral frequency channels and calculated thecorrelations between channels. Using this feature,they synthesized the voice of insects and severalother sounds successfully. Anyway, if a certainenvironmental sound could be recognizedefficiently by some extracted features, thetechnology might have various application.However, environmental sounds consist of a largevariety of sounds, which means the features mayhave to be researched specially foe some certainsounds.
展开▼