Environmental sound extraction and incremental learning approach for real time concepts identification

Audio classification has been becoming very important in the field of multimedia researches dealing with audio processing and pattern recognition. Although major of them are focusing in “how”? Audio classifications should be semantic; the majority of them have neglected the importance of preprocessing step of environmental sound recognition, or using simply very classic sound classifiers. The originality of this paper is to construct a complete three modules process, acting dependently, with well defined functions. New method of acoustic sources separation are offered by our process, as well as, a sophisticated encapsulation of binary classifiers is used to promote environmental sound classification, leading to a real time audio concepts identifier. The main finding is that our system was able to recognize accuracy more than 90% of the introduced audio concepts.