Environment Recognition from Audio Using MPEG-7 Features

In this paper, we introduce a full use of MPEG-7 audio features for environment recognition from audio for different multimedia applications. Environment recognition from audio files is a growing area of interest, however, compared to other branches of multimedia it is a less researched one. To recognize environment, we utilize total of 17 temporal and spectral MPEG-7 audio low level descriptors as features. The performance is compared with Mel-frequency cepstral coefficients (MFCC) features. Experimental results show a significant improvement of the proposed MPEG-7 based environment recognition over that of the conventional MFCC based features. Zero-crossing rate is appended with the MPEG-7 based feature to yield even better performance. The best performance is achieved with combined MFCC, MPEG-7 and zero crossing features.