Audio content based feature extraction on subband domain

Abstract

Content-based audio feature extraction is key to obtaining important message from audio information. Research in the past several years has focused on the use of speech recognition techniques that are not directly applicable to compressed audio bit stream. However, subband coding based MPEG-1 audio layer III (MP3) is now useful for any system with limited channel capacity for its high quality to bit rate ratio. It has been widely adopted in audio-on-demand, music link via ISDN and digital satellite broadcasting. Message collection is easier if audio content can be extract directly on subband domain. Several useful algorithms are proposed here to manifest this idea.