Concept Framework for Audio Information Retrieval: ARF

The majority of researches on content-based retrieval focused on visualmedia. However audio is also an important medium and informationcarrier from the viewpoint of human auditory perception, so it is neededto retrieve for audio collection. Audio is handled by conventionalmethods as an opaque stream medium, which is not suitable forinformation retrieval by its content. In fact, audio carries rich auralinformation with the form of speech, musical, and sound effects, so itcould be retrieved based on its aural content, such as acousticfeatures, musical melodies and associated semantics. In this paper, aconcept framework (ARF) for content-based audio retrieval is proposedfrom systematic perspectives, which describes audio content model,audio retrieval architecture and audio query schemes. Audio contentsare represented by a hierarchical model and a set of formaldescriptions from physical to acoustic to semantic level, which depictacoustic features, logical structure and semantics of audio and audioobjects. The architecture consisting of audio meta-database, populatingand accessing modules presents a system structure view of audioinformation retrieval. The query schemes give generalized approachesand modes concerning how users deliver audio information needs to audiocollections. Finally, an audio retrieval example implemented is used toexplain and specify the application of the components in the proposed ARF.