Concept framework for audio information retrieval: ARF

Abstract

The majority of researches on content-based retrieval focused on visual media. However audio is also an important medium and information carrier from the viewpoint of human auditory perception, so it is needed to retrieve for audio collection. Audio is handled by conventional methods as an opaque stream medium, which is not suitable for information retrieval by its content. In fact, audio carries rich aural information with the form of speech, musical, and sound effects, so it could be retrieved based on its aural content, such as acoustic features, musical melodies and associated semantics. In this paper, a concept framework (ARF) for content-based audio retrieval is proposed from systematic perspectives, which describes audio content model, audio retrieval architecture and audio query schemes. Audio contents are represented by a hierarchical model and a set of formal descriptions from physical to acoustic to semantic level, which depict acoustic features, logical structure and semantics of audio and audio objects. The architecture consisting of audio meta-database, populating and accessing modules presents a system structure view of audio information retrieval. The query schemes give generalized approaches and modes concerning how users deliver audio information needs to audio collections. Finally, an audio retrieval example implemented is used to explain and specify the application of the components in the proposed ARF.