An audio signatures indexing scheme for dynamic content multimedia databases

This paper describes a method of content indexing for dynamic multimedia databases using binary signatures of the audio signals. The analysis scheme focuses on the audio objects of the multimedia environment, extracts semi-stationary audio segments and analyzes their key features, These are: the fundamental frequency F0, the first three formants F1, F2, and F3, the associated phoneme and speaker identification information. Binary signatures are associated with the audio segments which they describe. For each audio segment these key features are encoded within the signature and information retrieval can be performed on the multimedia database querying the corresponding signatures of the streamed audio. The signature structure serves as an index of the associated audio objects assisting their retrieval.