Spoken content metadata and MPEG-7

The words spoken in an audio stream form an obvious descriptor essential to most audio-visual metadata standards. When derived using automatic speech recognition systems, the spoken content fits into neither low-level (representative) nor high-level (semantic) metadata categories. This results in difficulties in creating a representation that can support both interoperability between different extraction and application utilities while retaining robustness to the limitations of the extraction process. In this paper, we discuss the issues encountered in the design of the MPEG-7 spoken content descriptor and their applicability to other metadata standards.