Structuring and indexing digital archives of radio broadcasters

This paper describes a pilot project being undertaken by Westdeutscher Rundfunk (WDR) and Deutsche Welle (DW) in cooperation with Fraunhofer Institute for Media Communication (IMK). The project goal is to ascertain the practical usefulness of automatic approaches for the structuring and indexing of digital audio archives of radio broadcasters. Automatic approaches have an enormous potential to complement the conventional annotation methods of radio archivists, who are rapidly becoming overwhelmed by ever-increasing amounts of material that must be archived and growing demands for completely searchable audio collections. Automatic segmentation methods can set cue points in audio broadcasts, which make it possible to skim audio quickly using large intuitive jumps. Classification of segments as speech or non-speech and clustering of speech segments into groups of segments spoken by the same speaker further facilitates browsing. As an additional step, a sort of automatic indexing can be implemented by feeding structured audio through a syllable-based speech recognizer, and performing full-text searches for query words on the resulting syllable transcripts.