AN AUTOMATED TAGGING APPROACH TO IMPROVE SEARCH AND RETRIEVAL IN A RADIO ARCHIVE

摘要

While a manual process guarantees accurate tagging of archive material, it is very time-consuming. Hence not all media content in a big broadcast archive can be annotated sufficiently. A more automated tagging process could reduce the amount of unannotated content. This paper describes how speech technology is applied on one year of VRT Radio 1 Dutch-speaking news archive content to address the issue above. Different options are discussed and an automated approach is suggested. Techniques such as speech recognition, keyword spotting and keyword extraction are combined to generate automatic annotations. A search engine prototype was implemented to assess the findability of the radio content. Results from the prototype show that the proposed automated approach can improve annotation and search efficiency significantly while still maintaining high precision. The results and lessons learned, which will be presented, are not only valuable to archivists, but moreover to other professional users such as journalists and even the end user.