The audiovisual collections at the National Library of Sweden (KB) currently contain in excess of 10 million hours (more than 1000 years) of recorded audio and video. The collections are available to verified researchers, authors and journalists, but due to strict legal requirements, only within KB’s premises. The collections make up an unfathomable resource. They can provide invaluable and incomparable insights to Swedish culture, history, literature, art, society and politics, to mention but a few fields. They are equally valuable for meta-studies of fields in which the materials was produced, from broadcasting in general to specific genres, and for technical research areas geared towards (semi-)automated analysis of multimodal materials. This inconceivably large data set constitutes a primary as well as a secondary source in dozens of areas. In practice, the collections are not used. The reason is precisely the size of the material - it takes too long to investigate. This is doubly wasteful: a great resource potential is left dormant, while KB spends substantial resources to maintain and add materials. Speech technology is a key to these collections, as it provides a means to make them searchable. The project develops speech technology methods that can be applied to the audiovisual collections at KB, but also to other, similar materials. The methods and their reference implementations are made freely available omn the national research infrastructure Språkbanken Tal.