This demo currently provides Dutch large vocabulary speech recognition transcripts of television items with medical topics (such as "Vinger aan de pols") only. Search functionality and other nice stuff will be implemented soon.
This demo currently provides Dutch large vocabulary speech recognition transcripts of television broadcast news programmes (NOS 8 uur journaal). Search functionality and other nice stuff (such as cross linking) will be implemented soon.
The multimedia retrieval demo demonstrates a speech-based search facility on a few (KRO) television items. Speech recognition was done at UT, search and web-interface at CWI.
A daily recording of the Dutch Radio 1 Journaal is recognized using the ABBOT speech recognizer (see also below). The recognized text is used as a query for a journalist searching related documents in a large collection of newspaper items. The five best matching articles found are shown together with the original recognized text of Radio 1. It is also possible to listen to the original audio segment (password is required).
This demo shows that documents related to the spoken news items can be found using a standard information retrieval technique, even though the search queries will contain unrelated words because of errors in the speech recognition.