We've been writing about Nexidia for years on ResourceShelf. They're a multimedia engine but unlike others that use either closed captioning
or speech-to-text they use phonetics (phonemes) to break the spoken word down to phonetic sounds (about 40 English phonemes) and 400 in all of spoken language so yes, the technology can work with any language for faster and overall more accuate indexing and retrieval.

It's been cool to write about and see but now a public demo is available from an Atlanta TV station.

Look for the search box (news and some exclusive web content) at
http://www.wxia.com

More about the technology and related links here:
http://www.resourceshelf.com/2006/10/08/multimedia-search-technology-unveils-public-demo-of-technology-on-atlanta-tv-station-site/