> I suspect this is using the closed caption subtitles to look for keywords,
> not doing any kind of automated voice-to-text conversion. Using closed
> captions would be much, much easier, and this doesn't exactly look like
> a high-budget operation. If so it doesn't have much relevance to the
> spy agencies and their Echelon program.
Sorry, should be doing more factchecking next time. I know of a software product, however, which does make automatic transcripts of audio streams using speaker-independant speech recognition, albeit perhaps not in realtime.