New software from Fast-Talk lets you index and search audio voice-streams
much faster than in the past

     http://www.technologyreview.com/articles/print_version/wo_harney043003.asp

This has interesting possibilities for locating material in voice
recording, based on phonetic query of audio, without transcript.

Their web-page:

      http://www.fast-talk.com/

"Fast-Talk's software doesn't require speech-to-text conversion. With our
unique phonetic technology, you can retrieve any word,
name or phrase from voice data, regardless of speaker or dialect, with up
to 98 percent precision. We enable you to search your
digital assets 100,000 times faster than real time, or 30 hours in less
than one second. The result: Fast-Talk significantly improves
the way people search and access valuable audio and voice data."

It requires typing the phonetic representation of what you seek. It has no
visual component, given access to the
voice recording -- presuming that what is returned from the query is the
audio stream where that phonetic pattern is found.
It does not permit proximity searching.

I wonder if it might be able to accept a vocal representation of the
phonetics -- the phrase sought? -- would need to
normalize pace and possibly frequency; as those are not available in the
phonetic representation.

Regards/Harvey Bingham