Fast-Talk Gets Right to the Point

Audio processing engine lets listeners sift through recordings to find relevant terms-by sounding them out.

While carrying out testing for the virtual meeting eVal extravaganza that appears in the July 1 issue of eWEEK, one of the product features that most appealed to our Corporate Partner judges was the capacity for recording meetings—presentation slides, whiteboard doodles, audio comments and all.

It makes great sense—if youve brought together your employees and partners for a meeting of the minds, why shouldnt those not in attendance also benefit from the productivity-packed proceedings?

It sounds great until its time for those absent from the meeting to plow through hours of recordings, throughout which the relevant information lies scattered and not easily accessible. Hey, if someone didnt have the time to attend a meeting initially, they may not have the time to sit through a canned version of it, either.

Recently, Ive been testing a technology from Fast-Talk Communications Inc. that could help companies cut through the clutter of recorded materials to pick out the parts they need more quickly than a full viewing could permit.

The system takes .WAV audio files, breaks them up into their individual phonemes, and generates a large text file where these phonemes are indexed for text searching.

Software algorithms convert search terms to their constituent phonemes, enabling users to locate specific words in these recordings. Alternatively, I could generate a phoneme index while recording live or streaming content.

I used a technology demo version of Fast-Talks Phonetic Processing Engine to process and search through the audio recording of our most recent IT Roundtable discussion—I was too busy writing that virtual meeting eVal to attend in person, honest.

I was impressed by how quickly and how flexibly the Fast-Talk technology worked, particularly with odd product names and other proper nouns, and herein lies one of the major strengths of this approach compared to a strategy like speech to text.

A speech-to-text engine relies heavily on its dictionary. When one of these engines turns a spoken word into a written one, it must judge which of the terms in its internal lexicon best matches the word its heard. Computers make lousy judges, and many terms relevant to our lives and work wont appear in any speech-to-text lexicon. By focusing on sounds rather than words, the Fast-Talk engine sets the bar on the sort of judgments it attempts much lower. As a result, no matter how oddly a company has named itself in a run for one of the few remaining combinations of syllables that havent yet been claimed as a .com Web address, Fast-Talks engine can help you find it—you need only spell the name the way it sounds, and the software will locate the word.

This technology would make an excellent addition to any of the virtual meeting products we tested, or for any application that works to make multimedia content more accessible. For more information, check out

How do you manage your multimedia resources? Drop me a line at