It’s the new ‘new thing’ and it’s coming – fast.
by Pete Fasciano, Executive Director 04/30/2023
Artificial Intelligence. It’s the current high-techie buzz – a fancy polysyllabic term. Fake Info? Says the same thing – just not in buzzy breathy brochure language.
When the ChatGPT Open A.I. interface surfaced recently, I challenged it with a simple request – Write a one-page essay on Franklin, MA history. The result started off reasonably enough, but half-way along, the rising degree of fable over fact was clear. The A.I. was fabricating – faking it, instead of making it. Open-ended A.I. is a big high-tech ‘boil the ocean’ challenge. It’s wikkid hahd. I’m also mistrusting of any endeavor where the experts state. “We actually don’t know how or why it works.”
That said, there is a subset of A.I. that does work quite well. It’s G.I. Guided Inquiry. This subset of A.I. software development rests on a simple premise. If you code the right questions and goals in software, you can get the right answers. The computer can produce useful outcomes for specific tasks through training – exercising the specified questions and goals repeatedly, and through lots of trial and error – ultimately, to produce a beneficial, dependable, emergent behavior.
This week I tested Cleanvoice, a smart audio editing system that automatically removes ‘filler sounds, stutters, hesitations and the like from speech recordings. This is a task that is well suited to G.I. designed software, and the company that offers this intelligent task-specific service appears to have gotten it right.
I uploaded a test audio file that was about 2 minutes long. Only 10 seconds later it returned a processed file that was 7 seconds shorter.
The accompanying report stated that Cleanvoice removed:
12 filler sounds, 2 stutters, 7 seconds of dead air, and a mouth sound.
I listened to their processed audio clip. I was very impressed. I was also humbled. Were I to attempt manually editing and cleaning up all those momentary blips in my 2-minute sample, it would take perhaps 30-40 minutes. Tidying up a full hour of audio to this level of error-free speech would take a long day.
When spending a little money can save a lot of time – that’s a real intelligent move.
Thanks for listening to 102.9 wfpr●fm.
And – as always – thanks for watching.
Get this week's program guide for Franklin.TV and Franklin Public Radio (wfpr.fm) online http://franklin.tv/programguide.pdf
 |
| Franklin.TV and Franklin Public Radio (wfpr.fm) |
No comments:
Post a Comment