Author: Ricardo Gibert
Date: 19:46:59 10/01/99
Go up one level in this thread
On October 01, 1999 at 21:31:04, Gareth McCaughan wrote: >[Ratko Tomic:] >> I think the key >>underplayed point in this article was: >> >> In benchmark testing using just a few spoken words, USC's Berger-Liaw >> Neural Network Speaker Independent Speech Recognition System not >> only bested all existing computer speech recognition systems but >> outperformed the keenest human ears. >> >>i.e. the phrase "few spoken words." That sheds the light how it was >>possible. The net had a handful of choices to decide among the words >>amid noise. > >Yes, I agree. It's notable, for instance, that most of us are >especially good at picking out our own names amid noise and >distractions... this result seems to be comparable to that. > >It's still quite impressive; picking out anything with 1000 times >as much noise as signal seems like quite an achievement. But it >doesn't sound to me as if it's going to change the world. "Remarkably, USC's neural net system uses an architecture consisting of just 11 neurons connected by a mere 30 links." This is very interesting. Previous attempts used 1000 neurons with 10,000 interconnections, which did not do nearly as well. Tomic appears to be right about the limited vocabulary. Even so the demonstration offered at http://www.usc.edu/ext-relations/news_service/real/real_video.html is still impressive. It outperforms humans not by a small margin, but by an overwhelming margin.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.