However, an advantage of dealing with visual input is that so much of human language deals with visuospatial metaphors. So, as compared to a soley-sound-focused system, a vision-savvy AGI perceptual subsystem is substantially more useful for helping an AGI system learn human language...
-- Ben > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] > Behalf Of J. Andrew Rogers > Sent: Thursday, October 14, 2004 6:59 PM > To: [EMAIL PROTECTED] > Subject: Re: [agi] Sensory Front-End Design Specs > > > > > Last time I checked, speech recognition still performed badly. > > I guess the audio "resolution" has to be drastically reduced > > just like in video. > > > The problem with speech recognition has nothing to do with the quality > of the audio source, but the algorithms trying to analyze the data > given. The algorithms currently used for handling speech patterns are > brittle, same as any other common current "machine intelligence" > technology. A good AGI design won't have this problem. Humans have no > problem understanding speech that has been run through the same kind of > signal processing I described above; the point of the low-resolution is > so that the designs do not have to worry about the volume of data. In > other words, I was building generic data flow reduction into the design > since it needs to be done anyway and little will be lost by it. Raw Red > Book is overkill. > > You can do "interesting" and useful things with AGI that are less > complex to solve and easier on the hardware resources when dealing with > sound rather than video. To me, it would seem like a more logical > progression. You can work on video after you have the simpler problem > of audio figured out. If you can't solve audio, I very much doubt you > will solve video. > > > j. andrew rogers > > > ------- > To unsubscribe, change your address, or temporarily deactivate > your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] > > ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
