Re: What did that thread indicate?


Traveler wrote:
> On 26 Sep 2005 19:42:44 GMT, curt@xxxxxxxx (Curt Welch) wrote:
>
> >SO, just by adding feedback loops to a network like mine, I know it's got
> >the basic system needed to the "cocktail party" attention focus task. But
> >I don't know if it can actually do it.
>
> I doubt very much that it can. The cocktail party problem require a
> system with the ability to anticipate intervals between events. It is
> not that much more different than a dog chasing a rabbit. If the
> rabbit suddenly disappears behind a hedge, the dog will not stop. It
> will correctly predict that the rabbit will reappear on the other side
> of the hedge after a short interval. The anticipatory mechanism of the
> dog can instantly predict this interval.


Hmmm. I wonder, regards the cocktail party problem, whether anyone [ie,
some psychologist] ever did studies correlating success in hearing
against visual observations of lip movements of the speaker. IOW,
correlating the timing of lip movements with the temporal
characteristics of the stream of words might be the key to improving
intelligibility in this situation. Multi-sensory temporal integration.
In contrast, all of the background clutter will not have the same
temporal correlation.

Also, regards the comment about prediction from a couple of msgs back,
one thing I've noticed is it's "much" easier to understand certain rock
singers/songs if you can read the lyrics as you go along. EG, recently
I was listening to Dark Side of the Moon album by Pink Floyd, the
british rock group. Given the vowel shift and different
syllabic-emphasis compared to typical american english, it's not allows
easy to understand what they are saying, let alone singing with loud
background instrument accompaniment. IOW, if you know what they're
going to say, then it's much easier to make out the actual words. This
may seem like an obvious observation, but it also implies something
about the deep neural mechanisms.

.

Quantcast