[MV] Where are we now?
s.brostoff at cs.ucl.ac.uk
Mon Nov 8 17:03:13 PST 2004
How are you guys going to choose what to enhance? Low hanging fruit vs.
highest productivity gains?
I've not studied it, but I'd be amazed if the biggest improvement in
performance could be achieved by anything other than improving the
correction process, and that's speaking as a Dragon user about Dragon.
Have you guys (or anyone else) sat down and worked out how long/much
effort/detraction-from-the-task/ etc. each step of document production with
SR takes people? Fascinating stuff!
Regards and appreciation,
From: macvoice-bounces at listserver.themacintoshguy.com
[mailto:macvoice-bounces at listserver.themacintoshguy.com] On Behalf Of Chuck
Sent: 08 November 2004 17:27
To: A place to discuss speech recognition on Macintosh.
Subject: Re: [MV] Where are we now?
Brian (and everyone else):
Your question is sort of a skewed "horse or cart" question. The answer
is "both" in the sense that the cart can only go as fast as the horse
(or horses) pulling it, but the cart must also be built to handle both
the speed at which the horse(s) will take it, as well as the roads on
which it will travel.
As computers get faster, engines will become more capable, and software
will be written to take advantage of increased computing power. In the
case of iListen, we still have work to do on the cart (in terms of
enhancements to the application) that we think are more worth the
results we will see than building a new cart for the faster horse(s).
To refer back to my original message, the horses are not significantly
faster than they were two years ago in terms of speech recognition.
Will you get better accuracy from a G5 than a G4? In the vast majority
of cases, yes (as long as you keep in mind there are other factors
besides just processing speed that affect recognition). But we aren't
at the point yet that we should forgo building enhancements to the
existing app and turn our attentions to a new speech engine.
We are at the point where we are looking at what that engine might be,
and starting to scope out how we might migrate to it over time. Once
that is in place, we will have a better handle on when we start writing
code for a new engine.
So to answer your question directly: the major roadblock towards better
accuracy is indeed processing power. But a 4Ghz G5 (or even G6 or G7)
probably won't provide significant boost there either. You are really
looking at computers 5 or 10 years from now before an engine can really
be thought of differently in terms of how it processes information. The
existing engines will continue to be improved over this time, but such
changes will be incremental, not monumental.
In terms of ease of use, there is a lot we can do on our end now,
without the need for faster processors - and some of the enhancements
to which I have been referring will certainly be in that direction.
Chuck Rogers, Chief Evangelist
On Nov 8, 2004, at 11:01 AM, Brian Braunschweiger wrote:
> At 5:29 PM -0600 11/06/2004, Chuck Rogers wrote:
>> But the time will come soon when we will need to replace the engine
>> that works underneath iListen. Are we working on it? Let's just say
>> we are starting to carefully weigh our options. As to when a version
>> of iListen will be available with a new speech engine, that is
>> anyone's guess - but I would say we are at least a year or more away.
>> We still have quite a few things we would like to introduce with the
>> current engine first.
> A "just curious" question I have had:
> Is the major road block to better accuracy and ease of use more on the
> side of the speech engine or more on the side of the user's computer's
> horsepower? Should I be cheering on a 4ghz Mac or for Phillips and
> iListen programmers to make major advances?
> Brian Braunschweiger
> "Worry does not empty tomorrow of its sorrow. It empties today of
> its strength."
> - Corrie Ten Boom
> MacVoice mailing list
> MacVoice at listserver.themacintoshguy.com
MacVoice mailing list
MacVoice at listserver.themacintoshguy.com
More information about the MacVoice