[MV] cs ilisten, interface issues for correcting

[Chuck Rogers]

Elizabeth (and everyone else):

Please see my replies below:

> []tell the machine during learning that I misspoke a word

This actually wouldn't be very helpful. The software has some degree of 
checks and balances built-in. While reading the stories, the software 
is actually listening to the way you pronounce phonemes, not words. 
Those phonemes are repeated in various words throughout the training 
process, so you would gain very little (if any) benefit if you were to 
correct just one word here and there during the training process.

> []ask the machine what it thinks a proper pronunciation of a word is.

Unfortunately, this wouldn't work very well either. Apple's 
Text-To-Speech voices are very understandable, but they sometimes don't 
pronounce words the way you would pronounce them. They have their own 
unique "accent," in other words. But if you know a little about the 
phonetic alphabet, you can see the way iListen thinks things should be 
pronounced as you add new words. We do provide a guide to the phonetic 
alphabet in the 1.6 addendum, which is included inside the 
Documentation folder, located inside your iListen folder.

> []ask the machine all the ways it has heard me speak a particular word

Again, this wouldn't be all that practical. It would require an immense 
amount of storage space on your hard drive (to store all the audio 
samples). The way the software is built, it is not even possible, since 
the software does not listen to words, per se, but phonemes, and then 
phrases for context.

> []confirm-command mode; command mode I find nearly unusable because to
> interpret "mouse click" as "file close" (which happened repeatedly 
> since my
> buying mouse-clicks-extra) is horrible!  There needs to be a very easy 
> way to
> prevent bad things from occuring.  Speech is already slower; an extra 
> second
> for "yes/no" (which I think usually aren't misrecognized) would really 
> be in
> order.

Some people have problems with some commands more than others. This is 
especially true of shorter words. Try changing the name of the command. 
For instance, instead of "mouse click," change it to "click mouse" or 
something else. You can rename any command. If you need help for 
knowing how to do this, instructions are in the User's Guide, or you 
can email <questions at macspeech.com>.

> []show some kind of CPU thermometer that explains why words get eaten. 
>  Does it
> stop dictating because it hung?  Or I spoke too fast?  Not knowing why 
> it
> suddenly stops listening is really hard.  I have a hard time getting 
> back to
> where I want to be.

This would be really difficult to do within iListen. There are several 
programs out there that can show you in real time what the CPU usage 
is, but to ascertain why a certain word is not recognized or why there 
is a longer pause this time then last would be virtually impossible. 
All sorts of factors come into play. The application you are using when 
dictating may be doing something (such as the Fast Save process in 
Word), or your Mail program may be checking email in the background, or 
a server mounted may have become disconnected, or perhaps a web page is 
reloading. Any of these things could cause the program to "hang" 
momentarily.

Also, if you pause and wait for it to catch up, iListen may pause 
because it doesn't have enough information with which to work. For best 
results, don't look at the screen while you are dictating. There is a 
preference setting where you can adjust the program to recognize faster 
(which is fine for some people who have great accuracy), or more 
accurate (which would be slower, but work harder to be accurate). If 
you need assistance setting this preference, please email 
<questions at macspeech.com>.

I know these answers aren't what you want to hear, but computers just 
aren't powerful enough yet to develop the sort of software that would 
be necessary to address most of these issues.

Chuck Rogers, Chief Evangelist
MacSpeech, Inc.