Re: [Accessibility] resident evil

[Eric S. Johansson]

 On 7/31/2010 5:56 AM, Richard Stallman wrote:
>      I disagree that we should work to make a functional replacement of
>      Naturally Speaking.  Instead, we should work towards competitive
>      free-software continuous large-vocabulary speech recognition, in
>      parallel with developing/enhancing speech-enabled application
>      interfaces.  However, let's not duplicate Naturally Speaking's
>      limitations, and let's keep it focused on the needs of people who
>      can't type.
>
> When I say a "free replacement", that doesn't mean we should aim to
> preserve any limitations or misfeatures.  I mean a replacement that
> can do the same job.

No one has for preserving a limitation or misfeature. Every description I've 
seen of using NaturallySpeaking is as a baseline, a reference point, something 
to supersede.

> However, there is no reason to focus narrowly on the needs of people
> who can't type.  The more people the program is usable by, the better,
> and that will mean more contributions of a kind that are directly
> useful.  That ought to benefit all the users.

If we have focused on the needs of people who can type it's because people who 
can't type have been ignored by almost every other disability project out there. 
It's almost all focused on the blind or extremely disabled.

> People said we need a corpus of speech samples.  I read that there was
> a project to make a free one, and it needs help.  We should be in a
> position to help people contribute.  Can anyone find the name of the
> project?

yes I did.  the product has gotten better organized since I last looked at  it 
and it makes me feel hopeful that you might get a useful corpus out of it. 
http://www.voxforge.org <http://www.voxforge.org/>

I am curious as to why the limitation of 8 kHz bandwidth limitation. If you go 
12 to 16, you get more  information/features you can use in the recognition process.

Also, on understanding the speech recognition process, I highly recommend that 
you walk over and speak with Jim glass. He's not that far from your office in 
that topsy-turvy building and if you don't know him, Gerry J. S.  knows him and 
can provide an introduction. As I hinted at earlier, Jim may be able to help you 
acquire a horde of grad students with lots of knowledge about speech recognition 
systems in addition to a body of existing work at MIT that may be made free for 
this kind of project. He can also help calibrate you on the level of effort 
necessary. They may have a corpus available for tell you why they didn't try to 
gather one.

Jim is good people, really smart, and not shy about (politely) telling you where 
you may need to acquire more knowledge. He did that favor for me with 
programming by voice techniques and I'm grateful to him for that.