speech recognition help

[Tim Chase]

> do u know which one
> 1. cmu sphinx
> 2. natural speaking
> 3. windows sapi
> 
> is best ( in accuray and speed ) for predefined vocabulary.. and worth for
> learning as well.?

For a pre-defined vocabulary, they should all be pretty good.  In 
general (for non-predefined vocabularies), I've heard that NS 
beats both Sphinx and SAPI in terms of recognition quality.  I 
haven't seen any comparisons between Sphinx and SAPI.

Other considerations may also come into play:

  - do you expect all your users to have a copy of NS?  Then you 
might have to look into something other than NS if they won't. 
Or perhaps NS has a developers' toolkit that allows for 
redistribution when built into your app.

  - do you expect all your users to run Windows (as I type this 
from my Linux box...)?  If not, then SAPI won't be available to 
your Mac/Linux/BSD/BeOS/Amiga/whatever users.

  - do you need TTS (text-to-speech) capabilities as well?  SAPI 
includes them.  (there are other TTS libraries available such as 
mbrola/festival that are available for this if needed)

  - how easy is it to bind to Python?  Sphinx has Python bindings 
detailed on the website.  I can't speak for NS or SAPI.

  - how much time/money do you have to invest?  Sphinx is 
open-source and free.  SAPI is free-ish for developers, but I 
don't know if Microsoft incurs a licensing cost for distributed 
apps or other redistribution restrictions.  I can't speak for 
NS-for-developers.

Just a few thoughts,

-tkc