Voice Activated Controls

jeff moe at blagblagblag.org
Tue Feb 27 02:00:54 CET 2007

Dean Collins wrote:
> The answer is no, the neo processing power is too limited.

Or perhaps the answer is more like "maybe". ;)  I'm still holding my breath for 
pocketsphinx...   The page does say it works with StrongARM, which if I'm 
reading wikipedia correctly was just 206MHz.

> You need to start thinking bigger guys.
> Why does the processing need to occur on the handset itself? What about
> building a 'cloud' application where the application and processing
> occurs with only a sip or gsm connection to the handset?
> Check out http://www.voip-info.org/wiki/view/Tellme for an idea on what
> I am describing.

This is a nice idea to offload processing to another box and spit back the 
results. One big problem I have with the Tellme approach is that it isn't free 
as it's using some service provider. Blah for me there. That said, there's no 
reason the ideas behind it couldn't be implemented freely.

I've done speech recognition with sphinx and asterisk. It doesn't work too 
great--the main problem being that GSM is a pretty low quality sample for it to 
work with. You know how you can never understand anyone talking on a cell 
phone? ;)  Well, sphinx can't understand it too well either. But pocketsphinx 
(which I haven't tried) lists on their page "telephone-bandwidth models"--this 
looks promising.

For SIP, you need some sort of net connection, of course, and I think we can 
assume that that is coming. Then you can use alaw/ulaw instead of GSM and that 
will help recognition quite a bit...

-Jeff    /    #jebba

More information about the community mailing list