[maemo-developers] One question about speech2text poor performance

From: David Huggins-Daines dhuggins at cs.cmu.edu
Date: Thu Jan 17 05:17:17 EET 2008
Marco Solari wrote:
> Hi!
> I have a question about the voice recognition problem on the Nokia platform,
> hoping it's not too off-topic...
> I see on this platform some voice recognition tool is available, but it's
> extremely resource demanding, slow and performs with quite a low success
> percentage...
Sorry to flood everybody with messages here, but another reason that 
PocketSphinx performs poorly is that the default language model (which 
is the part of the recognizer that decides whether a string of words is 
a likely sentence or not) is really not good at all.  If you have a 
small vocabulary in mind I encourage you to build your own language 
model, following these instructions:

http://www.speech.cs.cmu.edu/cmusphinx/moinmoin/LanguageModelHowto


More information about the maemo-developers mailing list