VoxForge
VoxForge
Hi all,
I know that my question can be a bit out of line in this forum, but I would like to do an isolated word speaker in/dependent recognition system with a quite small vocabulary on a Windows mobile/CE system.
Does anyone know which engine can I use? I think PocketSphinx is the more suitable but I do not know if it is supported on these Windows versions. What about HTK??
Thanks, best regards!!
--- (Edited on 11/14/2007 6:42 am [GMT-0600] by coriscow) ---
Hi coriscow,
PocketSphinx is probably your best bet from an Open Source perspective.
HTK provides the source code, but you cannot distribute the source or binaries. However you can distribute models created with the HTK toolkit. Julius runs on HTK acoustic models ... but I am not sure how well it might run on Windows CE or Mobile. You might give it a try if you are familiar with HTK.
Have you looked at Speech API (SAPI) technology for Windows Embedded CE or Voice Command for Windows Mobile? These are not open source, but might fit your needs.
Good luck!
Ken
--- (Edited on 11/14/2007 1:41 pm [GMT-0500] by kmaclean) ---
--- (Edited on 11/14/2007 12:54 pm [GMT-0600] by nsh) ---
Hi all,
thanks for those fast answers. Ken, as you suggest, my first choice was indeed PocketSphinx but due to the lack of good documentation (I am to read SphinxII's) I searched for some others.
About HTK I think the main problem is to migrate to fixed point arithmetic and the fact that I will not be able to distribute the decoder along with program.
Concerning Julius, I am not confident too that it will run properly on such limited platform.
I have come up with another idea: Matlab is able to generate COM for .NET and that would be a really good approach due to Matlab and Voicebox libraries and docs. I am a newbie in that of ASR but I do know Matlab, so that would be an easier start point.
best regards!
Corisco
--- (Edited on 11/15/2007 3:20 am [GMT-0600] by coriscow) ---