[Olympus developers 162]: Kalliope performance in embedded system

[Νίκος Κατσαούνος]

Dear Brian,

I am running an Olympus-based dialog system in my desktop pc (Intel Core 2, 2.4GHz, 2GB ram). Everything is OK!
My problems start when my partners try to run it to the real embedded system of the project (VIA PICO-ITX board  --> 1GHz VIA C7, 1GB RAM).  TTS engine (KalliopeFlite) is too slow..That means that play-back of the prompts sounds too "sparsely"..(it sounds like a robot saying words syllable-by-syllable..)
I noticed that KalliopeFlite occupies a significant amount of cpu (in my desktop pc reaches up to 25-29%) even when a short-length prompt is played back.  I also noticed that approx. the same cpu amount is occupied even when KalliopeSAPI is used instead of KalliopeFlite. So, I suspect that cpu effort has to do with kalliope core-code and not the specific tts I may use..

What's your opinion? Am I "missing" something?
Has kalliope and/or flite indeed had such performance problems?
If so, is there any way to make kalliope faster? 
What are the minimum requirements (cpu / ram) in order to have satisfactory tts performance?  Are cpu 1GHZ  / ram 1 GB "enough" for having an acceptable performance?

Mention here that I use the "official" Olympus version (v. 2.2). 

Thanks in advance

Nikos Katsaounos
WCL / University of Patras - Greece
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.srv.cs.cmu.edu/pipermail/olympus-developers/attachments/20090923/2994b480/attachment.html