Hi,

if you are interested in high-quality text-to-speech, systems based on 
recorded speech instead of formant synthesis may be worth a look. We 
have very good experience with Mary from DFKI (http://mary.dfki.de/) 
that produces very natural speech. The disadvantage of these systems is 
that they require quite a lot of memory (some hundred MB).

Mary is open-source software, we wrote a shallow ROS wrapper [1] that 
makes it read everything published on a topic and allows to select 
different voices.

cheers
Moritz

[1] http://code.in.tum.de/pubsvn/knowrob/tags/latest/3rdparty/mary_tts/

-- 
Dipl.-Ing. Moritz Tenorth      | tenorth@cs.tum.edu
Technische Universität München | Karlstr. 45
80333 München                  | Germany
ias.cs.tum.edu/people/tenorth  | Tel: +49 89 289 26912