I worked at Nuance Communications for about a year doing voice application development. I'm not familiar with Sphynx, but I did talk to a number of people at t

I worked at Nuance Communications for about a year doing voice application development. I'm not familiar with Sphynx, but I did talk to a number of people at the company about it (this was 2004-2005). From what I gathered it would be pretty difficult to get speech synthesis up to their level. A single "voice" will be generated by taking tens of hours of audio and using algorithms to splice them together based on the text. The only significant monetary constraint is going to be if you want to have a real voice talent doing your recording in a studio, but I wouldn't try to tackle the technical issues without a subject matter expert