Hmmm.. Here's more voices; my presumption about the "American Female" was wrong; that's one of the Arctic voices. And they're much slower synthesizing than the default voice, at least.

MBROLA doesn't seem to be compiled for this platform; there's some voices over here with yet a different engine, but I've no idea if they're faster or slower...

There's lots of room to work on this now, so I guess I'll have to get g++ and start digging in.