I looked into this myself a few months ago. It should be fairly
straightforward; from what I can tell, two things need to be done to
get any of the several freely-available TTS (text to speech) engines
to speak Lojban:
1) Write some code that converts written text to phonetic
representation (diphones or phenomes, including stress and pause
markers). For many languages this is very challenging, but for
Lojban it should be trivial.
2) Record and process lots of voice samples containing all the
basic sounds of the language in all their possible 2-sound
combinations. (The combinations are necessary in order to get
the transitions.) This is the time-consuming part. It should
be done entirely by one speaker.
Alternatively, find a voice recorded for another language that
has all the same sounds as Lojban. English won't work (doesn't
have Lojban's 'x'). The resulting voice will sound a lot like
the language the voice was originally prepared for.
I would be interested in helping out on such a project. Specifically,
I'll handle the first part (the easy part) if somebody will get the
recordings together for the second part.