We've all heard the output of text-to-speech systems, and for anything but very short phrases, it sounds very machine-like. The ultimate goal of speech synthesis systems is to pass a Turing test of hearing. Clearly, the state of the art in text-to-speech has much to improve.
However, speech synthesis isn't restricted to just text-to-speech systems, and I'm wondering if other approaches have been tried with better success. Are there better ways to attack the problem?