Mandarin speaker here. I'm more impressed that it was able to reconstruct the sentences properly (where traditional translation tools typically fail). The output was fine. Obviously doesn't sound like a human speaker, but the tones are correct.

Yea, it sounds a bit mechanical. With that said, based on the priming he's given with the talk on waveform, I'm guess they are simply breaking the English speech into the waveform with corresponding frequency and simply mapping that over to the Chinese counterpart.

It would be even cooler if they created a distribution of possible sound frequency for each syllables in both English and Chinese, and determined where in the distribution his speech pattern lies, and transfer the "ranking" in the distribution. Hence you get a subjective transformation instead of a objective one. :)