> I also continue to find it amusing that we can build a person like that and sell it as commercial tech, but we still have to record people talking. (Though TTS has taken an interesting turn lately, after years of not much.)

I suspect part of the problem is that dialog in games is still largely "static"; if the writing is pre-canned then it does not make that much sense to try to develop advanced tts to act it out. The situation will become interesting if we manage to produce sufficiently dynamic dialog system where it is not feasible to use pre-recorded voice acting anymore.