The stress and intonation examples are what puts this on the far side of the uncanny valley. I've not heard computer speech like this, where the voice seems understand the idea underneath the words.
Perfect speech synthesis from unannotated text will require much more complex system. It will need database of common knowledge in some form, ability to relate text to that database, ability to retain and use this semantic information to guide synthesis. That is, it should be a step closer to artificial general intelligence.
Still, the quality is such that I'm only observing this in the spirit of scientific analysis. It's some good quality.
[0]https://deepmind.com/blog/wavenet-generative-model-raw-audio...
"WaveNet (van den Oord et al., 2016) is a powerful generative model of audio. It works well for TTS, but is slow due to its sample-level autoregressive nature. It also requires conditioning on linguistic features from an existing TTS frontend, and thus is not end-to-end: it only replaces the vocoder and acoustic model. Another recently-developed neural model is DeepVoice (Arik et al., 2017), which replaces every component in a typical TTS pipeline by a corresponding neural network. However, each component is independently trained, and it’s nontrivial to change the system to train in an end-to-end fashion."