synthesise speech waveforms instead of using pre-recorded segments

[De7vID]

This feature request will involve a long-term rewrite of the entire engine.

For an example of what's being suggested, see:
https://cloudplatform.googleblog.com/2018/03/introducing-Cloud-Text-to-Speech-powered-by-Deepmind-WaveNet-technology.html

[De7vID]

An advantage of the concatenation approach is that several other projects have adopted it, so it works across multiple platforms, e.g., https://hol.kag.org/

So probably both systems will have to be maintained.