Better text-to-speech support

[johnfactotum]

Currently text-to-speech (TTS) support is pretty bad. It should be improved.

Specifically, it should

• Use SpeechSynthesis API

• "Polyfill" SpeechSynthesis API (which isn't supported by WebKitGTK) by posting messages to the GJS process, which will call speech-dispatcher.

• Parse and handle assisting markups

  • SSML
  • PLS
  • CSS speech?

• Segment text, with Intl.Segmenter

• Highlight currently spoken word/sentence

  There is a "boundary" event on SpeechSynthesisUtterance. AFAICT there is no such event in speech-dispatcher. Instead we need to add SSML marks and read the INDEX_MARK event notification from speech-dispatcher. This is how Calibre does it, judging by its source code. Also, not all synthesizers supports this (or even SSML).

• UI for controlling playback and configuring voice, speed, etc., and make the features more discoverable (see Make text-to-speech feature more discoverable #672)

[ghost]

I've also come across a better sounding (IMO) TTS system than festival called mimic https://github.com/MycroftAI/mimic1

You can hear a samples comparing it to festival and espeak-ng here https://linuxreviews.org/Mimic