Making a complete synthesizer to autoswitch language in NVDA seemed a bit too much. I understand NVDA much better now and ported to language prediction and switching to a new plugin, that works with any speech-synthesizer. It’s a bit of a hack. It replaces the built in speak function from NVDA and modifies the speech-sequence before it’s given to the speech-synthesizer. Works great though. I tested it with IBM-TTS and onecore Windows. It does not work with Codefactory’s Vocalizer or Eloquence voices.
Here is the link to the initial release 0.0.1.