Ask HN: State of offline speech synthesis and software recommendations

Seems offline text to speech engines have all stopped development around 5 years ago as everyone moved to deep learning solutions with cloud tts, pay per character models.

https://en.wikipedia.org/wiki/Comparison_of_speech_synthesizers#Technical_voice_details

I read that Google will be offlining machine learning tts onto neural cores, but that still leaves desktop market unaddressed - they're doing great work in accessibility in general, but I can't do productie work on a phone/tablet. Wondering if there are any other developments in this area, especially with reference to desktop hardware.

In terms of recommendations: looking for some sort of offline speech engine that can handle multi-lingual seamlessly i.e. English/Chinese.

Current setup: IVONA (stopped development after being acquired by Amazon) which sounds pretty natural, but skips non foreign characters. It works in Balabolka which has regular expression support to filter out cruft. Great workflow when TTS is triggered by clipboard.

Previously I used chrome Read Aloud extension hooked up to Google Wavenet TTS (expensive $16 per million character) and Amazon Polly (cheap but unreliable). Also supports Google Chinese voices which pronounces Chinese fine but switch to Chinglish when using English characters. Limited to chrome. On Android there's Voice Aloud + discontinued IVONA apks. Pocket uses Amazon Polly but can't parse all articles. Joey has TTS support for Reddit. No HN apps with TTS support that I can find, so trying to learn accessibility features but having every element read out loud is... subpar.