They might be interested in integrating Vosk, it's a speech-to-text engine that is just a shared library (.so file on Linux) and comes with API support for a variety of languages:

https://alphacephei.com/vosk/

https://github.com/alphacep/vosk-api

Still, I've found that the Big players have much better recognition models, and the post-processing that I assume they do (grammatical, maybe syntactical inferences that improve the end result) are probably much more powerful too.

Yes Vosk will definitely be part of Leon when the focus will be on implementing new voice solutions.

Please do know that new offline TTS/STT solutions are coming up.

https://trello.com/c/lEZG0VYS/259-%E2%9C%A8-100-implement-ne... and https://trello.com/c/nxEstkvo/260-%E2%9C%A8-100-implement-ne...

There aren't any good speech-to-text models that are open source. If you think there is one, please reply with a link. The cloud ones are far superior.

I fully agree and I would love to change that. I mean my company already funded work in that direction... but I sadly predict that we won't have good open source real-time speech recognition anytime soon.

My napkin calculation is that you need about $100k for each attempt at training a Conformer-Transducer. There's a pre-trained NVIDIA model but it appears to have a bad choice of hyperparameters and performance is much worse than what one would expect based on research literature and I believe you're not allowed to execute it on non-NVIDIA hardware.

A skilled team will maybe need 5-10 attempts for discovering a good set of hyperparameters. So the price to create the AI model will likely be around $1 mio. But if you have such large expenses, you have to plan things as a business venture. And that means an open source release is highly unlikely.

(unless, of course, someone like stability.ai is happy to bankroll 200 A100 GPUs for a few months each per target language. In that case, please contact me)

Mozilla DeepSpeech is not bad, and supported by Mycroft: https://mycroft-ai.gitbook.io/docs/using-mycroft-ai/customiz...

You can contribute to improving it here, too: https://commonvoice.mozilla.org/en

It is terrible. Try it against any of the cloud providers. The problem is the lack of data.

Leon was using Mozilla DeepSpeech before. But then we moved to Coqui STT instead.

Right, and that's fine. The point is that if that's the case, it's incredibly disingenuous to say that you are in control of your own data if you use Leon.

You can choose what voice solutions you want to use. So you can choose the ones that run offline (which are the default ones).

That doesn't make it right to lie.

Where is the lie here? You are free to pick voice solutions, and the default ones are the ones that run offline. Your call.

I don’t think the open source ones need to be superior to the cloud ones, or even as good. If they come close enough for the most common, let’s say, 80% of use cases, that’s good enough for many people.

Currently, they are like 5x the error rate, which is significantly worse.

I don't find this such a big deal. Don't use the speech to text and just write messages to it instead. You still have way more control of your data using Leon compared to the commercial funded alternatives.