3 ms·

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.

jjulius 4y ago
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.
- Louistiti 4y ago
  You can choose what voice solutions you want to use. So you can choose the ones that run offline (which are the default ones).
RussianCow 4y ago
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.
- fxtentacle 4y ago
  Currently, they are like 5x the error rate, which is significantly worse.
fxtentacle 4y ago
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)
prmoustache 4y ago
That doesn't make it right to lie.
- Louistiti 4y ago
  Where is the lie here? You are free to pick voice solutions, and the default ones are the ones that run offline. Your call.
heliostatic 4y ago
Mozilla DeepSpeech is not bad, and supported by Mycroft: https://mycroft-ai.gitbook.io/docs/using-mycroft-ai/customizations/stt-engine#mozilla-deepspeech https://mycroft-ai.gitbook.io/docs/using-mycroft-ai/customiz... You can contribute to improving it here, too: https://commonvoice.mozilla.org/en https://commonvoice.mozilla.org/en
- spullara 4y ago
  It is terrible. Try it against any of the cloud providers. The problem is the lack of data.
- Louistiti 4y ago
  Leon was using Mozilla DeepSpeech before. But then we moved to Coqui STT instead.

jjulius 4y ago

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.

Louistiti 4y ago
You can choose what voice solutions you want to use. So you can choose the ones that run offline (which are the default ones).

RussianCow 4y ago

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.

fxtentacle 4y ago
Currently, they are like 5x the error rate, which is significantly worse.

fxtentacle 4y ago

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)

prmoustache 4y ago

That doesn't make it right to lie.

Louistiti 4y ago
Where is the lie here? You are free to pick voice solutions, and the default ones are the ones that run offline. Your call.

heliostatic 4y ago

Mozilla DeepSpeech is not bad, and supported by Mycroft: https://mycroft-ai.gitbook.io/docs/using-mycroft-ai/customizations/stt-engine#mozilla-deepspeech https://mycroft-ai.gitbook.io/docs/using-mycroft-ai/customiz... You can contribute to improving it here, too: https://commonvoice.mozilla.org/en https://commonvoice.mozilla.org/en

spullara 4y ago
It is terrible. Try it against any of the cloud providers. The problem is the lack of data.
Louistiti 4y ago
Leon was using Mozilla DeepSpeech before. But then we moved to Coqui STT instead.