Crow Translate

[Kevin Kofler]

Hennadii Chernyshchyk wrote:
> Regardless of the warning. I think that the only translation engine that
> fully respects privacy is LibreTranslate.
> But it doesn't support automatic language detection and can't do TTS. It's
> also disabled on the main Mozhi instance (instances can disable specific
> engines if they want).
> So I can add a warning, but there is not much to choose from :(
> Do you still think it's a good idea?

The issue is inherently that your application is relying on a web service 
(Mozhi) to do the translations. That service then either forwards the 
requests to yet another (typically proprietary) web service (e.g., Google 
Translate) or to something local to the Mozhi server instance (e.g., a 
LibreTranslate instance running on the same server as the Mozhi instance). 
In the first case, there are two services that see your data, in the second 
case, there is still one service that sees your data and that you have to 
trust.

The only thing that is guaranteed to be privacy-friendly is to run (either 
as a separate process or embedded as a library) a local instance of 
something like LibreTranslate (or Argos Translate directly) or Bergamot (or 
Marian directly) on the same device as your application. And libmozhi is 
probably not the library you would want to use for that (though technically 
I guess you could theoretically run and point libmozhi to a local Mozhi 
instance and in turn point that to a local LibreTranslate instance).

As for:

> But it doesn't support automatic language detection and can't do TTS.

there are separate libraries for that.

For automatic language detection, see, e.g., cld2 or cld3, or one of several 
Python or Perl modules.

For TTS, see eSpeak NG (or its library libespeakng) if you want something 
simple and compact that supports many languages out of the box, or one of 
the local modern AI-based TTS tools (Mozilla TTS, OpenAI Whisper, etc.) if 
you want something that sounds better.

And, while you did not ask for it, for STT, there is the old PocketSphinx, 
the newer Mozilla DeepSpeech, etc.

The Free Software implementations that you can run locally all focus on 
doing one thing well, not on offering a complete solution like the Google 
Translate web service does. Integrating that all into a complete solution is 
then your job as the application developer.

        Kevin Kofler