Skip to main content

Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
TagsGame Engines

GameTranslate

In-game translator at your disposal · By Godnoken

[Possible Workaround] RegEx support?

A topic by IDLT created 37 days ago Views: 99 Replies: 6
Viewing posts 1 to 4
(1 edit)

Just a slight suggestion while the models increase in accuracy

Hey Godnoken, I just bought your software. Love the Rust. Love the opportunity to collaborate and make it even better. I play games professionally and I need a tool that is reliable, fast, lightweight and not invasive (thanks for the offline option).

I, however, have noticed that some translations are very off, but consistently! Meaning that the model believes with high confidence that this is the correct translation. Therefore, if it was possible to have a RegEx find and replace for these erroneous translations, this issue would go away every single time.

While I know that the ideal solution is having better models. But since you say that these are not going to increase in accuracy that fast, What do you think about adding such a functionality?

Best Regards, IDLT

(1 edit)

After playing around with it for a bit, I do think the best solution would be to improve the Translation models.

I use an Open Source voice dictation app that uses the FasterWhisper and the Ctranslate2 library, which is wicked fast (less than 500ms to transcribe and translate more than 30secs of voice recording) on a RTX4070. I’m saying this because I saw that you were unsure of how heavy models would play. I’m confident there must be a way to try out CUDA compatible implementations to see how a “heavier” model would run.

Developer

No GPU support for translation yet, but as of 0.4.4, RapidOCR can now be run with the GPU. Hopefully that increases the speed somewhat on your PC!

(12 edits)

Fortunately, or unfortunately, the slowdown isn’t OCR (which is already impressively fast); it’s the DeepL translation that’s killing responsiveness. I know you’re focused on OCR perf, but from my end, that’s not the bottleneck.

Examples (0.4.4)


⬆️ The translation Online with DeepL
 
 
 
 


⬆️ The translation Offline with Internal Engine
 
 

This is one of the many reasons why an internal model is the preferred solution in my use case.

Developer

Hi there mate,

Thank you so much for purchasing my app and giving me all these suggestions & feedback, invaluable!

I have actually had Regex options in the pipeline since Early 2023 (took a 1.5-year complete break from the app), but never implemented it. It does certainly have its use for this app, but the reason it is still not implemented is just that I realized that a very small percentage of all users would know how to write regex lines themselves and most people would not even bother even if they knew how to do it.

It would certainly not require too much dev time to implement this, but I also think I have much more important things to address at the moment, unfortunately.

Regarding new, heavier models - Using CUDA or other similar tech is definitely possible and 100% worth looking into, which, in fact, I have actually already done. In that case, it was in regard to speeding up OCR. The attempt came to a halt due to some issues I had with the implementation. Can't remember off the bat what it was. But it is certainly something I will look into again in the future.

The tech we use for CPU translations is already pretty damn wicked fast, even on older hardware. Although the models themselves could of course do with much better quality output. The thing is, I am currently trying to focus on tech that can support as many devices as possible, as more paying customers means more development time & increased quality of the app, as well as bigger organic growth. Niche or top end hardware focus is not a priority, but be assured that these will be addressed too, in time.

Could you for testing purposes please get a free DeepL API key and see if the translation quality is much improved? It could very well be that it is the OCR and not the translations that are at fault, most of the time.

Thank you once again, I will respond to the rest of your feedback in other threads as time allows! :)

(1 edit)

” Could you for testing purposes please get a free DeepL API key and see if the translation quality is much improved? It could very well be that it is the OCR and not the translations that are at fault, most of the time. “

I have tested DeepL API translation. And while it is indeed better, the additional time each translation takes - roughly 2x longer than local - is just not worth it. I’m lucky enough to be bilingual so the tool is only there to support/correct my live translations. But this makes speed paramount.

Developer

Good to know! I think that is how most people would prefer it - granted the local models are good enough. I think they are, but they can be  a lot better and they will be.. in time.