Transcription on serbian language for base model is not working
So here is some strange issue with this and so I tried to use on my native language and I didn't succeed to make it. On English It worked with base model for me and I tried to use Serbian language and in some cases it wrote words on English if i said something in Serbian, or it just wrote something different again on English and Auto Translate to English is off.
Log in to comment and vote
Comments6
Vowen
Jan 5
Hey @Stef , we’ve released 0.1.9 version that addresses this issue. Please try updating the app to the latest version by going to “Settings > General” and clicking on “Check for updates”. You can verify the changes by selecting one of the models and changing the language to “Auto-detect” or “Serbian”. Thank you!
Stef
Jan 6
It's great that you managed to fix the error, thank you. I uploaded to the latest version, now it manages to do the transcription, but it can be quite slow. Until about 30-40 seconds when I release the hotkey and get the transcription, I don't know if the option of connecting an AI model can speed things up, and as for the AI model, I think it would be a good option, you have the option to choose models that are not only limited to what is in the menus. For example, on Openrouter there is 500+ models and when you choose the Openrouter application option, you only have Google fast 2.0 and gpt 4 mini.
Vowen
Jan 6
In the next release we will be adding support for GPU acceleration on Windows which should make the transcription much much faster. But for most use cases, the “base” model should work fine without any hit to accuracy. You can try switching to the base model and see if that helps with the transcription speed. We use the base / tiny model ourselves on a daily basis. Do let us know if it’s still slow on base model as well, we will check out what could be causing this.
Regarding OpenRouter, we will be adding support for it in the next release. Thanks!
Stef
Jan 6
Yes, when I set the model to "Base" the transcription is faster but the text quality is very poor, when it is set to "Small" it is much better but as I wrote it takes 30-40 seconds. Great news about OpenRouter support, thanks!
Vowen
Dec 20, 2025
Hey @Stef, I’ll check it out specifically for Serbian on base model and see if we can do something. In the meanwhile, could you try on a larger model / parakeet, I believe the smaller models are inaccurate for other languages. Thanks for the feedback, this really helps improve the product.
Stef
Dec 21, 2025
I forgent to mention that i tried almost on all models in app and it is the same.When you finish whit dictation it just start to circle the circle for a very long time and then nothing happend it like shot in blank.and thank you for an app like this, this has great potential.