Request to add a better model

[Rex000]

First off, thanks a lot for this app. Now, the issue is that I'm trying to generate subtitles for an old tv series but the subtitles generated are way off in terms of accuracy. I've found the models used by other apps such as PowerDirector to be pretty accurate. So it's my humble request to please consider adding more accurate models like them. Thanks.

[Serkali-sudo]

Well we are using smallest vosk model so that is kind of expected. I used the smallest model because we run the model on device and i wanted to keep the app size small and run fast.
But i guess you can try running the biggest vosk model. The model is 1.8GB though so i am not sure if it will work on device .
You can find the model here: https://alphacephei.com/vosk/models
After downloading it replace the old model-en-us folder in models then put the new one but make sure it is the same name as before. Then without changing anything in the code just run the project in android studio and see what happens.

Btw i checked powerdirector and i dont think it is doing auto subs on device like we do . They probably have a server running a big whisper model. The app is probably just sending the audio of the selected video to server and it is processed by the server.

I will try to implement this when i have the time.

[Serkali-sudo]

Btw it is probably best to download vosk model files at runtime instead of embedding it to the apk.
That way we can add other languages vosk supports

[Rex000]

Thanks for the reply. You're right, PowerDirector is indeed doing it on its server. What disappointed me though is that it doesn't allow you to export the subtitles to an srt file.

And yes I agree that it's best to download the model files at runtime. I've seen some apps actually do that.

I would try what you are suggesting(downloading and running the biggest vosk model) but I don't have any knowledge of Android Studio.