Add larger language model support

the current model is already somewhat powerful, but pales in comparison to medium or large Whisper models. This is becoming especially apparent when using other languages (in my case German). Let users choose what they prefer: A transfription that's fast or one that is significantly more precise but slightly slower.