Good point but the problem with local hosting is that if you want to use the larger models it will take a long time to transcribe a file. We use multiple gpus and we do speaker detection, sound detection and it is has a rich audio editor.