First off, love the project! Thanks for building something that is allowing access to some great open-source models 🙏
I am finding that local semantic correction appears not to be doing anything. I have a Apple M3 Max running Tahoe 26.2. I have tried all 4 different models (correctly downloaded on HF hub) and different categories. When I transcribe, it does show that semantic correction is processing in the UI and takes longer than when it is turned off, but the output is exactly the same as when it is turned off – including all ums, stutters, "wait no", etc.
First off, love the project! Thanks for building something that is allowing access to some great open-source models 🙏
I am finding that local semantic correction appears not to be doing anything. I have a Apple M3 Max running Tahoe 26.2. I have tried all 4 different models (correctly downloaded on HF hub) and different categories. When I transcribe, it does show that semantic correction is processing in the UI and takes longer than when it is turned off, but the output is exactly the same as when it is turned off – including all ums, stutters, "wait no", etc.