emcodem

yup, i didnt really look into that part yet, its a minor issue for me

i don't, i'm happy to get 99% output of a 3 hour clip :D In my case, a human corrects the stuff afterwards anyway, the timecodes and lots of words...

Just adding another puzzle piece from the whisper.cpp project but this is only for the last 500 samples, not very helpful i fear (except for streaming mode maybe?) // if...

Update: the patch above works but we can do much better. This updated version only deletes a single token from the prompt past instead of the whole prompt which can...

Nah sorry, i just realize this code was totally garbage, it needs to remove much more from the history. I am currently testing a more sophisticated method and post here...

@zlbbme sorry not sure if i understand the question. The .zip file download i provided contains my modification, there is no parameter to turn this off or on...

OK so here my youngest attempt. **Download Whisper Desktop and main.exe for users:** [WhisperCM_prevent-and-remove-duplicated-tokens-from-prompt.zip](https://github.com/Const-me/Whisper/files/12254499/WhisperCM_prevent-and-remove-duplicated-tokens-from-prompt.zip) Instead of just detecting repeated text, i now prevent repeated stuff to be put into prompt...

@softlypink if you are asking for "security/virus" related reasons, there is no way for anyone to tell if any compiled exe file is safe to use. Antivius software attempts to...

Closed because solved for me in the code i posted above. Final note: all "forever" related issues (might be same sentence, lowercase, counting up forever) can only be solved by...

In my mind such functionality has no place in a library like this. This library is speech-to-text. Other libraries are routing audio from application to device. Yet other libraries are...