LaurinmyReha
LaurinmyReha
Any updates on this? i ran into the same problem
Thank you very much. This made it work! :)
Check out this variant of whisper that was specifically designed to improve timestamps and halucinations. https://github.com/nyrahealth/CrisperWhisper Feel free to also checkout the paper: https://github.com/nyrahealth/CrisperWhisper
Sure, the most comprehensive and complete explanation will probably be reading the accompanying paper and the additional notes in the README.md of the repo: https://arxiv.org/pdf/2408.16589 If it is still unclear...
Check out this variant of whisper that was specifically designed to improve timestamps and halucinations. For english this should improve things dramatically. https://github.com/nyrahealth/CrisperWhisper Feel free to also checkout the paper:...
The text in a box is just to visualize the timestamps. I am sorry if it does not directly solve your specific problem. I would be curious to see this...
look, if you want to be like this i cant help you. Good luck with your problem!
Well since the model provides decent timestamps and transcription accuracy for english as proven in the paper and provides great timestamps converting these to subtitles format should be a piece...
nice! Well the easiest would be if you could upload it to google drive and share the link with my under [email protected] :)
Maybe this variant will solve your problems. https://github.com/nyrahealth/CrisperWhisper Timestamps around pauses are notoriously bad for the whisper model when using DTW due to the tokenizer. More details can be found...