Hey! Thank you for your answers! I'll investigate this more. I'm indeed not using the Voice Fixer, I find it a bit unstable and it messes with non verbal sounds.
Hey, after prolonged usage, I’m having the same problem I think. I cannot generate anything else through Mrq without having the OOM Cuda allocation error. What’s even weirder, I just shut down…
Nevermind, the issue came back with me as well, it seems quite random...
Hey! Well it does sound pretty good, even though of course it’s a bit random, but I’d say it outputs good results 40% of the time. What is a bit frustrating though is that the results are NOT…
Great news!! I've managed to get a very good validation curve over almost 200 epochs, only by changing the learning rate from 1e-05 to 1e-06. The training loss mel is still around 2.0, but I think…
Hello!
Allow me to share my own experience about this matter. I, too, have been struggling for like a week on the problem of the validation curve going up almost immediately, like as soon as…
Hey! I noticed the same problem, and I am also certain it used to work before!
Okay thank you for these clarifications!
I'm really struggling with the validation though, the validation yellow line starts going up almost as soon as the training starts :/ I've seen [this…
Oh okay I’ll try working with the text length then for now. Write talking about the number of characters right, spaces included?
Thank you!
Okay I'm closing the issue because I figured out my process. For those who are interested, here's what I think you have to do:
Case 1 - You want to add more audio clips after having started…
Thank you very much for all that information! I think I have all I need for now, I'll make another post should more questions arise. A thousand thanks for your patience and for the great work you…
Hi! Is it possible to resume a training and add new clips? Or should that rather be another finetuning altogether?
After preparing the dataset but before actually training, under train.txt, add in the additional non-verbal terms into there.
Should I also add them in the actual whisper.json transcription?