Shared process or pipeline
Hi!
I've recently been experimenting with your RVC voice models. I'm really impressed with the models you created, especially the TF G1 models. I wanted to also start creating models for 80s characters that I haven't been able to find. Could you tell me what your process has been to create the models? I'd like them to match your quality.
Thanks!
Hey Darkjedv! Sorry for the late reply, I didn't see your message. Thanks for the compliments on my models, I appreciate it!
To clarify, I've made the G1 models a long while back, back when I was a beginner to model making so the models presented now could be worked on much better, I just usually pass on reworking old models lol.
So my general method of creating models is making sure the dataset used is clean and accurate to the original source. That means trying your best to finding the raw vocals for the dataset first and foremost. If you can't do that, the next option would be cleaning up the dataset using UVR or any other program that helps with isolating vocals. I don't recommend using datasets with background noise/music as it can degrade the quality of the model overall, it can add up quickly.
This next step isn't needed but sometimes you can alter the dataset to sound better with Audacity, sometimes by adding bass or just amplifying the volume of the dataset. Altering the dataset varies from model to model, if it helps make it could clearer or better, more power to you.
If you'd like to know more about my methods, try to dm me on the Ai Hub Server from Discord. It's Starry Night there.
Awesome, thanks for taking the time to respond. I'll message you when I get a chance!
Alright.