#What AI?
1 messages · Page 1 of 1 (latest)
its a copy of post malones voice
but it has every style of him
in the song
here is like the most i have video wise
apart from that the audio files is all i have
That’s RVC v2 most likely trained on a Titan 48k or OV2Super pretrain given how natural the vocal texture sounds. What makes these realistic isn’t just the architecture though, it’s the dataset quality and the inference settings.
The reason it captures every style of him is because whoever trained it used a well-curated dataset covering his full vocal range different energy levels, falsetto, his raspy chest voice , not just one type of delivery. That coverage is what gives the model emotional range.
If you’re trying to identify or recreate something similar, the pretrain + clean diverse dataset combo is the key. Happy to point you in the right direction if you want to go deeper
So the software is Applio (free, open source) that’s the training environment. But the software alone won’t get you that quality, it’s really about:
1. The pretrain Titan 48k or OV2Super is what separates natural-sounding models from robotic ones
2. The dataset clean, diverse vocals covering the full range of the artist
3. The inference settings index rate, pitch, F0 ,most people get this wrong
Honestly getting all three right is where most people struggle.
Do you think you would be able to chase down this exact model?
Yeah absolutely