Transfer Basic-Pitch model weights to PyTorch

TeeJayBaker / PolyDDSP

Polyphonic generalisation of DDSP

16 stars 0 forks source link

Transfer Basic-Pitch model weights to PyTorch #3

Open TeeJayBaker opened 7 months ago

gudgud96 commented 6 months ago

Hey! Just randomly saw this issue when searching for BasicPitch... Hope my repo helps - https://github.com/gudgud96/basic-pitch-torch

TeeJayBaker commented 6 months ago

hey @gudgud96, thanks for pointing me to your repo! I've actually also just finished implemented and transferring weights myself too, but will definitely be making use of your implementation for inference benchmarking etc.

gudgud96 commented 6 months ago

@TeeJayBaker Ah, that's super wonderful! Glad you have made it too. Yes feel free to use it as reference or benchmark, I find mine has a max absolute error of around 6e-4 on the model output (I find that the floating point error originates from the normalized_log part), do let me know if you have similar or better findings!

Meanwhile, just wanna also say PolyDDSP is a super cool project! I starred it a long time ago. It will be a huge contribution to DDSP.