For the user's pitch OnKey uses CREPE-tiny (https://github.com/marl/crepe) but ported to Rust with the model running on ONNX Runtime. For the singer's pitch also CREPE but the full version.

The vocal separation for the song uses UVR: https://github.com/Anjok07/ultimatevocalremovergui

You can find more details here: https://singonkey.app/docs/how-it-works