An implementation of "Towards Improving Harmonic Sensitivity and Prediction Stability for Singing Melody Extraction", in ISMIR 2023
Will update training/inference instructions soon. Basically python feature_extraction.py for caching CFP/z-CFP before training. Then python main.py train will call tonet.py and start the main training loop. tonet.py in turn calls the PianoNet model in piano_net.py.
Standalone testing can be done using python main.py test
The data used for the experiments can be found here: https://drive.google.com/file/d/1QKX6rpuRxMPt54HOqNQztmLQqGlCALZ4/view?usp=sharing