This repository contains the offical implementation for the paper accepted by ICASSP 2021 《Frequency-Temporal Attention Network for Singing Melody Extraction》.
The codes are implemented by Keras 2.
-
We add a pytorch implementation version in the network directory. Special thanks to Dr. Ke Chen.
-
Please use 8K sampling rate, 768 window-size and 80 hop-size, when computing CFP. Other settings may cannot reproduce the results of the paper.