Vocal Melody Extraction via HRNet-Based Singing Voice Separation and Encoder-Decoder-Based F0 Estimation

Vocal melody extraction is an important and challenging task in music information retrieval. One main difficulty that, most of the time, various instruments singing voices are mixed according to harmonic structure, making it hard identify fundamental frequency (F0) a voice. Therefore, reducing interference accompaniment beneficial pitch estimation In this paper, we first adopted high-resolution network (HRNet) separate vocals from polyphonic music, then designed encoder-decoder estimate vocal F0 values. Experiment results demonstrate that effectiveness HRNet-based voice separation method on melody, proposed (VME) system outperforms other state-of-the-art algorithms cases.