We present a simple heuristic-based Sinusoidal Partial Tracking (PT) algorithm for singing analysis. Our PT algorithm uses a heuristic of minimal frequency and magnitude difference to track sinusoidal partials in the popular music. An Ideal Binary Mask (IBM), which is created from the ground truth of the singing voice and the music accompaniment, is used to identify the sound source of the partials. In this justifiable way, we are able to assess the quality of the partials identified from the PT algorithm. Using the iKala dataset along with the IBM and BSS Eval 3.0 as a new method of quantifying the partials quality, the comparative results show that our PT algorithm can achieve 0.8746 ~ 1.7029 dB GNSDR gain, compared to two common benchmarks, namely the MQ algorithm and the SMS-PT algorithm. Thus, our PT algorithm can be considered as a new benchmark of the PT algorithm used in singing analysis.
Cite as: Lin, K.W.E., Anderson, H., So, C., Lui, S. (2017) Sinusoidal Partials Tracking for Singing Analysis Using the Heuristic of the Minimal Frequency and Magnitude Difference. Proc. Interspeech 2017, 3038-3042, doi: 10.21437/Interspeech.2017-17
@inproceedings{lin17b_interspeech,
author={Kin Wah Edward Lin and Hans Anderson and Clifford So and Simon Lui},
title={{Sinusoidal Partials Tracking for Singing Analysis Using the Heuristic of the Minimal Frequency and Magnitude Difference}},
year=2017,
booktitle={Proc. Interspeech 2017},
pages={3038--3042},
doi={10.21437/Interspeech.2017-17}
}