Real-time performance comparison of tuning frequency estimation algorithms

In this paper, a comparison of the real-time performances of different algorithms for the estimation of the concert pitch (tuning frequency or reference frequency) of music recordings is presented and discussed. The unavailability of ground-truth datasets makes this kind of evaluation on real music recordings less trivial than what it may initially appear. Hence, in this paper we investigate the algorithms best-case performances using simple generated sounds and than we study the estimation reliability and the real-time performances using real world recordings. In particular, we focus on the standard deviation of the estimation for various length of the signal, the reliability of the on-line estimation and the computational complexity.