Multipitch Detection Based on Weighted Summary Correlogram

In this paper, we introduce a multipitch detection algorithm which is based on weighted summary correlogram. The weight is described as a conditional probability which models the relationship between fundamental frequency (FO) of periodic sound and response frequency of its dominated channels. Modified by this weight, SACF obtains more robustness to noise and to sub-harmonic error. The proposed algorithm can be used to track single or multiple pitches under noisy environment. Its performance is evaluated on 100 mixed sounds which comprise 10 voiced speeches and 10 different kinds of noises. The results show that our model has better performance than existing algorithms.