Unsupervised Classification Techniques for Multipitch Estimation
In this paper, we present a fast and efficient technique for multipitch estimation of musical signals. We deal with mixtures where several instruments are present in a monophonic recording. The approach consists in clustering the spectral peaks of the mixture to obtain a spectral representation of each musical note. These spectra are then used to estimate the fundamental frequencies. We compare two techniques for the classification of the spectral peaks: a K-means procedure and a simpler aggregation technique associated to a criterion that represents the closeness to harmonicity for any couple of frequency peaks. This comparison is made on complex mixtures holding various musical instruments and piano chord mixtures. The effectiveness of the two estimation methods is presented using computation of pitch recognition rates and mean source number estimate.
Click to purchase paper or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members, $5 for AES members and is free for E-Library subscribers.