Tempo Estimation from Urban Music Using Non-Negative Matrix Factorization
Automatic tempo estimation is a useful tool for DJs. Several algorithms have been introduced during the last years. In this paper, a system for tempo induction of urban music is presented. While other algorithms are designed to work on all kinds of music, this one utilizes characteristics that are typical for urban music, e.g., hiphop, like constant tempo and a time signature of 4/4. Activation functions are obtained from non-negative matrix factorization of the spectrogram, and then used as periodicity detection functions. An initial pool of dominant periodicities is collected, using a comb grid approach on a combined autocorrelation representation of the periodicity detection functions. From this pool, the most promising periodicity and it’s power-of-two-multiples are determined and rated. Based on this rating, the periodicity candidates are sorted and eventually transformed into bpm values. The system’s performance is close to perfect (99.85% accuracy), if octave errors are accepted as correct estimates. Two reference systems are outperformed by the proposed approach.
Click to purchase paper as a non-member 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 and is temporarily free for AES members.