This paper describes a novel onset detector for music signal based on the smoothly varying-Q transform, where the Q-factors vary following a linear function of the center frequencies. The smoothly varying-Q factors allow the time-frequency representation to coincide with the auditory critical-band scale. As the analysis basis of the input signal, the time-frequency image generated by smoothly varying-Q transform indicates the frequency evolution. On the detection stage, a bidirectional mismatch procedure is designed to estimate the discrepancies of frequency partials between currently processed frame and its bidirectional neighboring frames. An onset strength signal is obtained by measuring the mismatch error between the neighboring frames. The evaluation of the proposed algorithm is performed on a fully onset annotated music database and the results show that the proposed algorithm can achieve high detection accuracy and satisfied results.
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 free for AES members and E-Library subscribers.