Computing predominant local periodicity information in music recordings

The periodic structure of musical events plays a crucial role in the perception of tempo as well as the sensation of note changes and onsets. In this paper, we introduce a novel function that reveals the predominant local periodicity (PLP) in music recordings. Here, our main idea is to estimate for each time position a periodicity kernel that best explains the local periodic nature of previously extracted note onset information and then to accumulate all these kernels to yield a single function. This function, which is also referred to as PLP curve, reveals musically meaningful periodicity information even for non-percussive music with soft and blurred note onsets. Such information is useful not only for stabilizing note onset detection but also for beat tracking and tempo estimation in the case of music with changing tempo.