A novel method to segment stereo music recordings into formal musical structures such as verses and choruses is presented. The method performs dimensional reduction on a time-azimuth representation of audio which results in a set of time activation sequences, each of which corresponds to a repeating structural segment. This is based on the assumption that each segment type such as verse or chorus has a unique energy distribution across the stereo field. It can be shown that these unique energy distributions along with their time activation sequences are the latent principal components of the time-azimuth representation. It can be shown that each time activation sequence represents a structural segment such as a verse or chorus.
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.