This paper presents a structure mining scheme for music recordings. The term adaptive refers to the fact that the method relies on an adaptive scheme to detect similarity on the diagonals of the self-similarity matrix of the recording and removes the need for hard thresholds during this processing stage. Structural analysis is subsequently cast in a clustering framework. The output of the adaptive scheme is used to initialize a hierarchical data clustering algorithm whose output is a representation of the recording in terms of non-overlapping repeating patterns. The proposed method has been evaluated on a corpus of popular music recordings and various performance measures have been computed.