Methods and apparatuses are disclosed for automatically composing a song are disclosed. In an embodiment, a method includes receiving music performance data by a processor. The processor may then segment the music performance data based on at least one structural attribute into at least a first musical segment, where the first musical segment is associated with at least one musical attribute. The processor may then determine an affinity value for the first musical segment based on the at least one musical attribute. The affinity value represents a degree of similarity between the first musical segment and a second musical segment having the at least one musical attribute. The processor may then generate a musical composition based on the affinity values associated with the first musical segment and the second musical segment.