We explore different approaches for generating music from the flocking dynamics of groups of mobile autonomous agents following a simple decentralized control rule. By developing software that links these dynamics to a set of sound wave generators, we study how each approach reflects sonically the transition to collective order and which produces musically interesting results. We consider three qualitatively different ways of translating flocking dynamics into music: (1) A direct approach that maps agent positions to sounds, (2) a synchronization approach where each agent carries an oscillator that couples to neighboring agents, and (3) a physics-inspired approach that mimics the sound that would result from an effective friction between neighboring agents. We find that all approaches allow the listener to discriminate between different phases in the system, but that the second and third can yield more musically interesting and appealing results.