- Music summarization involves the process of identifying and presenting melody snippets carrying sufficient information for highlighting and remembering a song. In many commercial applications, the problem of finding those snippets is addressed by having humans select the most salient parts of the song or by extracting a few seconds from the song's introduction. Research in the automatic creation of music summaries has focused mainly on the extraction of one or more highly repetitive phrases to represent the whole song. This paper explores whether the composition of multiple "characteristic" phrases that are selected to be highly dissimilar to one another will increase the summary's effectiveness. This paper presents three variations of this multi-phrase music summarization approach and a human-centered evaluation comparing these algorithms. Results showed that the resulting multi-phrase summaries performed well in describing the songs. People preferred the multi-phrase summaries over presentations of the introductions of the songs.