In this paper, we present methods for automatically producing representative excerpts of music. We maximize the segment similarity to the entire work to find the most similar excerpt, and maximize the segment novelty score to find the most meaningful excerpt, and combine the two measures to find the most representative excerpt. We discuss variations on the method, and present experimental results. These results demonstrate that the method can find significantly representative excerpts, just using very few assumptions about the source music.
展开▼
