A compositional tool classifies and indexes loops in a library of audio loops by generating a time-stamped series of the durations of the percussive events that comprise each loop. The duration of a percussive event is based on spectral masking, in which a subsequent event having a spectral similarity to a prior event terminates the prior event. A composer queries the library with a query loop, and the system returns loops ranked according to the distance of their rhythmic signatures from that of the query loop, the distance determination being based on dynamic time warp analysis. Rhythmic signatures may also be used to classify and index video sequences.