Rehearsal Audio Stream Segmentation and Clustering

Rehearsal recordings are valuable for music ensembles to improve their performance. However, since rehearsal recordings are typically hours long and contain disordered, disrupted, and unclassified musical content (mixed with different sections of different music pieces, conductor’s talk, all kinds of noise between rehearsal intervals and so on), they’re hard to use directly. For example, recordings for a particular song are not easily found. This paper presents a variety of approaches to extract all the musical segments from the disarrayed rehearsal audio stream and then cluster the segments belonging to the same piece of music together. The procedures discussed herein have the potential to be applied in a large scale music database to accomplish the music information retrieval (MIR) tasks.