The number of audience recordings of concerts on the internet has exploded with the advent of smartphones. This paper proposes a method to organize and align these recordings in order to create one or more complete renderings of the concert. The process comprises two steps: first, using audio fingerprints to represent the recordings, identify overlapping segments, and compute an approximate alignment using a modified Dynamic Time Warping (DTW) algorithm and second, applying a cross-correlation

more »

... round the approximate alignment points in order to improve the accuracy of the alignment. The proposed method is compared to two baseline systems using approaches previously proposed for similar tasks. One baseline cross-correlates the audio fingerprints directly without DTW. The second baseline replaces the audio fingerprints with pitch chroma in the DTW algorithm. A new dataset annotating real-world data obtained from the Live Music Archive is presented and used for evaluation of the three systems.