A Proposal to Compare the Similarity Between Musical Products. One More Step for Automated Plagiarism Detection?

Abstract

In previous works, the authors presented a measure of similarity between melodies by identifying them with sequences of ordered vectors and using a clustering process based on fuzzy logics. Along the same line, we propose a measure of musical similarity between fragments of digital audio. We present the SpectroMap algorithm that allows us to detect the local maxima of the audio spectrogram representation (also known as constellation map) and we compared the similarity between different maps belonging to different audio excerpts. As a result, it is obtained a value that represents the resemblance between two musical products. This procedure could be used as a non-subjective tool in automatic plagiarism detection. To illustrate this method, three experiments have been carried out comparing different versions famous pop songs. The results point to the usefulness of the method, although this should be contrasted with an analysis of the human perception of this similarity.