A new technique identifies versions of the same song
October 22, 2009
A new technique identifies versions of the same song. Credit: SINC
A team of researchers from Pompeu Fabra University (UPF, Spain) has developed a system to identify common patterns in versions of songs, which will help to quantify the similarity of musical pieces. The technique, which appears in the New Journal of Physics, could be applied to analyse time series of data in other fields, such as economy, biology or astronomy.
"What we propose is a measure to quantify cross recurrences between two songs, that is, to be able to analyse repetitions of different musical patterns that have previously been identified from the tonal or harmonic content of the audio recording" Joan Serrą, co-author of the technique and researcher of the Musical Technology Group (MTG) of UPF in Barcelona, explains to SINC.
Serrą and his team have developed this method, based on mathematical equations, which makes it possible to identify the concurrent presence of tonal events on two song tracks (taken from a CD or other device). The results can be visualised using Cross Recurrence Plots (CRPs).
For example, therefore, the researchers have taken the tonal profile from the song Day Tripper by the Beatles and have compared it with the version performed by the group Ocean Colour Scene, as well as with a different song, I've Got a Crush on You by Frank Sinatra. In the first case, the CRP shows oblique lines that reveal matches between the two versions, but in the second this pattern does not appear.
Serrą indicates that "The identification of versions of the same song (whether or not it is by the original artist, with the same instruments, with the same or different lyrics or language, in the studio or live) may be very interesting for scientific, commercial and intellectual property reasons, or simply for the interest of the end user".
The researcher points out that the cross recurrence plots and their quantification measures are "powerful tools for analysing and comparing time series of any type of data", which means they can be used in disciplines such as astrophysics, biology, engineering or the economy. For example, it would be possible with this technique to analyse over a period of time the correlations between the Ibex and the Dow Jones or with other stock exchange indexes.
The possible applications of this study in different fields have led to its publication in the New Journal of Physics, a journal of generalist and multidisciplinary physics.
More information: J. Serrą, X. Serra y R. G. Andrzejak. "Cross recurrence quantification for cover song identification". New Journal of Physics 11: 093017, 2009.
Source: FECYT - Spanish Foundation for Science and Technology



Besides, this musical analysis was used just as a test for the technique, if I understand correctly. I'm sure there are better uses for it besides tormenting would-be composers.
Did they test the new method with the Haydn Variations by Brahms? Or with the dozens of versions of "La Folia"? Did it find the cross recurrences within BWV 1080? Or the originals of Bernd Alois Zimmermann's "Musique pour les Soupers du Roi Ubu" which consists of recurrences only?
I doubt it.
Further, where the music is discovered to be accidentally similar to existing music (which will happen the majority of the time) the musicians can consult a database for the terms and conditions of reusing the music they have incorporated in their new work
I have never met a (real) musician who did not want the ability to do just these things. Nothing worse than believing you are being original, breaking new groud, and then having some bubble-bursting music agent tapping you on the shoulder
Congratulations to the UPF team. I hope that you have, or will discover, the right commercial, judicial and governmental support to make your tool widely available