kane

Can we judge an artist by their covers?

Within the pop music landscape, artists often record a version of another artist’s material (known as a ‘cover song’ or simply a ‘cover’). Often, this interpretation involves many changes to the music and often there are major differences to aspects such as arrangement, instrumentation, tempo, dynamics and style. This project provides an analysis and comparison of musical data drawn from cover versions by the same artist. The material selected for analysis is from the Johnny Cash album ‘American IV: the Man Comes Around’. The album is a collection of ‘cover songs’ from the pop music cannon and each song features heavy interpretation from the original.

Goals & Motivation

The main overarching aim of this project is to identify if there are any interesting patterns that arise from the data drawn from the cover version repertoire of one artist. The project plans to investigate to what extent, if any, patterns exist across a specific cover version repertoire (i.e. a cover album) by the artist in question and whether these patterns point to any overarching approaches to interpretation by the covering artist.

Context

Within the MIR community, the task of cover song detection has undergone several years of evaluation and development (Serra et al. 2008, Foucard et al. 2010). In 2007, analysis of the Music Information Retrieval Evaluation eXchange (MIREX) Audio Cover Song Identification (ACS) tasks were published by Downie et al. These tasks were designed to widen MIR research activity to include the conceptualisation that ‘musical works retain their identity notwithstanding variations in style, genre, orchestration, rhythm or melodic ornamentation’. This project takes this idea further, by proposing that when an artist ‘covers’ existing material, they bring their own musical style, expression and interpretation to the piece. The project aims to attempt, in a small way, to identify and quantify these contributions from one artist within a basic empirical framework.

It has been documented within the landscape of cover song identification that the top performing algorithms use chromagrams and/or pitch class profiles as the main feature representation (Ellis & Poliner 2007, Muller et al. 2011). These tools, alongside key and tempo change identification methods, seem to best identify possible musical interpretations. The worst performing algorithms seemed to be based upon timbre analysis. However, these tools can still provide useful results regarding music similarity. Utilising these tools it is possible to provide a basic analysis of one artist’s approach to interpretation within the confines of a set of suitable cover versions (i.e. a cover album). Due to the myriad of influences and circumstances that lead to any artist’s interpretation of existing work, this expectation may not be able to be extended to a provide a overarching definitive analysis that encapsulates all cover versions by any given artist.

Data Set

The data set of the project comprises of 6 songs, with an audio file for each original and each cover version:

• • Personal Jesus (original by Depeche Mode)

  • Hurt (original by Nine Inch Nails)

  • First Time Ever I Saw Your Face (original by Peggy Seegar)

  • Bridge Over Troubled Water (original by Simon & Garfunkel)

  • I Hung My Head (original by Sting)

  • In My Life (original by The Beatles)

Each song has been edited into first verse and last chorus sections, with the edit points in matching places between each audio file. This comprises of a total of 24 audio files. Annotations have been conducted for each audio file via Sonic Visualiser and include bars and beats time instances, tempo curves, loudness curves (via Mazurka Power Curve), chord transcription (via Chordino) and spectrograms (including traditional and melodic range). All .csv data was imported into MATLAB for use in Elias Pampalk's MAToolbox and the MIRtoolbox, developed by The Centre of Excellence in Interdisciplinary Music Research at the University of Jyvaskyla.

Analysis: Tempo & Dynamics

Over 240 initial plots were computed via the MAToolbox, which were meaningfully scaled down to 4 plots per piece and used to determine comparisons between originals and covers and between adjacent covers.

For the Verses of material that had an ‘electronic original’ (i.e. Depeche Mode & Nine Inch Nails), Tempo in Score & Performance Time plots indicate that Cash often displays more variance with regards to tempo and loudness than originals (see Figure 1 below for tempo). For the Choruses of an ‘electronic original’ (i.e. Depeche Mode & Nine Inch Nails) Tempo in Score & Performance Time plots indicate that Cash still has variance but is often more stable (i.e. less variance) with regards to tempo and loudness (see Figure 2 below for tempo).

The above appears to hold true but in a lesser manner for the Verse and the Choruses of material than had a ‘pop original’ (i.e. Sting, Beatles or Simon & Garfunkel - see Figures 3 and 4 below for tempo).

This shift in variance could possibly be down to the repetitive ‘hook’ nature of the Choruses in Cash's versions and the fact that Cash often employs Blues ‘vamping’ throughout his Choruses, delivering a more solid, stable tempo and loudness curve.

Verse and Chorus loudness plot comparisons can be downloaded here.

Analysis: Key & Structure

Over 504 initial plots were computed via the MIRToolbox, which were meaningfully scaled down to 8 plots per piece and used to determine comparisons between originals and covers and between adjacent covers.

Five out of six songs featured key transposition, with Cash's interpretation of The Beatles ‘In My Life’ the only song not transposed. This is evidenced by Chromagram (Pitch Class) plots, Chromagram Similarity Matrix plots, the output of the Sonic Visualizer Chordino VAMP plug-­in and by ear (see Figures 5 and 6 below).

Path-like structures in chromagram, spectral & MFCC similarity matrices seem to indicate that Cash utilises similar repetition and homogeneity-based properties as the original versions (i.e. structural elements such as Verses and Chorus have similar lengths), even though the material was transposed (see Figure 7 below). More chromagram contrast is apparent in the Electronic and Pop originals than Cash's interpretations (see Figure 8 below). This is indicated by chromagram novelty curves (i.e. moments in time where there is the most important contrasts). This could possibly be due to Cash's less complex arrangements which feature less instrumentation than the originals and exhibit a more raw blues vibe. The production methodology of Cash's interpretations seem to focus on the core of the song, as apposed to some of the originals which leverage production techniques as much as songwriting techniques (i.e. Depeche Mode, Nine Inch Nails, Simon & Garfunkel etc).

Conclusion

As previously expected, due to the myriad of influences and circumstances that lead to any artist’s interpretation of existing work, it is not possible to a provide a overarching definitive analysis that encapsulates all cover versions by Cash. However, from the above analysis several interesting patterns have emerged:

Cash's Verse Interpretations:

Cash often delivers a stripped back version mostly driven by acoustic guitar and voice. This often provides musical variance of tempo and dynamics in Verses, giving a 'blues' inspired feel.

Cash's Chorus Interpretations:

Cash often builds more instrumentation during his Choruses to provide more of a pop friendly ‘hook’, but there is still less multi track layering than the originals. There is often less musical variance of tempo and dynamics in Cash's Choruses than his Verses - this possibly could be down to the repetitive ‘hook’ nature of the Choruses in Cash's versions and the fact that Cash often employs Blues ‘vamping’ throughout, delivering a more solid, stable tempo and loudness curve. There is less chromagram contrast than the original versions – possibly due to Cash's less complex arrangements which feature less instrumentation than the originals and exhibit a more raw blues vibe. The production methodology of Cash's interpretations seem to focus on the core of the song, as apposed to some of the originals which leverage production techniques as much as songwriting techniques (i.e. Depeche Mode, Nine Inch Nails, Simon & Garfunkel etc).

Cash's Tonality Interpretations:

Cash transposed five out of the six songs studied into a new key. This may possibly be due to the need to shift the material into a singing register he is more comfortable performing in. However, Cash retained the tonality from the original versions (i.e. there was no major to minor tonality changes). No specific transposition pattern appears but Cash seems to favour the keys C, E, F, A.

Cash's Instrumentation Interpretations:

Cash's version exhibits major instrumentation changes. His work features the re-interpretation of the originals via the arrangements of instrumentation that are more closely associated via the Blues idiom (i.e. acoustic guitar, piano, organ, mellotron, slide guitar etc).

Cash's Rhythm Interpretations:

Cash's versions employ tempo changes but none are too drastic. The most major rhythmical change is a meter change from compound time 9/8 to 4/4 for the song ‘I Hung My Head’.

Cash's Form Interpretations:

Throughout his versions, Cash focuses on delivering the main song sections that are driven by the lead vocal (i.e. Verse / Chorus) and seems to have reduced the use of instrumental sections that are more prevalent in the electronic and pop original versions.

To summarise, Cash uses a series of techniques to interpret the original material into his own style, with the above patterns being identified via analysis.

Downloads

A 74 page PDF containing both Tempo & Loudness and Chroma & Spectral Plot Comparisons of the data set can be downloaded here (10mb).

Sonic Visualizer annotations and corresponding audio files (.wav, .sv, .svl and .csv formats) can be downloaded here (684mb).

Performance Worm and Performance Curve videos for four songs created via www.VMus.net can be downloaded here (87mb).

Automated MATLAB scripts for MAToolbox and MIRToolbox plot generation can be downloaded here (8kb - no audio files - please see link above).

The MAToolbox can be downloaded here.

The MIRToolbox can be downloaded here.

Bibliography

Ellis, D., Poliner, G., Identifying ‘Cover Songs’ With Chroma Features And Dynamic Programming Beat Tracking, in Proc. IEEE Int. Conf. Acoust., Speech, Signal Process. (ICASSP), Honolulu, HI, Apr. 2007, vol. 4, pp. 1429-1439.

Ellis, D. & Bertin-Mahieux, T. “Large-scale cover song recognition using hashed chroma landmarks,” in 2011 IEEE Workshop on Applications of Signal Processing to Audio and Acoustics.

Downie, J., Bay, M., Ehmann, A., Jones, M. (2008) Audio Cover Song Identification: MIREX 2006-2007 results and analyses, ISMIR 2008: Proc. 9th International Conference of Music Information Retrieval

Foucard, R., Durrieu, J., Lagrange, M., Richard, G., Multimodal Similarity Between Musical Streams For Cover Version Detection, in Proc. ICASSP, Dallas, TX, Mar. 2010, pp. 5514–5517

Muller, M., Ellis, D., Klapuri, A., Richard, G. Signal Processing for Music Analysis, IEEE Journal Of Selected Topics In Signal Processing, Vol. 0, No. 0, 2011

Muller, M., Jiang, H. & Grohganz, H., SM Toolbox: MATLAB Implementations for Computing and Enhancing Similarity Matrices, AES 53rd International Conference, UK: London, 2014

Muller, M., Kurth, F., Enhancing Similarity Matrices for Music Audio Analysis, In Proceedings of the International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2006.

Muller, M., Clausen, M., Transposition-Invariant Self-Similarity Matrices, In Proceedings of the International Conference on Music Information Retrieval (ISMIR), 2007

Muller, M., Kurth, F., & Clausen, M., Audio Matching via Chroma-Based Statistical Features, In Proceedings of the 12th International Conference on Music Information Retrieval (ISMIR), 2011.

Serra, J., Gómez, E., Herrera, P., Serra, X., Chroma Binary Similarity And Local Alignment Applied To Cover Song Identification, IEEE Trans. Audio, Speech, Lang. Process., vol. 16, no. 6, pp. 1138–1151, Aug. 2008.