vassallo

Recognising Swing Ratio as a Genre Specific Rhythmic Characteristic

Thomas Vassallo

    1 Introduction

One of the ways in which a drummer’s performance deviates from the written representation of a piece is the timing. If a piece of music was to be played exactly as it was written there would not be slight temporal "shifts" of isolated notes with respect to their theoretical locations [1]. Research has revealed that these “shifts” take place on the metrical structure of rhythm [2]. This metrical structure contains anchor points that allow performers to employ systematic timing deviations. For example, funk drummers are known to often play “behind the beat” - a term that was coined in jazz music. “swing” refers to a particular metrical level: eighth notes. [3] This “swing” may be described as consecutive eighth notes which are performed as long short patterns. [4] This micro-timing produced by human drummers may be described as the “swing ratio” which is the “the duration of an eighth-note divided by that of the consecutive one.” [5] Swing ratio, the long short patterns found in consecutive eighth notes is sometimes employed as an artistic choice and this research investigated whether or not these artistic decisions display genre specific characteristics. 

    2 Data Collection

Three musical genres were initially selected for investigation. Funk, RnB/Soul and Reggae were chosen due to them often being rhythmically driven and groove drummers often being found performing and recording within these genres. A dataset of 15 track, consisting of 5 tracks per genre was selected. The songs were selected under two criteria: the piece is easily labelled as the genre it is being labelled with and that it is known to be a rhythmically driven song.  A four bar section of each song of steady rhythm was then selected, avoiding drum fills and song section transitions. The eighth notes of these extracts were then hand annotated within Sonic Visualizer. A CSV file was extracted from these annotations, providing the time instants position over time allowing for the swing ratio to be calculated. The Tempo and Beat Tracker plugin was used in Sonic Visualiser in order to calculate the tempo of each track.

Figure 1: Example of hand annotated 8th notes

The 15 tracks that make up the dataset are as follows:

      3 Early Results

A scatter plot showing the relationship of the swing ratio against the tempo distinguished each genre by the marker style and colour. Initial observations reveal that the majority of all tracks consist of little to no swing ratio at all. As the tracks are hand annotated only significant deviation from a straight 1:1 swing ratio may be considered as convincing results. The general lack of swing ratio found within tracks across all genres; the indication that there is a lack of obvious genre specific patterns; and the time consuming nature of the hand annotation of the eighth notes all resulted in a quick determination that this approach would not present the desired results. 

                                                                                                                                                              Figure 2: Scatter plot of hand annotated examples' swing ratio against tempo

4 The GTZAN-Rhythm Test-Set

The GTZAN-Rhythm dataset [6] is made up of 1000 audio track each 30 seconds in length. It consists of 1000 songs which may be broken down into 100 songs for each of the 10 genres. It contains high-level annotations, such as genre, artist, tempo, meter and most importantly swing. The dataset provides an ideal source of information with which to extract genre specific information, including that of swing ratio.

By simply observing the ratio of tracks with swing to tracks with no swing, one may deduce that the mere presence of swing ratio, is potentially a genre specific characteristic in itself.  Certain genres such as blues, country, jazz and reggae present significantly larger numbers of audio tracks with swing ratio, while genres such as pop disco and hip-hop present almost no tracks at all.  Another scatter plot is created in order to observe the relationship between swing ratio. Some clusters may be observed, particularly for the reggae songs as well as a clear presence of a number of tracks which contain a swing ratio of exactly 2:1.  This swing ratio is often referred to as the “triple feel” [3].

Figure 3: Scatter plot of hand annotated examples' swing ratio against tempo

5 Swing Ratio in Relation to Tempo

Research from previous studies analysing swing ratio as a relationship to tempo in jazz drummers indicates that swing ratio varies considerably for different tempi [7]. In order to easily observe each individual genres relationship with tempo, separate plots were created, and a first order polynomial was fit to each genre dataset. A general trend of a decreasing degree in swing ratio occurring as the tempo increases is observed. This is true for all genres excluding Blues, which displays an almost entirely constant fit.

Figure 4: Individual genre swing ratio against tempo with first order polynomial fit

In order to further investigate the relationship swing ratio has with different tempi the same data was used again but plotted differently. The Y axis now represents the “short” eighth note while the X axis again represents the tempo. The plots reveal that, as one would expect with increasing tempo, the duration of the “short” eighth note found within the swing ratio decreases in a fairly linear fashion. The curious finding is that this relationship changes when the tempo reaches roughly 150bpm. It is at this point that the note duration stabilises at about 100ms for all genres. The observation is less clear for that of the Reggae dataset due to its general lack of tempo variation between different songs. This value coincides with the nerve response times for both visual and auditory stimuli.

Figure 5: Individual genre "short" eighth note against tempo.

These tables aids in the representation of the minimum, maximum, maximum difference and mean of the swing ratio and tempo for each of the four genres in an attempt to extract further information. The mean, maximum and minimum of the swing ratio of all genres have relatively similar values. 

6 Reggae - Further Investigation

Due to Reggae presenting the most obvious clustering in both the swing ratio against tempo as well as the “short” eighth note duration against tempo plots, further investigations were made in order to determine the cause. First the GTZAN dataset is observed, quickly revealing that more than 50% of the reggae dataset consists of Bob Marley songs. This may indicate that this clustering is due to artist specific patterns rather than genre specific patterns. When comparing the swing ratio of all Bob Marley songs against all other reggae songs a slight pattern emerges. The swing ratio for the Bob Marley tracks cluster between the 1.4 and 2 range, while the rest of the reggae songs general reside in the 1.2-1.4 range. This is an indicator of artist driven artistic rather than genre driven choices found within swing ratio.

Figure 6: Swing ratio against tempo for Bob Marley and all other reggae examples.

7 Discussion

This project aimed to extract any relationship between a songs genre and the swing ratio it was likely to contain. Swing is a tool often utilised by human drummers as well as implemented as a form of humanisation when sequencing drum patterns. Having clear parameters would allow for sequenced drum patterns to be humanised in a genre specific manner. At this level of analysis there seems to be a general lack of genre specific characteristics to swing ratio. However, the research has reinforced the notion that swing ratio is reduced as tempo increases over most genres, and that there is a lower threshold of about 100ms for the “shot” eighth note duration across all genres. 

8 References

[1]  Bilmes J. “Techniques to Foster Drum Machine Expressivity.” In Proc. ICMC, 1993. 

[2]  Honing H. “From time to time: The representation of timing and tempo.” Computer Music Journal 35(3), 2001. 

[3]  Friberg A. and Sundström J. “Swing ratios and ensemble timing in Jazz performances: Evidence for a common rhythmic pattern.” Music Perception 19(3), 2002. 

[4]  Laroche J. “Estimating tempo, swing and beat locations in audio recordings.” In Proc. IEEE WASPAA, 2001.

[5] Gouyon F. “Rhthmic Expressiveness Transformations of Audio Recordings: Swing Modifications.”  DAFX-03, 2003

[6] U. Marchand, G. Peeters. “Swing Ratio Estimation” in Proc. of the 18th Int. Conference on Digital Audio Effects (DAFx-15), December 2015.

[7] Anders Friberg, Andreas Sundström. “Jazz Drummers' Swing Ratio in Relation to Tempo”, March 1999.