Microtiming

Microtiming refers to small-scale timing deviations of musical events from an underlying regular metrical grid: notes that sound slightly earlier or later than strictly even spacing would place them.^[1:1] In many musical traditions these deviations are systematic rather than random, and their interaction with other aspects of rhythm has been described as contributing to the sense of swing or groove.^[1:2]

Microtiming is studied in music psychology, ethnomusicology, music theory, and music information retrieval. Examples include the long–short eighth notes of jazz swing, the unequal beat subdivisions of West African jembe drumming, and the timing patterns of Brazilian samba.^{[2:1][3:1][1:3]} Whether microtiming makes music feel more groove-laden to listeners is contested: listening experiments have produced mixed results.^[4:1][5:1]

Definition and related concepts

Magdalena Fuentes and colleagues note that, depending on the musical context, microtiming can take the form of gradual tempo variations, such as rubato or accelerando, or of small deviations of individual events from a regular grid. Studying such deviations requires knowing both where events are expected to fall within the metrical grid and where they actually occur.^[1:4]

A closely related idea is the ethnomusicologist Charles Keil's concept of participatory discrepancies. Keil argued in 1987 that "the power of music is in its participatory discrepancies," which he divided into processual discrepancies (associated with beat, drive, groove, and swing) and textural ones, and that music must be "out of time" and "out of tune" to be personally involving.^[6:1] Vijay Iyer applied theories of embodied and situated cognition to microrhythmic techniques in African-American music, arguing that cognitive universals grounded in bodily experience are tempered by the cultural specificity of how the body is used in performance.^[7:1]

In musical traditions

Jazz swing

In jazz, consecutive eighth notes are typically played as long–short pairs. Measuring drummers' ride-cymbal patterns on jazz recordings, Anders Friberg and Andreas Sundström found that the ratio between the long and short notes (the swing ratio) varied with tempo, from as high as 3.5:1 at slow tempi to 1:1 at fast tempi. The often-cited 2:1 "triple feel" appeared only at one particular tempo, and the short note stayed at about 100 milliseconds across medium to fast tempi. At slow tempi, soloists tended to play their downbeats after the cymbal while staying synchronized with it on the off-beats, which can give an impression of playing "behind the beat".^[2:1]

West African jembe music

Rainer Polak's analysis of jembe drum ensemble music from Mali found that in a piece set in a 12-pulse cycle (often notated as 12/8), each beat is divided into three unequal subpulses, short, flexible, and long, while other pieces use a long–flexible–flexible pattern. Polak argues that these timing patterns distinguish individual meters and are fundamental to the metric system of jembe music, rather than being deviations from it.^[3:1]

Samba and candombe

In Brazilian samba and Uruguayan candombe, timekeeper instruments show microtiming deviations at the sixteenth-note level, while the beat positions themselves are not displaced. Fuentes and colleagues found that the resulting microtiming profiles differ between the two styles.^[1:3]

Shetland fiddle music

In traditional fiddle music from the Shetland Isles, microtiming variations are often called lilt, and they contribute to the music's suitability as an accompaniment for dancing. Analyzing 27 tunes, Estefanía Cano and Scott Beveridge found distinct timing patterns that may characterize lilt, as well as patterns specific to individual performers.^[8:1]

Perception and groove

The relationship between microtiming and the experience of groove is an open research question. Experimental studies, some testing Keil's participatory-discrepancies idea directly, have produced differing results.

• Matthew Butterfield ran two experiments on bass–drum asynchrony in jazz, one using synthetic swing grooves with varied asynchrony and one using recordings of professional musicians; they found little evidence for the participatory-discrepancies framework. He proposed instead that, under metric entrainment theory, such discrepancies have more limited and local effects than previously thought.^[9:1]
• Matthew Davies and colleagues synthesized jazz, funk, and samba patterns with idiomatic microtiming at magnitudes from none to about double the natural level. Contrary to a common belief, systematic microtiming lowered ratings of groove, liking, and naturalness, except for a simple short–long jazz shuffle; the effect was stronger among expert listeners.^[4:1]
• In a study by Jan Frühauf, Reinhard Kopiez, and Friedrich Platz, 93 music students rated a simple rock drum pattern highest when it was perfectly quantized. Early shifts were rated worse than late ones, and shifts on the snare drum worse than on the bass drum. The authors concluded that at least some groove-oriented styles are characterized by an aesthetic of "exactitude".^[10:1]
• Lorenz Kilchenmann and Olivier Senn measured the head movements of 160 listeners to funk and swing recordings with timing discrepancies scaled up or down. Expert listeners moved more to versions with discrepancies reduced by 60% than to fully quantized versions, which the authors describe as partial support for participatory-discrepancies theory; the effect was small to moderate and appeared only among experts.^[5:1]

Taken together, these studies suggest that any effect of microtiming on groove depends on musical style, the size of the deviations, and listener expertise, and that no single direction of effect has been established.^{[4:1][10:1][5:1]}

Computational analysis

Analyzing microtiming in recordings requires estimating both the metrical grid and the actual times at which notes begin. Researchers in music information retrieval infer these from note onsets, tempo, or beat tracking.^[1:4]

This can be sensitive to errors. Christian Dittmar, Martin Pfleiderer, and Meinard Müller found that estimating jazz drummers' swing ratios from automatically detected ride-cymbal onsets was unreliable because small onset-detection errors propagated into the estimates; they instead used a log-lag autocorrelation function, which gave more reliable results.^[11:1] Fuentes and colleagues developed a model that tracks beats and microtiming jointly in samba and candombe recordings, using deep neural networks to estimate beat and onset likelihoods.^[1:1] Cano and Beveridge described their analysis of Shetland fiddle music as a first step toward using MIR techniques to model lilt.^[8:1]