EP44

EP44: Microtiming & Groove

Swing Ratio, Participatory Discrepancy, and the Perceptual Threshold Window

▶ 3:30 Music CognitionSignal ProcessingStatistics

前置知识

EP13 Polyrhythm and Rhythm-Pitch Duality

Overview / 概述

Put every drum hit exactly on the grid — zero error per beat.

How does it sound? Stiff. Mechanical. Lifeless.

Shift those same hits by twenty or thirty milliseconds, and groove emerges. This episode asks: what is the mathematics of that shift? Three tools answer the question — the swing ratio, the participatory-discrepancy curve, and the perceptual-threshold window.

The central insight is that musical timing is not a contest to minimize error. It is a precisely engineered deviation from the metronomic grid. A deviation of zero produces a drum machine. A deviation that is purely random produces instability. The narrow corridor between those two extremes — deviations that are structured, directional, and bounded within roughly 10–50 ms — is where groove lives.

中文: “把鼓点精确对齐在网格上，每一拍零误差。听起来怎样？死板，机械，没有生命。但偏移二三十毫秒——律动感来了。今天用三个数学工具，量化这种「不完美」。”

The three tools unify into one framework: the swing ratio specifies the direction of deviation; the participatory-discrepancy curve specifies the shape of deviation across a cycle; and the threshold model specifies the boundary beyond which deviation ceases to feel like groove and becomes instability. Together they replace the folk intuition of “feel” with a quantitative language that can be measured, compared, and designed.

Prerequisites / 前置知识

Polyrhythm and Rhythm–Pitch Duality (EP13) — integer ratios, LCM/GCD, and the rhythmic grid that microtiming deviates from

Definitions

Definition 44.1 (Metronomic Grid)

Fix a tempo $T_{\text{BPM}}$ (beats per minute). The metronomic grid is the sequence of ideal beat onsets

$G_k = k \cdot \frac{60}{T_{\text{BPM}}} \quad \text{seconds}, \quad k = 0, 1, 2, \ldots$

For subdivisions (e.g., eighth notes at subdivision level $n$ ), the grid points are $G_{k,n} = (k/n) \cdot (60/T_{\text{BPM}})$ . A completely quantized performance places every attack exactly at some $G_{k,n}$ .

Worked example. At 120 BPM, the beat grid places onsets at 0, 0.500, 1.000, 1.500 s. The eighth-note grid adds 0.250, 0.750, 1.250, 1.750 s. Any attack not at one of these positions constitutes a microtiming deviation.

Definition 44.2 (Swing Ratio)

In a swing or shuffle feel, pairs of nominally equal subdivisions are performed with unequal durations. The swing ratio is

$r = \frac{d_{\text{long}}}{d_{\text{short}}}$

where $d_{\text{long}}$ is the duration of the first (lengthened) subdivision and $d_{\text{short}}$ is the duration of the second (shortened) subdivision within one beat.

$r = 1$ : perfectly even subdivisions (straight feel).
$r = 2$ : hard triplet swing — the long note occupies exactly two-thirds of the beat, the short note one-third.
$1 < r < 2$ : the continuous spectrum of swing feels used in practice.

Jazz musicians typically operate in the range $r \approx 1.2$ to $r \approx 1.7$ , varying with tempo and expressive intent. Swing is not a binary switch between “straight” and “triplet” — it is a continuous knob.

Worked example. At 120 BPM, one beat lasts 0.500 s. With $r = 1.5$ , the beat divides into $d_{\text{long}} = 0.300$ s and $d_{\text{short}} = 0.200$ s. The “off-beat” eighth note lands 300 ms after the downbeat rather than the straight-feel 250 ms — a shift of 50 ms forward relative to the straight grid.

中文: “定义摇摆比率，等于长音符时值除以短音符时值。比率一是直拍，比率二是硬三连音。但真实的爵士乐手在比率约一点二到一点七之间滑动——快慢段落各有不同。摇摆不是开关，是连续旋钮。”

Definition 44.3 (Microtiming Deviation)

Let $a_k$ denote the measured onset time of the $k$ -th attack in a performance, and let $G_k$ denote its corresponding ideal grid time. The microtiming deviation at beat $k$ is

$d(k) = a_k - G_k \quad \text{(seconds)}$

Positive $d(k)$ means the attack is late (laid back); negative $d(k)$ means the attack is early (pushed).

The deviation sequence $\{d(k)\}_{k=1}^{N}$ over an $N$ -beat passage is the primary object of microtiming analysis.

Worked example. A drummer’s snare hits at (in ms relative to a 120-BPM grid): $d(1) = +18, d(2) = +22, d(3) = +17, d(4) = +21$ . The mean deviation is $\bar{d} = +19.5$ ms and the variance is small — a systematic, consistent layback.

Definition 44.4 (Participatory Discrepancy (PD) Curve)

(Charles Keil, 1987.) The participatory-discrepancy curve of a performance is the time-ordered deviation sequence

$\text{PD}(k) = d(k), \quad k = 1, 2, \ldots, N$

viewed as a signal. Three qualitative regimes are distinguished:

Quantized: $\text{PD}(k) \approx 0$ for all $k$ (drum-machine output).
Structured: $\text{PD}(k)$ has recognizable patterns that repeat with the metric cycle (bar period). This is the groove regime.
Random noise: $\text{PD}(k)$ is uncorrelated white noise with no cyclic structure — perceived as rhythmic instability, not groove.

Keil’s insight: groove arises not from minimizing $|\text{PD}(k)|$ , but from the structure of the PD curve. Structured deviation is participatory — it invites the listener to lean into the rhythm.

中文: “一九八七年，查尔斯凯尔提出：律动感来自偏差，不是来自精确。他叫它参与性差异。偏差若是纯随机噪声，那只是不稳。但好的律动中，偏差有结构——和小节周期同步，每个循环重复相似的偏差模式。”

Definition 44.5 (Perceptual Threshold Window)

(Kilchenmann & Senn, 2015.) Empirical listening studies identify three zones of deviation magnitude $|d(k)|$ :

Zone	Magnitude range	Perceptual effect
Sub-threshold	<span class=“math-inline” data-latex="	d
Groove window	<span class=“math-inline” data-latex=“10 \lesssim	d
Instability zone	<span class=“math-inline” data-latex=”	d

The groove window $[10, 50]$ ms defines the operational space of microtiming. Boundaries shift with tempo: faster tempos narrow the window because the same deviation represents a larger fraction of the beat duration.

中文: “低于约十毫秒，人耳难以察觉。十到五十毫秒之间，偏差被感知为律动感——不在网格上，但不觉得错。超过约五十毫秒，听感被评价为不稳定——具体边界因风格速度而异。律动的操作空间极窄。”

Main Theorems / 主要定理

Theorem 44.1 (Swing Ratio Continuum)

Let a beat of duration $\tau$ be subdivided into two unequal parts with ratio $r \geq 1$ . The off-beat deviation relative to the straight-eighth grid is

$\delta_{\text{swing}}(r) = \frac{\tau}{2} \cdot \frac{r - 1}{r + 1}$

At $r = 1$ : $\delta_{\text{swing}} = 0$ (straight). At $r = 2$ : $\delta_{\text{swing}} = \tau/6$ (hard triplet). The function $\delta_{\text{swing}}(r)$ is strictly increasing and continuous on $[1, \infty)$ .

Proof.

The long subdivision has duration $d_{\text{long}} = r \tau / (r+1)$ and the short subdivision $d_{\text{short}} = \tau / (r+1)$ , since they must sum to $\tau$ and maintain ratio $r$ .

The straight-eighth grid places the off-beat at $\tau/2$ after the downbeat. In the swung version the off-beat lands at $d_{\text{long}} = r\tau/(r+1)$ .

The deviation is

$\delta_{\text{swing}}(r) = \frac{r\tau}{r+1} - \frac{\tau}{2} = \tau \left(\frac{r}{r+1} - \frac{1}{2}\right) = \frac{\tau(2r - r - 1)}{2(r+1)} = \frac{\tau(r-1)}{2(r+1)}$

Monotonicity: $\partial \delta_{\text{swing}}/\partial r = \tau/(r+1)^2 > 0$ for all $r$ . At $r=2$ : $\delta_{\text{swing}} = \tau/6$ , confirming the hard-triplet result. $\square$

Numerical check. At 120 BPM, $\tau = 0.500$ s. Typical jazz swing at $r = 1.4$ : $\delta_{\text{swing}} = 0.500 \times 0.4/4.8 \approx 41.7$ ms. This falls near the top of the groove window — consistent with the characteristic “lilt” of medium-tempo jazz.

Theorem 44.2 (PD Structured-vs-Random Decomposition)

Let $\{d(k)\}_{k=1}^{N}$ be the deviation sequence over $N$ beats, and let $P$ be the metric period (number of beats per bar). Decompose

$d(k) = \mu_p + \varepsilon(k)$

where $p = (k-1 \bmod P) + 1$ is the position within the bar, $\mu_p = \mathbb{E}[d(k) \mid \text{position } p]$ is the systematic component (position-dependent mean), and $\varepsilon(k) = d(k) - \mu_p$ is the residual. Then:

Groove regime: $\|\{\mu_p\}\| \gg \|\{\varepsilon(k)\}\|_{\text{rms}}$ — the systematic pattern dominates the noise.
Instability regime: $\|\{\varepsilon(k)\}\|_{\text{rms}} \gg \|\{\mu_p\}\|$ — the noise dominates; deviations appear structureless.
Mechanical regime: $\|\{\mu_p\}\| \approx 0$ and $\|\{\varepsilon(k)\}\|_{\text{rms}} \approx 0$ — the drum machine.

Proof.

The decomposition

d(k) = \mu_p + \varepsilon(k)

is a one-way ANOVA model with factor “metric position.” By construction,

\mathbb{E}[\varepsilon(k) \mid p] = 0

. If the metric position explains most of the variance — i.e.,

\eta^2 = \text{SS}_{\text{between}} / \text{SS}_{\text{total}}

is large — then deviations are structured by the bar cycle. A random (white-noise) sequence has

\eta^2 \approx 0

by definition, since the position label carries no predictive power. The three regimes follow by inspection of the magnitudes.

\square

中文: “屏幕上三种风格，三条不同的偏差曲线。有结构的偏差 ≠ 随机误差。”

Theorem 44.3 (Layback as Positive-Mean Low-Variance PD)

A layback groove is characterized by a deviation sequence with:

Positive systematic component: $\mu_p > 0$ for all metric positions $p$ (every attack is late).
Low residual variance: $\text{Var}(\varepsilon) \ll \bar{\mu}^2$ .
Magnitude in the groove window: $10 \lesssim \bar{\mu} \lesssim 50$ ms.

The signature distinguishes layback from random lateness (high variance) and from pushes (negative mean).

Proof.

The three conditions encode distinct geometric properties of the distribution of

d(k)

. Positive mean rules out pushes; low variance rules out random instability (the distribution concentrates around a positive center); magnitude in the groove window rules out sub-threshold deviations and out-of-window errors. These three conditions jointly constrain

d(k)

to a distribution with support predominantly in

[10, 50]

ms and small spread — exactly the observable signature of laid-back performers such as Questlove (D’Angelo’s Voodoo album). The three constraints are independent and each eliminates one confounding class.

\square

中文: “以巫毒专辑的鼓点为例：鼓手系统性地将击打延后约二三十毫秒——每一拍都往后靠。偏差分布：均值为正，方向高度一致——每一拍都往同一方向偏移。这就是后靠律动。刚好在律动窗口内，精确地慵懒。”

Prop 44.1 (Tempo Scaling of the Groove Window)

If the groove window is

[d_{\min}, d_{\max}]

ms at beat duration

\tau_0

, then at a new tempo with beat duration

\tau

, the window scales to approximately

[d_{\min} \cdot (\tau/\tau_0), d_{\max} \cdot (\tau/\tau_0)]

in units of the beat fraction, though the absolute millisecond bounds remain roughly fixed due to peripheral auditory processing.

Proof.

The JND for time intervals scales proportionally to the interval duration at short timescales (Weber’s law for time:

\Delta t / t \approx \text{const} \approx 0.04

for durations above ~100 ms). For a beat of duration

\tau

, the just-noticeable deviation is

\approx 0.04\tau

. At 120 BPM (

\tau = 500

ms), this gives ~20 ms, consistent with the lower threshold. The upper bound at ~50 ms is set by categorical perception: beyond this, the brain reclassifies the timing as a distinct rhythmic event rather than a deviation. Both bounds have weak tempo dependence in the 60–200 BPM range studied by Kilchenmann & Senn (2015).

\square

Numerical Examples

Swing Ratio at Various Tempos

At tempo $T$ BPM, beat duration $\tau = 60/T$ s. The off-beat deviation for a swing ratio $r$ is $\delta = \tau(r-1)/[2(r+1)]$ . Converting to milliseconds:

Tempo (BPM)	$\tau$ (ms)	$r = 1.2$	$r = 1.5$	$r = 1.7$
80	750	45.5 ms	75.0 ms	95.5 ms
120	500	30.3 ms	50.0 ms	63.6 ms
160	375	22.7 ms	37.5 ms	47.7 ms

Observation: at 80 BPM with $r \geq 1.5$ , the deviation exceeds 50 ms and risks entering the instability zone. This is why slow jazz ballads typically use lower swing ratios ( $r \approx 1.2$ ) while medium-tempo bebop allows higher ratios ( $r \approx 1.5{-}1.7$ ).

Layback Profile: D’Angelo Voodoo (Illustrative)

The following values are community-derived empirical descriptions, not peer-reviewed measurements. They illustrate the mathematical structure of a layback groove.

Suppose a 4-beat bar with $P = 4$ and measured deviations (ms):

\{d(k)\} = \{+22, +28, +25, +20, +23, +27, +26, +21, \ldots\}

Systematic component estimates: $\mu_1 \approx 22, \mu_2 \approx 28, \mu_3 \approx 25, \mu_4 \approx 21$ ms.

All four are positive (laid back on every beat). Residual standard deviation $\approx 2.5$ ms — small relative to the mean of 24 ms. The groove ratio $\bar{\mu}/\sigma_\varepsilon \approx 9.6$ , strongly structured. The mean magnitude 24 ms lies well within the $[10, 50]$ ms groove window.

中文: “不同律动风格排在一条量化光谱上。一端是完全量化的鼓机，另一端是完全自由的人类演奏。凯特拉纳达的醉态量化在中间偏量化的位置，爵士摇摆更偏自由，迪安杰洛的后靠律动最自由。不是好坏之分，是设计选择。”

The Three-Component Unification

The three tools of this episode map onto a clean decomposition of the deviation signal:

Tool	What it specifies	Mathematical object
Swing ratio $r$	Deviation direction (late vs. early, and how much)	Mean systematic component $\bar{\mu}$
PD curve $\{\mu_p\}$	Deviation shape across the bar cycle	Position-dependent mean profile
Threshold window $[d_{\min}, d_{\max}]$	Deviation boundary (groove vs. instability)	Feasible set for $\bar{\mu}$

中文: “三个工具，一个框架。摇摆比率是偏差的方向，参与性差异曲线是偏差的形状，阈值模型是偏差的边界。”

Musical Connection / 音乐联系

音乐联系

The Mirror of EP13

EP13 (Polyrhythm and Rhythm–Pitch Duality)

showed that integer ratios create rhythmic order: two voices with period ratio 3:2 produce a polyrhythmic pattern that repeats every LCM(3,2) = 6 beats. Integer ratios between onset times are the skeleton of rhythm.

EP44 is the mirror: departing from integer ratios — by deviating from the rational grid — creates groove. The swing ratio $r = 1.4$ is not a simple integer fraction; it is an irrational-ish number chosen by the performer’s ear. The participatory discrepancy lives precisely between the grid points that EP13 identified as structurally meaningful.

This is not a contradiction. The grid provides the reference frame without which deviation has no meaning. Groove requires the skeleton of EP13 and then bends it.

Three Archetypal Groove Styles

The quantization spectrum places three canonical styles:

Full quantization (drum machine): $d(k) = 0$ for all $k$ . No groove; maximum metric precision. Used in early electronic dance music (Roland TR-808, 1980).
Drunk quantization (Kaytranada): Deviations pushed slightly off-grid in a style-specific, repeating pattern. The groove is imposed by the DAW’s “humanize” function or by manual nudging. $\bar{\mu} \approx 10{-}20$ ms, low variance.
Layback (D’Angelo Voodoo, drummer Questlove): Consistent positive deviations of ~20–30 ms across every beat. The whole pocket sits behind the grid. Feels relaxed, heavy, and “in the pocket.”

Why Structured Deviation Feels Good

One hypothesis from music cognition: the listener’s motor system entrained to the beat continuously generates a prediction of the next onset time. When the actual onset is systematically slightly late — but in a repeating, predictable pattern — the prediction error has a regular structure that is processed as a “pull” or “lean.” The brain interprets this lean as an invitation to move, which is the subjective experience of groove.

This is formalized in the Dynamic Attending Theory (Jones, 1976): attention pulses synchronize to periodic rhythmic events. Microtiming deviations that are predictable within the cycle keep attention synchronized while introducing pleasurable micro-tensions.

The Narrowness of the Groove Window

The groove window of approximately 10–50 ms is remarkable: it spans less than one twentieth of a second. Within this narrow corridor, the difference between a mechanical performance and an emotionally compelling one can be a matter of 15 ms of systematic delay. No other parameter in music — not dynamics, not pitch intonation, not timbre — operates in such a tight quantitative range with such large perceptual effect.

Limits and Open Questions / 局限性与开放问题

Threshold universality. The 10–50 ms groove window is derived from Western music listeners in controlled laboratory conditions (Kilchenmann & Senn, 2015). Whether these boundaries are universal across musical cultures and rhythmic traditions — Afrobeat, Hindustani classical, Brazilian samba — is an open empirical question. Micro-timing studies in non-Western traditions are sparse.
Individual and instrument variation. Different instruments have different onset detectability. A kick drum has a sharp transient and a JND of ~2–5 ms; a bowed string has a gradual onset and a JND of ~15–25 ms. The groove window derived for percussion may not transfer directly to melodic instruments. The threshold model needs instrument-specific calibration.
Interaction between musicians. This episode models a single performer’s deviation sequence. In an ensemble, the PD curves of different instruments interact: the bassist may be slightly ahead of the drummer, creating a composite groove that no single performer generates. Multi-voice microtiming analysis requires modeling the covariance structure $\text{Cov}(d_{\text{bass}}(k), d_{\text{drums}}(k))$ , not just marginal means.
DAW quantization and “humanization." Digital audio workstations apply swing ratios and random offsets algorithmically. Whether algorithmic humanization is perceptually equivalent to authentic human timing at the same deviation statistics is contested. Participants in listening tests sometimes distinguish “fake” humanization from real performances at matched deviation levels, suggesting the PD curve has features beyond mean and variance.
Neural correlates. The perceptual threshold window presumably reflects a neural timescale — possibly the duration of a cortical oscillation cycle in the beta band (~20 Hz, period ~50 ms). The coincidence between the 50 ms upper groove boundary and the ~50 ms period of 20 Hz oscillations is suggestive but not established as causal.

Conjecture (Groove Window Tracks the Beta Oscillation Period)

The upper boundary of the perceptual groove window (~50 ms) is not an arbitrary psychoacoustic limit, but tracks the period of cortical beta-band oscillations (~20 Hz), which are known to entrain to rhythmic auditory stimuli. Specifically: deviations shorter than one beta period ( $\lesssim 50$ ms) are assimilated into the ongoing oscillatory prediction; deviations longer than one beta period disrupt entrainment and are perceived as errors.

Falsification criterion: A study that manipulates the listener’s beta oscillation frequency (e.g., via transcranial alternating-current stimulation at varying frequencies) and observes a corresponding shift in the upper groove threshold would support this conjecture. If the upper threshold is fixed at ~50 ms regardless of induced beta frequency, the conjecture is falsified.

Academic References / 参考文献

Keil, C. (1987). Participatory discrepancies and the power of music. Cultural Anthropology, 2(3), 275–283.
Kilchenmann, L., & Senn, O. (2015). Microtiming in swing and funk affects the body movement behavior of music expert listeners. Frontiers in Psychology, 6, 1232.
Friberg, A., & Sundström, A. (2002). Swing ratios and ensemble timing in jazz performance: Evidence for a common rhythmic pattern. Music Perception, 19(3), 333–349.
Iyer, V. (2002). Embodied mind, situated cognition, and expressive microtiming in African-American music. Music Perception, 19(3), 387–414.
Madison, G. (2006). Experiencing groove induced by music: Consistency and phenomenology. Music Perception, 24(2), 201–208.
Pressing, J. (2002). Black Atlantic rhythm: Its computational and transcultural foundations. Music Perception, 19(3), 285–310.
Senn, O., Kilchenmann, L., von Georgi, R., & Bullerjahn, C. (2016). The effect of expert performance microtiming on listeners' experience of groove in jazz, funk, and rock. Music Perception, 33(4), 489–510.
Butterfield, M. (2010). Participatory discrepancies and the anacrusis in jazz. Music Theory Spectrum, 32(2), 157–177.
Jones, M. R. (1976). Time, our lost dimension: Toward a new theory of perception, attention, and memory. Psychological Review, 83(5), 323–355.
Fitch, W. T., & Rosenfeld, A. J. (2007). Perception and production of syncopated rhythms. Music Perception, 25(1), 43–58.
Benadon, F. (2006). Slicing the beat: Jazz eighth-notes as expressive microrhythm. Ethnomusicology, 50(1), 73–98.
Danielsen, A. (Ed.) (2010). Musical Rhythm in the Age of Digital Reproduction. Ashgate.
Lerdahl, F., & Jackendoff, R. (1983). A Generative Theory of Tonal Music. MIT Press. (Ch. 2 for metric grid theory.)
London, J. (2012). Hearing in Time: Psychological Aspects of Musical Meter (2nd ed.). Oxford University Press.
Grosche, P., Müller, M., & Sapp, C. S. (2010). What makes beat tracking difficult? A case study on Chopin Mazurkas. Proceedings of ISMIR, 649–654.