Downbeat delays are a key component of swing in jazz

The paper addresses the long-standing question of what musical and psychoacoustical components create swing in jazz. While the uneven subdivision of quarter notes into long-short eighths (swing ratio) is established, the role of microtiming deviations (MTDs) has been controversial. Some theories emphasize participatory discrepancies as central to swing, while others stress rhythmic accuracy and suggest MTDs can impede swing. The authors adopt an operational definition of swing (expert listeners’ judgments) to test whether systematic MTDs—specifically slight delays of soloist downbeats relative to the rhythm section with synchronized offbeats—contribute positively to perceived swing. The study’s importance lies in resolving conflicting claims about MTDs and identifying timing mechanisms that musicians can use to enhance swing.

Prior work identified the swing ratio as a hallmark of jazz rhythm but left other components unclear. Keil and others proposed that small timing discrepancies among ensemble members (participatory discrepancies) create swing, whereas other researchers argued that timing accuracy is critical and MTDs may hinder swing. Listener-based studies predominantly examined groove (enjoyment and entrainment) rather than swing, often finding that random or systematic MTDs do not enhance groove and sometimes reduce it, with quantized performances rated similarly or better. A previous study by the authors’ group found that random MTDs do not increase swing. Only limited prior work addressed swing directly and anecdotal analyses suggested some soloists delay downbeats, but evidence was inconsistent and datasets small. The authors also distinguish swing from groove: groove is necessary but not sufficient for swing, implying swing has additional components beyond those that induce groove.

The study combines an observational corpus analysis with an experiment using timing manipulations of real performances rated by expert listeners. Observational analysis:

• Corpus: 456 jazz solos from the Weimar Jazz Database containing annotated note onsets and structure. Drum downbeats serve as reference points.
• Measures: For each solo, compute average soloist downbeat delay relative to drum downbeats and mean swing ratio as a function of tempo. Timing is expressed in milliseconds and MIDI ticks (default 960 ticks per quarter note). Conversion: ms = 1000 × ticks × 60 / (tpq × tempo). Swing ratio is the ratio of long to short eighth durations within downbeat–offbeat–downbeat triplets, averaged per piece.
• Additional analyses: Assess variability, tempo dependence, and sub-genre effects (bebop, swing, hardbop). Note that offbeat timings of the drums are not available in the database, precluding direct measurement of offbeat alignment.
• Data quality: Onset annotation RMS error estimated around 20 ms via repeated transcription of a reference piece. Experimental study:
• Stimuli: Live MIDI recordings of solo piano on four pieces (The smudge, Texas blues, Jordu, Serenade to a Cuckoo) with added quantized rhythm section (ride cymbal 8ths, hi-hat on 2 and 4, walking bass in quarters). All tracks first quantized to a grid with an optimized swing ratio guided by corpus results and prior literature; the soloist’s swing ratio was held constant across manipulations.
• Timing manipulations (implemented in Julia with MusicManipulations.jl, rendered to audio):
  1. Quantized original (baseline): Soloist MTDs suppressed; aligned to grid with chosen swing ratio.
  2. Both delayed: All soloist notes uniformly delayed by 85 ticks relative to the rhythm section.
  3. Downbeat delayed: Soloist track first delayed by 85 ticks; then rhythm-section offbeats are synchronized to the soloist’s offbeats, yielding soloist downbeat delays of 85 ticks without changing the soloist’s swing ratio, and increasing the rhythm section’s swing ratio (about 2.46–2.54 in examples) while keeping offbeat synchrony.
• Pieces and swing ratios: Example BPM ~150–175; optimized soloist swing ratios around 1.65–1.86; in the original experiment, Serenade to a Cuckoo yielded an excessively large rhythm-section swing ratio after manipulation (≈2.91) and was excluded from main analyses; a second experiment on that piece with adjusted ratios is reported in Supplementary Results.
• Participants: Expert listeners recruited via conservatories, universities, bands. Primary analysis includes 19 semiprofessional and 18 professional jazz musicians (mean age ~38.6 years; predominantly male). Additional groups (amateurs/non-jazz musicians) were analyzed in supplementary analyses.
• Procedure: For each piece, participants compared all three versions on one page, could switch freely, and rated swing and groove on 1–4 scales (1 = not at all, 4 = very much). If identical ratings were given, participants were asked whether they perceived any differences between versions. Headphones were requested; definitions of groove and relation to swing were provided; no formal definition of swing was given.
• Statistical analysis: Ordinal logistic regression modeling swing ratings as a function of manipulation (quantized original reference; downbeat delayed; both delayed), musician category (semiprofessional reference; professional), and their interaction. Reported odds ratios (OR), 95% CIs, and p-values. Discriminability assessed via ROC curves and area under the curve (AUC) between conditions for each piece; significance inferred if AUC CI excludes 0.5. Additional power and robustness checks performed. Ethics adhered to APA principles; informed consent obtained.

Observational corpus findings:

• Soloists frequently exhibit positive downbeat delays relative to the rhythm section across 456 solos; delays generally decrease with increasing tempo. In ticks, the trend is approximately linear within the studied tempo range.
• Typical magnitude: ~30 ms (~85 ticks), around 9% of a quarter note at ~150 bpm.
• Most cases show positive delays; variability exists across pieces and players; a minority exhibit small or negative delays. Sub-genre analysis did not alter the general trend.
• Soloist swing ratios are smaller than commonly assumed, mostly below 1.5 (i.e., not the triplet 2:1 myth). Trend vs tempo: decreases above ~160 bpm with increasing tempo; for medium-to-slow tempi (<160 bpm), soloist swing ratio tends to decrease with decreasing tempo, opposing trends reported for drummers. Experimental findings (three pieces analyzed: The smudge, Texas blues, Jordu):
• Downbeat delayed versions (soloist downbeats delayed by 85 ticks; offbeats synchronized) received significantly higher swing ratings than quantized originals. • Ordinal logistic regression: Downbeat delayed vs quantized original OR = 7.48, 95% CI [3.19, 17.54], p < 0.001. • Both delayed vs quantized original: OR = 1.30, 95% CI [0.67, 2.54], p = 0.440 (no significant difference). • Musician category: Semiprofessional vs professional OR = 3.94, 95% CI [1.26, 12.31], p = 0.019 (professionals rated lower overall). Interaction shows larger downbeat-delayed effect for semiprofessionals (OR = 7.00, 95% CI [1.32, 37.11], p = 0.022).
• ROC/AUC analyses show significant discriminability favoring downbeat delayed versions: • Downbeat delayed vs quantized original AUCs: The smudge 0.71 ± 0.13; Texas blues 0.70 ± 0.12; Jordu 0.69 ± 0.13 (CIs exclude 0.5). • Downbeat delayed vs both delayed also significant; both delayed vs quantized original not significant (AUCs ≈0.52–0.55).
• Groove ratings show a similar but weaker pattern; effects on swing are stronger, suggesting partial dissociation between swing and groove. Overall conclusion: Slight, systematic downbeat delays of the soloist with synchronized offbeats substantially enhance perceived swing; uniform delays without offbeat synchrony do not.

The study resolves a long-standing controversy by demonstrating that specific systematic microtiming deviations—soloist downbeat delays combined with offbeat synchrony—enhance swing as judged by expert listeners, whereas random or uniform delays do not. The corpus analysis establishes that many jazz soloists naturally employ positive downbeat delays whose magnitude decreases with tempo, and that soloist swing ratios are typically below 1.5, countering the common belief in a triplet (2:1) feel. The experimental findings directly link downbeat delays (with preserved offbeat synchrony) to higher swing ratings, with large effect sizes and robust discriminability across pieces. These results refine Keil’s participatory discrepancies hypothesis by identifying the specific timing pattern that is beneficial and distinguishing it from detrimental random deviations. The work has implications for understanding rhythmic perception, performance practice, and music technology (e.g., improving DAW swing features by incorporating downbeat delays). The effects generalize to some extent beyond expert jazz musicians, though are strongest among expert listeners. The phenomenon studied involves small, non-salient delays (~30 ms) distinct from the laid-back style with much larger, perceptible delays, and participants often reported a felt but not easily verbalized “pleasant friction” between soloist and rhythm section.

By combining large-scale corpus analysis with controlled listening experiments, the paper demonstrates that systematic microtiming deviations in the form of slight soloist downbeat delays, with offbeats synchronized to the rhythm section, are a key component of swing in jazz. The work challenges the triplet-feel myth for soloists and quantifies tempo-dependent trends in downbeat delays and swing ratios. The findings inform pedagogy and digital music production by highlighting the importance of downbeat delays beyond simple swing-ratio adjustments. Future research should extend to ensemble contexts (e.g., big bands), investigate microtiming of other rhythmic values (such as triplets), characterize laid-back timing patterns more precisely, and explore cross-genre generality, given initial observations of minimal or absent downbeat delays in Latin styles.

• Observational analyses average across pieces and players without modeling individual styles, leading to substantial scatter; the database lacks drum offbeat annotations, precluding direct measurement of offbeat alignment.
• Annotation noise: Onset RMS error ~20 ms in manual transcriptions may inflate variability.
• Experimental simplifications: Use of a quantized rhythm section and constrained manipulation conditions; focus on piano solos with many eighth-note pairs; exclusion of one piece due to an ill-chosen swing ratio for the rhythm section.
• Sample size and population: Primary analysis includes 37 expert participants; while effects were robust and supplemented by additional checks, broader populations and live ensemble settings were not directly tested.
• The study targets subtle delays (~30 ms) and does not characterize larger, stylistic laid-back playing in detail.