Transforming Cycling Through the Groove: How AI-Quantified Music Boosts Rider Performance and Coordination
Imagine this: You're grinding through a tough cycling session, legs burning, heart pounding—now add the perfect beat, and suddenly your body moves like it's part of a seamless rhythm. Sounds almost magical, right? But here's the game-changer—scientists have uncovered how high-groove music, measured by advanced AI, can literally rewire your cycling coordination. And this is the part most people miss: It's not just about feeling good; it's about your brain and muscles syncing up in ways that could redefine how we train. Ready to pedal into the details? Let's explore this groundbreaking study!
Impact of AI-Quantified Musical Groove on Cyclists' Joint Coordination and Muscle Synergy: Insights from a Repeated Measures Study
Research Article
Open Access (accessible via https://www.springernature.com/gp/open-science/about/the-fundamentals-of-open-access-and-open-research)
Published: 07 November 2025
Journal of NeuroEngineering and Rehabilitation (find more at https://jneuroengrehab.biomedcentral.com/) Volume 22, Article Number: 233 (2025) Cite this article
Abstract
Let's start with the basics: Music isn't just background noise during workouts—it's a powerful tool that connects what you hear to how you move. This connection, known as auditory-motor coupling, can supercharge your exercise performance. But what about the specific vibe of music, like its rhythmic pull or groove? Groove refers to that irresistible urge to move, driven by the beat's drive and sway. Researchers have long wondered how this groove tweaks the biomechanics of cycling, especially under high effort. This study dives deep into that mystery, exploring how varying groove levels—precisely defined by a cutting-edge deep learning model with an impressive accuracy score of R² = 0.85—affect lower body movements and muscle control in intense cycling sessions.
Our team gathered 24 skilled, right-handed cyclists, all young adults averaging 21.1 years old, to pedal under three different setups: a metronome for a neutral baseline, low-groove tunes, and high-groove tracks. We used high-speed 3D motion capture (200 Hz) to track coordination between hips and ankles, as well as pelvis and torso, employing vector coding techniques that break down movement patterns into angles and phases. On the muscle side, surface electromyography (EMG) recorded activity from 12 lower limb muscles, and we analyzed it with non-negative matrix factorization (NMF)—a smart way to unpack how muscles team up in synergistic groups. Think of NMF as a detective that decomposes complex muscle signals into simpler, fundamental patterns, revealing how the nervous system organizes movement. (For beginners, imagine it's like sorting a messy playlist into genre folders—it helps us understand the 'recipes' behind coordinated actions.)
The results? High-groove music stood out. Compared to low-groove or metronome conditions, it boosted hip-ankle in-phase coordination by a striking 28.7% (high-groove: 29.8% vs. low-groove: 23.2%, with a p-value of 0.020, indicating strong statistical significance). It also enhanced pelvis-torso synchronization by 27.1% (high-groove: 38.0% vs. low-groove: 29.9%, p = 0.048). Plus, muscle synergy got more intricate, with an average of 7 synergies in high-groove trials versus 6 in low-groove ones (p = 0.039). A key highlight: The soleus muscle (a calf stabilizer crucial for ankle support) showed much higher activation weights in high-groove scenarios (0.11 ± 0.03 vs. 0.04 ± 0.02, p = 0.030), pointing to better control at the ankle. And here's a unique twist—a new synergy linking the erector spinae (back muscles) with the gastrocnemius lateralis (another calf muscle) popped up in 54% of high-groove trials, hinting at improved trunk-to-limb connections.
In wrapping up, high-groove music fosters smoother, more synced cycling through two main avenues: refined joint coordination (focusing on hip-ankle and pelvis-torso links) and reshaped neuromuscular strategies, including richer synergies and prioritized stabilizer muscles. These biomechanics back up groove-enhanced auditory-motor training, though we'd need more tests on direct performance boosts like energy use or speed to confirm broader benefits. The AI groove measurement tool also pioneers personalized music picks for sports and rehab. But wait, is this all positive? Some might argue over-reliance on music could mask fatigue or distract from form—food for thought!
Introduction
Music's role in exercise is gaining serious traction, especially how it blends senses to boost workouts [1]. For instance, it can ramp up endurance by fine-tuning brain signals that fight fatigue [2]. This magic ties into music's structure, like its pulse, beat strength, and groove, which spark involuntary movements via auditory-motor links [3]. Rhythm itself is a multifaceted beast, blending acoustic cues and brain predictions to sync motions and spark enjoyment [4]. It works through syncopation (surprising beats), beat gradients (how beats build), cognitive engagement, and low-frequency vibes that resonate with our bodies [5]. High-groove tunes light up brain pathways involving the basal ganglia, turning listening into active movement planning via beta-wave rhythms [6]. Past studies often lumped music simply as matching or mismatching BPM (beats per minute) to motion [7], but that's overly simplistic—BPM measures tempo but misses groove's depth [8]. Ignoring groove can lead to inconsistent movement ties, even with the same BPM [9]. Dynamic Systems Theory frames motor-music sync as a complex dance of beat predictability, groove richness, and physical limits [10]. Groove's impact on motion is clear, yet its inner workings are tricky to pin down, mainly because quantifying something so subjective—and tied to acoustics, cognition, and experience—is no easy feat [11]. Groove isn't just BPM; it's a blend of features, needing a holistic eval of temporal, spectral, and personal vibes [12].
Early efforts used behavioral tests, like phase sync measures in finger-tapping to music, creating groove rating systems [13]. But these had flaws: Limited song pools from older eras, plus familiarity biases from memory and dopamine hits. Harmonic simplifications also stripped real-world features like compression or bass depth, hurting realism [14]. AI breakthroughs, especially in music info retrieval (MIR), changed the game. By pulling 21 acoustic traits (e.g., pulse strength, groove flow) and linking them to movement sync, we've hit high accuracy (R² = 0.704) for predicting music-driven feelings [15], paving the way for models that connect sound to motion. In exercise science, muscle synergies and joint coordination offer a dual lens for brain-driven control [16]. Synergies show the nervous system's clever trick: Blending modules instead of micromanaging muscles, simplifying complex tasks under constraints [17, 18]. (For newcomers, picture it as your brain using 'shortcuts' like muscle teams to handle running or cycling without overthinking every step.) This shines in repetitive activities like cycling, where limb muscles fire in rhythmic bursts [19, 20]. Coordination, measured via vector coding of joint phase ties [21], reflects how synergy shapes temporal links, with feedback loops refining activation [22]. (Vector coding, in simple terms, breaks down joint movements into angles that reveal if legs are pushing together or against each other.)
Music's groove nudges the motor system twofold. First, beat strength and bass harmonics sharpen synergy timing through beta-wave sync between hearing and motor brain areas, aided by the basal ganglia [23]. Second, pulse strength locks joints better, like optimizing hip-knee-ankle flows in cycling [24]. Together, these tweak neuromotor control: Synergy shifts boost joint forces, while timing adjustments via feedback polish kinematics [22]. Groove also aids internal-external rhythm matching [25], syncing contractions for smoother control. Plus, by cutting control complexity, music might slash energy waste in muscles and joints, aiding efficiency [26]. Cycling's rhythmic, full-body nature makes it ideal for groove tests [27, 28]. Yet, how groove alters cycling mechanics and muscle ties remains underexplored. This research aims to unpack groove's effects on cycling form, joint sync, and synergy, offering evidence for tailored music in training and therapy.
But here's where it gets controversial: Critics say music might just be a placebo—does it really change your body, or is it all in your head? And could forcing a groove override natural rhythms, risking injury? We'll unpack these debates as we go.
Participants and Methods
Participants
We recruited 24 university students (12 women, 12 men, average age 21.1 ± 1.4 years), all right-handed and experienced exercisers. To ensure solid stats, we ran a power analysis for repeated-measures ANOVA, aiming for effect size 0.01, 95% confidence, 90% power, and moderate correlation (0.5)—needing at least 23 folks. Criteria: Ages 18-25, 3+ weekly workouts of 30+ minutes, music-savvy, bike-ready, and cleared medical hurdles. Exclusions: Cardio/respiratory/neurological issues, recent competitive cycling, past music-exercise combos, fatigue problems, sensory deficits, or meds like antidepressants. Everyone wore standard gear (T-shirt and shorts). We explained risks fully and got signed consent, following Helsinki guidelines, with approval from Beijing Normal University's Ethics Committee (ICBIRB0213_001). Baseline details are in Table 1.
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab1)
Methods
Experimental Design
This repeated-measures setup had participants visit three times, spaced 24 hours apart. Visit one: Baseline on music-free rides (5 N-m torque, 3 minutes, then 30-minute rest) plus a metronome trial as control. Visit two: Low-groove music. Visit three: High-groove music. See the flow in Fig. 1.
Design of the research experiment
Full size image (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Fig1)
For consistency, EMG sensors were reapplied by the same expert each session, skin prepped (shaved, cleaned), and placed per SENIAM standards [29]. This avoided placement errors, ensuring differences stemmed from groove, not setup. We chose 100 BPM deliberately—off participants' usual 65-75 RPM—to enforce rhythm constraint for testing entrainment. It also fits popular music's 90-110 BPM sweet spot [29].
Deep Learning-Based Groove Music Selection and Validation
To compare high- vs. low-groove tracks, we trained an AI model on Janata's dataset: 264 clips rated by musicians and listeners on a 7-point groove scale, plus 21 acoustic features. (Janata's work mixes subjective grooves with objective sounds—check https://doi.org/10.6084/m9.figshare.30217846.v1 for details.) Our process: Build a Temporal Convolutional Network (TCN) for initial features, pre-train a Multilayer Perceptron (MLP) on acoustics, then fine-tune on Janata scores. Model arch in Figs. 2 and 3, SHAP analysis in Fig. 4 [31]. Tenfold cross-validation gave R² = 0.85, proving it rivals human ratings. Top features: Beat patterns, timbre, chords.
We rated 1,280 top December 2024 hits: High-groove (90-110 scores, top 25%), low-groove (30-50, bottom 25%). Picked 11 each, had 10 students rate subjectively. Excluded mismatches, ending with 8 high and 9 low tracks. Acoustic diffs in Fig. 5.
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab2)
Comparison of acoustic features between music with different groove levels. Note: Stats via t-tests or Mann-Whitney, p < 0.05, Cohen's d or rank-biserial. Top 5 from SHAP model. All via Python/SciPy/R.
Full size image (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Fig5)
BPM normalized to 100 via Abelton Live 11. Played through AirPods at comfy volume, starting with music and ending with it. Randomized order. Control: GarageBand metronome.
And this is the part most people miss: AI groove detection boosts reliability, cutting familiarity biases and highlighting acoustic drivers [31]. Traditional methods—manual features or ratings—fall short on complexity, like syncopation interactions or nonlinear elements. Our TCN end-to-end learning captures these, with 21 MIR features and ratings (R² = 0.85) preserving meaning while dodging collinearity. SHAP reveals key predictors (pulse clarity, flux). SHAP helps explain groove [31]. Standardized scoring of modern hits avoids outdated data. This sets a standard for movement-sound links in biomechanics. (For extras, see Appendix 1.) Note: Rhythm is temporal regularity/BPM; groove adds layers like syncopation, micro-timing, dynamics, timbre—same BPM, different groove!
Riding Protocol
Participants synced to groove, aiming to match RPM to BPM (smaller diff = better sync). Target 100 RPM enforced groove effects.
Used Lode Power bike (Fig. 6). Seat height: 88.3% inner leg length. Warm-up: 3 min at 2.5 N-m, 55-65 RPM. 30-s practice to 100 RPM via cues/display (hidden after). Then sequential tasks: Low torque (males 7 N-m, females 4.9 N-m), medium (11/7.7 N-m), high (15/10.5 N-m). Maintain 100 RPM, synced to groove.
We sequenced high/low fixed for logistics—no repeats, so no learning bias.
Measurement and Calculation of Joint Coordination
Recorded last 30 s of 3-min trials via Qualisys (8 cameras, 200 Hz). 25 markers per biomechanical model (Fig. 8). Crank marker for cycle division (Fig. 9). Processed in Qualisys/ Anybody 7.4 for joint angles. Peaks in Z-traj split into cycles.
Vector coding: Angle-angle plots for coupling angles (0-360°), classifying modes (Fig. 10): In-phase (sync same dir), anti-phase (opposite), proximal (upper leads), distal (lower leads). Focused on hip-ankle, knee-ankle, pelvis-torso vertical.
Vector coding intuitively reveals dynamic ties—better than simple plots for periodic moves like cycling [34, 35].
Acquisition and Processing of EMG Data
Delsys system: 1 kHz, 12 muscles (TA, GM, GL, SOL, VM, VL, RF, BF, ST, GMX, RA, ES). Preprocessed: Averaging removal, 20-400 Hz filter, rectification, smoothing (20 Hz). RMS envelope (20 ms windows, 10 ms overlap), normalized to max, time-normalized to 200 points [36]. Only high-torque, low/high-groove analyzed—metronome too different.
Calculation of Muscle Synergy Patterns
NMF decomposed EMG into synergies (W) and coefficients (C(t)), reconstructing as sum (Eq. 1). VAF (Eq. 2) picked optimal synergies (90% VAF). Clustered k-means (1000 reps), Gap stat for classes. Reference: High-groove's clusters. Compared weights/durations (>0.3 threshold).
For details, Appendix 2.
Statistical Analysis
SPSS 26.0: Shapiro-Wilk for normality, boxplots for outliers. Normals: ANOVA (Greenhouse-Geisser if needed), Bonferroni post-hocs, η²p. Non-normals: Friedman, all-pairs. EMG: t-tests or Wilcoxon. p < 0.05. Independent analyses per torque.
Results
Comparison of Joint Coordination in Riding to Music at Different Groove Levels
High-torque: Hip-ankle in-phase sig. differed (F=17.09, p<0.001, η²p=0.43). HG > LG (p=0.020), HG/LG > MT (p<0.001/0.040). Pelvis-torso in-phase: HG > LG/MT (p=0.048/0.041). No sig. diffs low/medium torque or hip-knee/knee-ankle. Tables 3-5.
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab3)
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab4)
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab5)
Comparison of Synergistic Patterns of Riding Muscles Under Different Groove Levels of Music
Total synergies: HG > LG (Z=-2.06, p=0.039). Figs. 11, Table 7. LG: 4 classes, HG: 5 (reference). Table 8: More HG in reference patterns.
Fig. 11: Synergies (SYN1-5), bars=weights (>0.30 marked), lines=activation % cycle.
Full size image (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Fig11)
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab6)
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab7)
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab8)
Table 9: SYN2 SOL weights HG>LG (Z=-2.17, p=0.030). SYN5 HG-only (GL/ES >0.3). SYN1/3/4 no diffs.
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab9)
Table 10: No activation time diffs SYN1-4.
Full size table (view at https://jneuroengrehab.biomedcentral.com/articles/10.1186/s12984-025-01778-7#Tab10)
Discussion
Neuro-Biomechanical Regulatory Mechanisms of Musical Groove on Joint Coordination in High-Load Cycling
Coordination is key in cycling, esp. high-intensity [42]. Groove boosts hip-ankle/pelvis-torso in-phase under high torque (28.7%/27.1% gains), aiding force transfer. This neuromodulation syncs timings, smooths torque, cuts waste [43, 44]. Groove refines control, optimizing for efficiency.
Effects of Musical Groove on Muscle Synergy Patterns in Cycling
HG triggers adaptive synergy shifts for complexity. Total synergies up, more patterns. HG shows trunk-calf links (ES-GL), SOL boost in SYN2 (ankle stability via type I fibers). Functional roles: SYN1 propulsion (GM/GL), SYN2 control (SOL), SYN3 knee/hip (VL/VM/RF), SYN4 stability (ES/GMX), SYN5 power (ES/GL). Increased synergies suggest brain fractionation for sync—reducing degrees of freedom [46-48].
Limitations: Small sample (young athletes), short-term focus, unmeasured familiarity/motivation/fatigue. Future: Larger groups, long-term, psych metrics, direct performance (VO2/HR).
Conclusion
HG music refines cycling via coordination and synergy tweaks, backing groove interventions. AI quantification enables personalization. Mechanisms need more probing.
Availability of data and materials: In manuscript/supplement.
What do you think—could groove music revolutionize your workouts, or does it just add hype? Is there a risk of over-relying on beats? Does this change how we view training aids? Drop your opinions below and let's debate!
References
Timmers R, et al. Neural multimodal integration... Neurosci Lett. 2020;721:134803.
Proost M, et al. How to tackle mental fatigue... Sports Med. 2022;52(9):2129–58.
Stanton TR, Spence C. The influence of auditory cues... Front Psychol. 2020;10(3001).
Snyder JS, et al. Theoretical and empirical advances... Nat Rev Psychol. 2024;3(7):449–62.
Levitin DJ, et al. The psychology of music... Annu Rev Psychol. 2018;69(1):51–75.
Brittain JS, Brown P. Oscillations and the basal ganglia... Neuroimage. 2014;85 Pt 2:637–47.
Karageorghis CI, et al. Ergogenic and psychological effects... Psychol Sport Exerc. 2010;11(6):551–9.
Dixon S, et al. Towards characterisation of music... ISMIR, 2004.
Womelsdorf T, Fries P. Neuronal coherence... J Physiol-Paris. 2006;100(4):182–93.
Reh J, et al. Loudness affects motion... BMC Musculoskelet Disord. 2022;23(1):586.
Large EW, et al. Dynamic models... Front Comput Neurosci. 2023;17(1151895).
Patel AD, Iversen JR. The evolutionary neuroscience... Front Syst Neurosci. 2014;8(57).
Elliott MT, et al. Analysing multi-person timing... Timing and time perception... 2018:177–215.
Collins T, et al. A combined model... Psychol Rev. 2014;121(1):33–65.
Abri F, et al. A comparative analysis... Electronics. 2021;10(20):2519.
Winters JMJHMS. How detailed should muscle models... Hum Mov Sci. 1995;14(4–5):401–42.
Cheung VC, Seki K. Approaches to revealing... J Neurophysiol. 2021;125(5):1580–97.
Latash ML, et al. Toward a new theory... Motor Control. 2007;11(3):276–308.
Wojtara T, et al. Muscle synergy stability... J Neuroeng Rehabil. 2014;11(129).
Raasch CC. Coordination of pedaling... Stanford University; 1996.
Bockemühl T, et al. Inter-joint coupling... Hum Mov Sci. 2010;29(1):73–93.
Dejnabadi H, et al. A new approach... IEEE Trans Biomed Eng. 2008;55(2):755–64.
Phillips-Silver J, Keller PE. Searching for roots... Front Human Neurosci. 2012;6(26).
Mondok C, Wiener M. An entrainment oscillator... bioRxiv, 2024.03.07.583955.
Bacon CJ, et al. Effect of music-movement... J Sports Med Phys Fitness. 2012;52(4):359–65.
Braun Janzen T, et al. Rhythm and music-based... Front Human Neurosci. 2022;15(789467).
SO C-H R. Fatigue of lower limb... 2009.
Symsack A, et al. Usability assessment... Mil Med. 2021;186(3–4):379–86.
Hermens HJ, et al. Development of recommendations... J Electromyogr Kinesiol. 2000;10(5):361–74.
Janata P, et al. Sensorimotor coupling... J Exp Psychol Gen. 2012;141(1):54–75.
Chen J, et al. The effect of perceived groove... Med Sci Sports Exerc. 2025;57(4):857–66.
Li H, et al. Newly compiled tai chi... PeerJ. 2023;11:e15036.
Sarvestan J, et al. Determining the optimum number... J Biomech. 2024;176:112375.
Robbins SM, et al. Inter-segment coordination... Gait Posture. 2024;107:324–9.
Persine S, et al. Transfemoral amputees adapt... Gait Posture. 2023;105:171–6.
Kibushi B. Muscle coordination patterns... Hum Mov Sci. 2023;90:103116.
Baifa Z, et al. Muscle coordination during... Eur J Sport Sci. 2023;23(1):54–61.
Seo G, et al. Developing new intermuscular... J Neuroeng Rehabil. 2023;20(1):112.
Saito H, et al. Muscle synergy patterns... J Neuroeng Rehabil. 2023;20(1):69.
Hafer JF, et al. Changes in lower extremity... J Biomech. 2024;177:112434.
Vendrame E, et al. Muscle synergies in archery... Annual international conference... IEEE Engineering in Medicine... 2021:6220–23.
Choi A, et al. Inter-joint coordination... J Sports Sci. 2016;34(20):1991–7.
Heuvelmans P, et al. Relationships between task constraints... Percept Mot Skills. 2024;131(1):161–76.
Dorel S. Mechanical effectiveness... Biomechanics of training... 2018:33–62.
Bini RR, et al. Lower limb muscle activation... Int SportMed J. 2011;12(1):7–16.
Hankinson K. Can rhythm change the brain?... 2021.
Turpin NA, et al. Upper limb and trunk... J Sports Sci. 2017;35(6):557–64.
Kassiano W, et al. Bigger calves... Int J Sports Med. 2024;45(10):739–47.
Acknowledgements: Not applicable.
Funding: Supported by Shanghai University of Sport's Key Lab and Graduate Innovation Program.
Author information: Haojie Li, Xinyu Lin: co-first. Authors: Haojie Li, Xinyu Lin, Xie Wu. All involved in design/writing.
Corresponding author: Xie Wu.
Ethics: Helsinki-approved, consent obtained.
Consent for publication: Agreed.
Competing interests: None.
Additional info: Neutral on maps.
Rights: CC BY-NC-ND 4.0.
Reprints: Via Springer.
About this article: Cite as Li H, et al. Impact... J NeuroEngineering Rehabil 22, 233 (2025).
Download: Citations/references.
