A song you have never heard before is half over, and you can already tell where the next chord is going. You may not be able to say why. When the music goes somewhere else, you feel it in your chest: a little lift, a little jolt. Take a single wrong note out of a familiar tune and it jumps out at you from across the room. Something in you has been predicting the tune all along.
Music cognition is the science of that something: how minds learn the patterns of music, predict them, enjoy being surprised, and how people become good at playing it. It is also a field full of claims that sound wonderful and have not survived testing, so this piece spends time on what held up and what did not. This piece describes research and health topics; it is not medical or psychological advice. If performance anxiety, pain or hearing problems are troubling you, a clinician or an audiologist is the right person; for emotional distress, findahelpline.com lists local services. The sound features play short synthesised tones: tap Play to start, keep the volume low, and every one also shows its numbers.
In 1956 the music theorist and philosopher Leonard Meyer published Emotion and Meaning in Music, arguing that musical feeling arises from expectation: music sets up a pattern, delays or denies what the listener expects, and the tension and relief that follow are the emotion. Fifty years later David Huron's Sweet Anticipation (2006) gave the idea a modern frame, with five stages in the response to an event: Imagination (expecting), Tension (preparing), Prediction (was I right?), Reaction (a fast gut response) and Appraisal (a slower evaluation), the ITPRA theory. The brain rewards good predictions and is jolted by bad ones, and a clever piece plays with both.
Where do the expectations come from? Largely from statistical learning: listeners absorb, without trying, how often each note follows each other in the music around them. Marcus Pearce's IDyOM model is a research version of this idea, a program that learns melodic patterns from a body of music and predicts what comes next; it has been tested against listeners' reported expectations. The sketch below is a toy version: a model that knows only which note followed which in four short public-domain tunes it has not seen the test tune from.
The tune stops here. Which note would you add next?
Sound isn't available in this reader; the staff and numbers still work.
Tunes (all public domain): Beethoven's "Ode to Joy" theme (1824), the French tune of "Ah! vous dirai-je, Maman" (published 1761), "Mary Had a Little Lamb" (US, 1830), "Au clair de la lune", and "Frère Jacques". For each test tune the model is trained only on the other four, using first-order note-to-note counts with a small smoothing. It is a toy, with far less data than a listener has had. Letter names with fixed-do solfège: C do, D re, E mi, F fa, G sol, A la, B si.
Try the tune first, then a very unexpected note (the B or the high C, say) and read the bits. A surprise is not "wrong": it is what makes the note feel like something. Notice also that the model is not the composer. In Mary the real tune jumps to a note the model rated as unlikely (over 5 bits); composers deliberately do that, and a theory of expectation does not say that expected is good, only that it is felt. What counts as expected also depends on what you have heard. A listener raised on other musical systems has different statistics in their head and different surprises (see Music Traditions of the World, 35/69). The chills and dopamine side of the story is in the easier piece Why Music Moves You (09/01).
Does every society have something we would call music? And does it share any structure? In 2019 Samuel Mehr and colleagues published, in Science, "Universality and diversity in human song". They combined an ethnographic corpus of descriptions of music in a representative sample of societies with a discography of recordings, and used listener ratings, machine summaries and transcriptions. They reported that music appears in every society sampled, that songs are tied to behavioural contexts such as dance, healing, love and caring for infants, and that the variation among songs can be summarised along a few dimensions (they named formality, arousal and religiosity). A song's acoustic features could predict its behavioural context, and tonality (some organisation of pitch around a centre) was widespread. That is universality as a statistical tendency, not a law that every song obeys, and the authors were explicit that the diversity is as large as the shared part.
Rhythm has a clean example. A team led by Nori Jacoby and Josh McDermott, published in Nature Human Behaviour in 2024, tested 39 groups in 15 countries, from large cities to Indigenous communities such as the Tsimane' in Bolivia. Each person heard a random pattern of four beats and tapped it back; the tapping was fed back as the next stimulus, and over several rounds the taps drifted toward the person's own mental favourites. In every group the favourites clustered at simple whole-number ratios between the beat gaps (1:1, 2:1, 3:2 and so on), but which ratios were most favoured differed by group, often matching local music. The authors suggest the bias may help a culture's music stay consistent when it is passed along; that is a hypothesis, not a settled result.
Sound isn't available in this reader; the timeline and numbers still work.
A two-click loop of 1.2 s; the first gap is the slider's fraction of it. Grey marks are the simple ratios 1:2 (0.333), 2:3 (0.4), 1:1 (0.5), 3:2 (0.6) and 2:1 (0.667). A listening illustration, not the published experiment.
Slide to 0.47 and play, then snap: the snapped version should sound steadier, a pattern you could clap along with, while the in-between one feels slightly off. That sense of "nearly regular but not quite" is the mind reaching for a category.
Absolute pitch (AP, "perfect pitch") is naming or producing a note with no reference: hear a car horn and say "that's a B♭". Relative pitch is hearing the distance between notes, which is what nearly all musicians use and what ear training builds (Ear Training & Musicianship, 35/64). The commonly quoted prevalence of AP in the general population is one in 10,000, but encyclopaedic reviews note that this figure may not be accurate; a 2019 review put it at "at least 4%" among music students. It is more common among people who began music training very young, and among East Asian music students fluent in a tone language such as Mandarin: Diana Deutsch and colleagues (2006) compared conservatory students in China and the US, and people of East Asian ancestry raised in North America are less likely to have AP than those raised in East Asia. One idea is a critical period, a window in early childhood; researchers disagree about whether training creates AP or whether the lack of it lets relative hearing take over, and whether adults can learn it (a 2019 study reported some adults performing like lifelong possessors, which remains debated). Family studies suggest some genetic influence.
It is not a superpower. A listener with AP can be confused by music "in the wrong key" or tuned to a different standard: Baroque ensembles often tune A to 415 Hz, about a semitone below 440, which can be disorienting. And playing in tune with others is about relative pitch, which most good musicians have.
Sound isn't available in this reader, so this test can't run here.
Random pitches from C4 to B4 in equal temperament, A4 = 440 Hz, one soft tone at a time. Scores live only on this page. One session on one speaker proves nothing about you, and anyone who has a familiar ringtone or a tuned instrument at hand can use it as a reference. The chance level is 1 in 12.
With no reference, expect to land near 8% (1 in 12). The p-value shows how often pure guessing would match your score. Switch to "Reference C given": now the task is hearing the distance from a known note, and a trained ear does far better. That gap is the difference between the two skills.
"Ten thousand hours" became famous through Malcolm Gladwell's Outliers (2008), which retold research by Anders Ericsson, Ralf Krampe and Clemens Tesch-Römer (1993) on violin students at the Berlin music academy: the top group, by the commonly told version, had accumulated around 10,000 hours of practice by age 20, against fewer for the less advanced groups. Ericsson himself objected to the retelling: 10,000 was an average for one group, not a threshold, and what mattered was deliberate practice, focused on weaknesses with feedback, not simply hours.
How much does practice explain? A 2014 meta-analysis by Brooke Macnamara, David Hambrick and Frederick Oswald found that deliberate practice explained about 26% of the variance in performance in games, 21% in music, 18% in sports, 4% in education and under 1% in professions. That is a lot, and it is also most of the story left unexplained. A 2019 replication attempt of the violinist study by Macnamara and Maitra, with 39 violinists and raters who didn't know each player's practice hours, found that the "good" group had actually logged slightly more practice than the "best" group (about 9,800 hours against 8,200, not a significant difference); Ericsson's camp argued the replication measured practice differently. Reasonable researchers still disagree about how much of the rest is age of starting, teaching, genes, motivation or opportunity. A few cautions: practice is mostly measured by looking back, good players may practise more because they enjoy it, and 21% is an average over studies, not a statement about you.
150 simulated players (invented data, same points each time) with practice and skill drawn so that the chosen share of skill variance is explained by practice. Axes are in standard deviations from the average. The percentages are from Macnamara, Hambrick and Oswald (2014); the picture is an illustration, not their data.
At 21% the correlation between practice and skill is about 0.46. Someone who practises one standard deviation more than average (more than 84% of players) is expected to be at about the 68th percentile in skill, and about 70 out of 100 such players are above average, so 30 are not. Practice matters; it is just far from a guarantee.
In 1993 Frances Rauscher, Gordon Shaw and Katherine Ky reported in Nature that 36 college students did slightly better on a spatial-reasoning task (paper folding and cutting items from the Stanford–Binet test) after ten minutes of Mozart's Sonata for Two Pianos in D major (K. 448) than after silence or relaxation instructions. The effect lasted about 15 minutes. The press turned it into "Mozart makes babies smarter": in January 1998 Georgia's governor Zell Miller proposed a state budget item to give every newborn a classical music CD, playing Beethoven's "Ode to Joy" for legislators and asking "Now, don't you feel smarter already?"
It did not hold up. A 2010 meta-analysis of 39 studies by Jakob Pietschnig, Martin Voracek and Anton Formann found little evidence for a Mozart effect, and studies linked to the original researchers reported effects more than three times larger than others. Kristin Thompson, Glenn Schellenberg and Gabriela Husain (2001) showed that the boost followed arousal and mood: any energetic, enjoyable listening could produce it, and it was not about Mozart or about intelligence. Music lessons over years are a separate question with their own mixed evidence. The deeper lesson is statistical: small studies are noisy, and the surprising ones are the ones that get noticed and written up.
Each study: n people each do the task after Mozart and after silence; the score difference is drawn from a normal distribution with the true mean shown. "Significant" is a one-sample t-test at p < .05, two-sided, with textbook critical values. Invented data from your device's random numbers, so each run differs.
Leave the true effect at zero and run it a few times with 36 people. Most of the 40 dots scatter around zero, but most runs also show one or two that land well to the right and reach "significance" (about 2.5% of studies do so by luck alone), and each one looks like a finding. With 40 labs and no effect, the chance that at least one gets a headline result is 1 − 0.97540 ≈ 64%. Move to 400 people and the dots tighten around the truth. Real effects, even small ones, then stand out; fake ones don't.
Many musicians know the dry mouth, shaking hands and racing mind that come before an audition or solo. Music performance anxiety is common, and estimates of how common vary widely with how it is defined and who is asked, so any single percentage should be treated with caution. Researchers separate the physiological arousal itself from how a person interprets it, and from the effect on the playing. Fine motor control on a difficult passage is more fragile than loud, broad playing.
The standard picture is the Yerkes–Dodson "inverted U": performance rises with arousal up to an optimum and then falls. Handle it with care. The original 1908 study by Robert Yerkes and John Dillingham Dodson used mice (the "dancing mouse") and found that the best level of stimulation depended on how difficult the task was. The tidy bell curve that appears in textbooks is a later popular version, criticised for leaving out that simple tasks may keep improving with arousal, and the causal story has never been settled. Treat it as a useful metaphor, not a measured law.
Illustrative curves, not measurements: simple tasks rise and level off, moderate tasks peak near the middle, difficult tasks peak earlier. Real arousal can't be read off a slider, and people differ. It describes; it does not treat anything. For distress, see a clinician (findahelpline.com lists services).
What the research tradition describes about performance anxiety, without being advice: it is common even in professionals; it is shaped by preparation, the setting and how the performer interprets their own arousal; and clinicians have evidence-based approaches, such as cognitive-behavioural therapy, that are for a clinician to suggest. Some musicians use prescribed medication such as beta-blockers; that is a medical decision for a doctor and is not discussed here.
Making music is physical work, and the research on musicians' bodies is mostly about three problems. This is a description of what studies report, not medical advice; any pain, numbness or hearing change is a reason to see a clinician.
| Problem | What it is | What studies say |
|---|---|---|
| Playing-related musculoskeletal disorders | Pain and strain in the neck, shoulders, arms and back linked to playing | Reviews give very wide prevalence (for orchestral players, point prevalence of 57–68% in a 2016 review; ranges of 26–93% in a 2006 one), mostly because studies define the problem differently. Heavy practice time and posture are commonly named as risks. |
| Hearing | Noise-induced hearing damage and tinnitus from loud rehearsal, amplified music and close-to-brass seating | It follows the same dose rule as any noise: about each 3 dB doubles the damage per hour (see the hearing budget in Music Technology & Production, 35/73). Musician's earplugs reduce level evenly across frequencies. |
| Focal dystonia | A loss of control of one or a few fingers or the mouth, only when playing | Uncommon but career-altering; often linked to overuse and heavy repetitive practice. Pianists Leon Fleisher and Gary Graffman are well-known cases; some returned to playing, some adapted, for example to left-hand repertoire. Multidisciplinary care is described as the most successful approach. |
A thread through all three is that musicians are workers whose bodies and ears are the instrument. That is why teachers talk about warm-ups, breaks and listening levels, and why research on performers sits alongside research on listeners in the same discipline.
Music is learned patterns, felt as prediction. It shares statistical tendencies across the world, rare gifts like absolute pitch are less magical than they sound, practice explains a real minority of skill, and the most famous claims (the Mozart effect, the neat 10,000 hours) shrink under testing. The honest scientific habit, ask how big, how many studies, and did it replicate, is the useful thing to take away.
Try this: this week, listen to a song you know well and try to hum, before it arrives, the next note at a key moment such as the end of a chorus. Then notice a moment where the song does something you did not predict. Where did the surprise come from: the melody, the chord, the rhythm or a missing beat?
Sources. Meyer (1956) and Huron (2006): the standard works on expectation, with Huron's ITPRA summarised from the book's well-known framework; IDyOM (Pearce) described briefly from its published description. Mehr and colleagues, Science 366 (2019): abstract-level summary; exact sample sizes and the listener-experiment details were not confirmed and are left out. Jacoby et al., Nature Human Behaviour (2024): 39 groups in 15 countries, confirmed from press summaries of the paper. Absolute pitch figures (1 in 10,000 questioned; "at least 4%" in a 2019 review; Deutsch et al. 2006; Baroque A = 415 Hz): encyclopaedia summary. Practice: Macnamara, Hambrick and Oswald (2014) percentages and the Macnamara and Maitra (2019) replication from encyclopaedia and press summaries; the exact 1993 practice-hour figures vary between retellings and are given as "around 10,000". Mozart effect: Rauscher et al. (1993), Pietschnig et al. (2010), Thompson et al. (2001), Zell Miller (1998): encyclopaedia summary. Yerkes–Dodson (1908): encyclopaedia summary. Playing-related disorders: systematic-review summaries (2006, 2016); focal dystonia: encyclopaedia summary. t critical values from standard tables. The interactive numbers (model probabilities, binomial tail, simulated correlations) were computed in node. Not medical advice.