← Can You Learn It?

Can Adults Learn Absolute Pitch?

Research from 2013 to 2025 shows that adults can develop absolute-pitch abilities under specific conditions — with extended training, with above-average auditory working memory, or with pharmacological intervention. The framing that best fits the evidence isn’t “the window is open” or “the window is closed”. Wong 2025 proposes a sensitive period rather than a critical one: early experience matters enormously, and the ability stays plastic for life, but adult acquisition is much harder than it is for a child. The difficulty is the whole content of the claim.

The pre-2010 consensus: a closed window

For most of the twentieth century, the prevailing view was that AP could only develop in early childhood. Takeuchi and Hulse’s influential 1993 review concluded, under what they call the early-learning theory, that the window for AP closes around ages 5–6 and that adult acquisition was effectively impossible given the evidence then available Takeuchi & Hulse 1993.

By 2005, Levitin and Rogers were still summarizing the field this way: studies suggested AP is acquired before age 9, and no convincing case existed of an adult successfully acquiring it Levitin 2005. That was the textbook position — and their wording is stronger than a paraphrase suggests — “no case exists of an adult successfully acquiring it” — though the line is one the review attributes to Ward (1999), not a conclusion of its own.

Even that review wasn’t as flat as it sounds. In the same pages it notes that adults can be trained, within a week, to attach a label to a single tone and then produce or identify it well above chance — just far less successfully than five- and six-year-olds given the same training. The consensus was never that adults learn nothing. It was that they never get all the way there.

It started changing about a decade later.

What changed

Three lines of evidence for adult acquisition

1. Extended training (Van Hedger 2019, Van Hedger 2015)

The most direct behavioral evidence comes from a 2019 study by Stephen Van Hedger and colleagues, titled — pointedly — Absolute Pitch Can Be Learned by Some Adults. They selected six adults who scored high on auditory working memory and put them through eight weeks of intensive training (32 total hours). After training, two of the six scored between 97.9% and 100% on the Chicago Test of Absolute Pitch — performance on par with people who developed AP naturally as children. At a four-month follow-up neither had practised, and both held their Chicago Test scores (93.8% and 91.7%). On the stricter external tests the picture is mixed: S2’s UCSD score fell from 94.4% to 77.8%, under the strict 85% cutoff — though never missing by more than one semitone, so under the more liberal criteria used in earlier studies S2 "would still be categorized as possessing AP." S5 had never cleared even the loosest UCSF threshold, missing it by 2.75 points — so only one of the two passed every AP measure Van Hedger 2019.

Three caveats the authors themselves stress. The sample was six people, all of them affiliated with the senior author’s own lab, pre-selected for high auditory working memory. Both successes were the strongest performers before training — “the two successful AP learners were performing better before training than the third highest-performing participant after training, suggesting that these participants had some degree of AP at pretest that was refined over the course of training.” And both had begun musical training before age 8, which the authors flag as “an age that would be more compatible with the critical period theory.”

The authors reject the strong version of that reading, though. They name “our successful participants were always ‘AP possessors’” as a non-learning explanation and argue against it: before training, the classifier gave S2 a 0 probability and S5 a 0.01 probability of belonging to the genuine-AP group, and “by all tests of AP and by most theoretic positions about AP, these individuals would never be considered to qualify as having AP.” Their own summary: “we interpret the present experiment as providing strong evidence for adult AP acquisition.” And the sentence usually quoted against the result is concessive, not conclusive: “while the present results cannot refute either the critical period or innate theories of AP acquisition, they suggest that aspects of both theories should be more fully integrated with a skill acquisition theory of AP.”

The 2019 result rests on an earlier 2015 paper by the same group, which showed that auditory working memory strongly predicts AP learning. In a single one-hour session, adults averaged 31.0% on the trained notes, and those in the top band of working memory reached 43.8%. Van Hedger and colleagues make the comparison to the valproate group themselves — and they make it to undercut the drug result, not to support it: one hour of ordinary feedback training, with no drug, matched a week of training on valproate (28.3%) in percentage points — though not in multiples of chance, since chance is 8.3% in one task and 16.7% in the other. So the drug might be acting on working memory rather than reopening any developmental window. “This alternative needs to be ruled out in order to make any definitive conclusions about reopening a critical period of learning” Van Hedger 2015.

So the headline isn’t “adults can learn AP.” It’s closer to: some adults — specifically those with above-average auditory working memory — can develop AP-level performance with eight weeks of focused training, and at least in the small sample tested, the ability holds up months later.

2. Scalable online training (Wong 2025)

The most accessible recent result is Wong and colleagues’ 2025 paper, Learning Fast and Accurate Absolute Pitch Judgment in Adulthood. They had 12 adult musicians complete an optimized eight-week online training protocol. Within the training program, participants averaged 7.08 of the 12 chromatic notes at 90% accuracy or better. On the independent post-test — 144 trials, two timbres, three octaves, no feedback at any point — accuracy on the trained timbre rose from 13.9% to 31.7% (chance is 8.3%), and 2 of the 12 reached performance the authors say would be classed as "true" AP in more than 80% of published papers that use an objective performance-based definition. On the untrained timbre, accuracy did not improve significantly — only the size of the error shrank. If you are wondering whether what you learn on piano carries to other instruments, that is the honest answer so far: partly, and not in accuracy. They framed the result as a challenge to the critical-period assumption, arguing for a sensitive period instead Wong 2025.

The whole protocol ran online, without a lab. But the numbers behind “N=12” matter more than the 12 does: 48 people signed up and 12 finished. Ten dropped out and 25 couldn’t keep the pace, after averaging under six hours of training. The dozen who finished were already unusual — an average of 12.8 years of musical training, and 21.4 hours of practice over eight weeks, unpaid. The authors raise the obvious worry themselves — “it is possible that the excluded participants were those who learned AP less effectively, and therefore the results reported in this study would have over-estimated the AP learning in the population” — and then test it: comparing the two groups across bins of training progress, they found no differences (all ps > 0.21) and concluded that “the risk of over-estimating the AP learning by investigating the included participants only was minimal.”

The attrition ratio still matters to you as a reader: three of four volunteers did not finish 21 hours of unpaid practice over eight weeks. That is a fact about how demanding the protocol is, not evidence that the ones who finished were secretly different.

3. A pharmacological proof-of-concept (Gervain 2013)

The strangest and most provocative line of evidence comes from a 2013 trial by Judit Gervain and colleagues. They ran a double-blind, placebo-controlled crossover study testing whether valproate — a drug commonly used for epilepsy and bipolar disorder — could reopen the critical period for AP learning in healthy adult men. In the first treatment arm, the valproate group identified an average of 5.09 of 18 correctly, versus 3.50 for placebo (p=0.02) — but chance here is 3.00, because the task used six categories, not twelve, and they were spaced two semitones apart. The placebo group was statistically at chance; the valproate group scored about 2.1 notes above chance, and 1.6 above placebo. The labels were invented names, not note names Gervain 2013.

That difference is real but modest. And there are several reasons not to take this as a recipe:

Still, the result matters scientifically: it is the first time any drug has been shown to change AP performance at all. Whether that means a developmental window was reopened is exactly what the Van Hedger group disputes — and the authors of the study itself call it a proof-of-concept, conditioning every generalisation on replication that has not yet been published.

The contradictions worth taking seriously

If the case were one-sided, this section would already be over. It isn’t.

Bongiovanni 2023: trained adults fail at the real test

The most pointed counter-evidence comes from a 2023 study by Bongiovanni and colleagues. They trained 177 adults to detect a single note, C4, among others. Only about a third reported playing an instrument or singing — a figure from the 90 participants whose questionnaire data survived a programming error, not from all 177. Accuracy within the trained context was high, but performance collapsed when the note was tested in other octaves: the discrimination index dropped to d′=0.23 — still above chance (p<.01), but a collapse from the trained context. Their reading is that the participants latched onto pitch height — how high or low the sound is — in addition to pitch chroma rather than exclusively on it — “the strategy we believe the present results demonstrate the participants were using.” A Gaussian mixture model still placed about 60% of the octave-test responses around the correct target. Not memorised audio files either: performance transferred in part to an untrained French horn timbre (d′=0.58 against 1.24 on the trained context, p<.001) — attenuated, but well above chance, and those are different recordings entirely. It was a real perceptual skill pointed at the wrong dimension Bongiovanni 2023.

This is a serious challenge. The question it raises isn’t whether trained adults can score well on a familiar test — clearly they can — but whether what they learned is the same thing a person with naturally-developed AP has.

Their published conclusion is broader and harder than their own experiment: “both present and prior paradigms that train a single target frequency are inappropriate measures of AP learning — even as proof-of-concept demonstrations.” And: “had an octave generalization test not been included in the present study, the conclusion that robust, adult AP learning… may have been erroneously reached.” That verdict cuts both ways — the paper argues the same flaw invalidates Crozier 1997 and Russo 2003, the two child-training studies most often cited for the critical period. It is a methodological broom, not a vote for one side.

One way to reconcile this with Van Hedger 2019 and Wong 2025: Bongiovanni trained a single frequency; Wong trained many notes over eight weeks. (Van Hedger 2015 was a single one-hour session, so it belongs on the "brief" side of that line, not the long one.) And the cross-octave test has been run on those longer protocols. Wong trained and tested across three octaves and reported that participants “learned to attach the same pitch name to all tones sharing the same chroma in three octaves, demonstrating that they learned the chroma of the tones rather than the specific pitch height.” Van Hedger 2015 tested an untrained octave and an untrained timbre — and performance fell when it did, from 36.2% on the trained notes to 21.7% on the generalization test in Experiment 1. Testing for generalization is not the same as finding it intact. Bongiovanni’s own paper acknowledges the point, naming Van Hedger 2015, Van Hedger 2019 and an earlier Wong paper (2020) as studies that “incorporated multiple timbres and octaves” — a 2023 paper cannot cite Wong 2025.

So the gap isn’t that nobody tried. It is that studies with different designs reached different conclusions, and no one has run the head-to-head comparison that would settle which design is measuring the real thing. Whether adult-acquired AP is the same as child-acquired AP remains open on those grounds.

Hedger 2013: plasticity isn’t acquisition

One easily-misread result: a 2013 study by Hedger and colleagues showed that just 45 minutes of listening to music being detuned — lowered at about 2 cents per minute, too slowly for anyone to notice, until it sat 33 cents flat — shifted AP listeners’ perceptual categories. The authors framed it as the first evidence that AP categories in adults can change with listening experience Hedger 2013.

That’s plasticity — AP isn’t fixed in stone — but it’s not the same thing as acquiring AP. The participants already had AP. The study shows AP can be modulated, not that it can be created.

The honest gaps

A short list of things the literature doesn’t yet answer:

  1. Long-term retention. The longest follow-ups on record are Van Hedger 2015’s retest of six participants at about six months (rote performance still above chance) and a single participant from Van Hedger 2019 retested at 16 months, who scored 88.9%. Beyond one person and a year and a half, nothing has been measured.
  2. Why working memory mediates learning. Van Hedger 2015 showed it predicts, but the underlying mechanism isn’t identified.
  3. Why some adults succeed and others don’t. In Van Hedger 2019 the two who succeeded were the sample’s best performers beforehand and both began music before age 8 — the authors offer that as a partial account, hedged (“suggesting that…”), while rejecting the stronger reading that the two already had AP. Wong 2025 looked for the same pattern across training studies and did not find it: one person who acquired AP had scored 0% at pretest, and Wong’s best non-tone-language learner started music at 24. So the honest answer is that nobody has identified what predicts success.
  4. Octave generalization. Bongiovanni 2023 found it largely fails after short training on a single frequency. Wong 2025 did train and test across three octaves and reported that participants held the pitch name across them — so the gap is not that nobody tried, it is that the two studies used different designs and reached different conclusions, and no one has yet run the comparison that would settle it.
  5. Which training components do the work. The protocols differ, and no head-to-head comparison isolates which elements drive learning.
  6. Tone-language speakers. Wong 2025 recruited both groups on purpose — 7 tone-language speakers and 5 non-tone-language speakers, pre-registered separately — but merged them for analysis and never compared the two. What it did report cuts the other way: the single best learner was a Hebrew speaker who started music at 24, which the authors read as evidence that early tone-language exposure “is not an essential condition for learning AP judgment in adulthood.” A direct comparison still hasn’t been run.
  7. Phenotyping standards. As Bairnsfather 2025 documented, AP thresholds vary from 20% to 100% across studies. The same person could be “AP” in one paper and “not AP” in another.

A calibrated answer

So, can you learn absolute pitch as an adult?

Two things are true at once, and most of the confusion comes from picking one.

Pitch perception trains. That part is not in dispute, and it is not small. Adults roughly doubled or tripled their note-naming accuracy in these studies; average error fell from 2.62 to 1.50 semitones in Wong 2025; even the sceptical paper, Bongiovanni 2023, measured real learning within the trained context. If your goal is to hear pitch better than you do now, the evidence says that is achievable, and says so repeatedly.

Absolute pitch, as the field defines it, is a much higher bar — and most people who try do not clear it. In Wong 2025, 48 enrolled, 13 reached the post-test and 12 entered the analysis, and the group averaged 31.7%, well short of the 85.9% mean that Bairnsfather’s review reports for AP groups. One of the twelve finished above that mark (89%); the second reached 78%. Both would count as “true” AP in more than 80% of published papers with an objective performance definition, which is the comparison the authors themselves make. The 2025 systematic review of 160 studies still describes AP as something that “in most studies cannot be reliably trained” — while naming Van Hedger et al. (2019) in the same breath as the exception (“although see…”), and reporting “strong evidence supporting a critical or sensitive period for AP acquisition.”

Does it only work for people who half-had it already? That is the obvious suspicion, and the authors take it seriously rather than dismiss it. In Van Hedger 2019 the two who succeeded were the strongest performers beforehand — and Wong 2025 concedes the same of its own two: they “did show higher AP accuracies than others during pretest (proportion correct = 0.36 and 0.36)”. Their argument is that 36% is less than half of what real-world possessors score, that the two took 18 and 21 hours with “slow and gradual improvement”, and that in another study of theirs someone who started at 0% also got there — “hence there was no evidence suggesting that pretest performance is an important factor in explaining AP training success.” The concession and the argument both belong here. Nor does starting young appear to be required: Wong’s best learner among the non-tone-language speakers began musical training at 24, which the authors read as showing that early onset “was not essential for AP development during adulthood.”

The framing that fits both facts is the one Wong 2025 proposes in writing: not a critical period but a sensitive one, where “experience has a particularly strong effect during early life but the psychological function concerned remains plastic throughout the lifespan. Under this definition, acquiring ‘true’ AP in adulthood should well be possible, albeit more difficult.” Difficulty is the whole content of the claim — and the scale of it is worth stating: children in the comparison studies reached all twelve pitches over one and a half to three years; the adults here trained 21 hours over eight weeks and averaged seven.

Several large questions stay open: whether adult-acquired AP is the same thing as native AP, whether it survives years rather than months, and whether it works for adults without high auditory working memory.

The headlines that say “anyone can learn perfect pitch in eight weeks” aren’t supported by the data. The textbooks that say “no adult can improve” aren’t supported either — and never quite were. The interesting middle is where the science actually is, and it is a genuinely useful place to stand: your ear is trainable, and the label “perfect pitch” is a poor way to measure whether it worked.

Curious where you stand?

absolutepitchpiano.com — pitch-identification exercises with immediate feedback. Most of the studies above trained this way — though their formal tests remove the feedback on purpose, because feedback is an external reference.

Try Classic Pitch →

← Browse all 21 studies