🧠 Learning Fast and Accurate Absolute Pitch Judgment in Adulthood

📋 Study Overview

🎯 Research Question

Earlier studies had already reported adults learning to name pitches. The question here was narrower and harder: does that learning survive once you close the loopholes that could explain it away?

The authors list six such loopholes in previous work — learning pitch height rather than pitch chroma, holding a reference tone in working memory, relative-pitch strategies helped by small intervals between trials, passing a final level by luck across unlimited attempts, pre-existing AP in the best performers of earlier studies, and selecting participants for high auditory working memory. In their words, the goal was to examine the learnability of AP judgment in adulthood “after addressing the issues described above.”

🔬 Methodology

Participants

48 people enrolled; 12 reached the final analysis. Ten withdrew (averaging 5.1 hours of training) and 25 could not keep up with the pace the protocol demanded (averaging 3.8 hours). Thirteen accumulated the 10 hours needed to be invited to the post-test, and one of those was excluded after the Zoom recording showed her listening to reference tones on an iPad during a break.

• The 12 analysed: trained musicians, mean age 27.8 (range 19–44), averaging 12.8 years of musical training. Seven were Cantonese speakers (a tone language), five were not.
• Pre-existing ability: none had AP at the level of a real-world possessor — pre-test accuracy ran from 0% to 36% (mean 13.9%; chance is 8.3%). Three participants finished with all 12 pitches, and only two of those three started at the top of that range — the authors report “no evidence suggesting that pretest performance is an important factor in explaining AP training success.”
• No cognitive screening — deliberately, since selecting for high working memory was one of the loopholes the study set out to close.
• No control group. The authors argue one was unnecessary because AP is not known to appear spontaneously even after decades of musical training.

Training Protocol

Self-paced online training over 8 weeks, 288 levels, adding one pitch at a time. Participants trained an average of 21.4 hours (range 12.6–26.5) across roughly 15,300 trials. Five safeguards correspond to the loopholes above:

• Three octaves (C4–B6), so that naming a pitch could not be done by pitch height alone
• No feedback in the later levels, and a descending Shepard tone before them to wipe any sample tone from working memory
• Adjacent trials in the pre/post tests always more than an octave apart, which — with the short response window — reduced the effectiveness of relative-pitch strategies
• Out-of-bound pitches presented alongside the trained set, so answers could not be reduced to a relative judgment of “highest versus lowest tones within the training set”
• Finishing required passing the final level four times, with at least 12 hours of no training between the third and fourth — not one lucky attempt

Measurement

Two different measures, which are easy to confuse and which give very different numbers.

• Inside the training program: how many pitches a participant could name at ≥90% accuracy, without feedback, within a response-time window. This is where the figure “7.08 of 12” comes from.
• The independent pre/post test: 144 trials — 12 pitches × 3 octaves × 2 timbres (trained piano, untrained guitar) — with no feedback at all, adjacent trials more than an octave apart, and a 5-second response limit. This is where the figure “31.7%” comes from.
• Chance level: 8.3% (1 in 12). A semitone error counts as an error, in both measures.

📊 Key Findings

1. High Accuracy Achieved

2. The learning was slow and gradual

The “fast” in the paper’s title refers to how quickly trained participants respond (1.3–2.0 seconds), not to how quickly they learn. The two who reached all 12 pitches took 18 and 21 hours, and the authors describe their curves as showing “slow and gradual improvement.”

Nor was this faster than earlier attempts: previous training studies used 12–40 hours and produced 7.2–9.4 pitches. This one used 21.4 hours and produced 7.08 — just below the floor of that range, though under a considerably harder test (three octaves, no feedback). What is new here is the rigour of the test, not the speed of the result.

3. Two participants reached AP level; ten did not

Ten of the twelve improved numerically in accuracy and eleven of twelve reduced their error. At the group level the gains were statistically reliable for accuracy and error, but only near-significant for response time.

• S1 and S2 finished at 89% and 78% accuracy, with mean errors of 0.42 and 0.31 semitones. The authors note this would count as “true” AP in more than 80% of published papers that use an objective criterion. Both had also started at the top of the pre-test range (36% each).
• A third participant passed every training level but failed the post-test. He reported afterwards that he had listened to the sample tones and used relative pitch to “legally game the training system.” With no reference available at test, his accuracy collapsed — a useful demonstration that passing the levels and having AP are not the same thing.
• The other ten finished well below S1 and S2.

4. Timbre generalisation was only partial

Training used piano tones; the test also included guitar. Mean error fell for both timbres, but accuracy improved only for the trained timbre — not for the untrained one. Whatever was learned did not fully transfer to a new instrument.

💡 Main Conclusions

"The ability to learn fast and accurate AP judgment in adulthood in all participants, ranging from a few to all 12 pitches, challenges the widely endorsed notion that the critical period constrains the development of AP." — Wong et al., 2025 (p. 1685)

Key Implications:

• Gains are real but partial. Accuracy roughly doubled on the independent test (13.9% → 31.7%, chance 8.3%) and mean error fell from 2.62 to 1.50 semitones. Two of twelve reached AP-level performance.
• A sensitive period, not a critical one. This is the authors’ own framing, and it is more careful than “age doesn’t matter”: early experience has a particularly strong effect, but the ability stays plastic through life, so acquiring AP as an adult should be possible — in their words, “albeit more difficult.” They also leave the question open, writing that whether a sensitive period exists “should be clarified by matched methods and measurements of AP learning between young children and adults.”
• It contradicts the two-component model. Levitin and Rogers (2005), also in this collection, held that almost everyone can perceive isolated tones but only a few can attach labels. Here every participant learned to name at least some pitches quickly and accurately. See that study →
• Early musical experience still matters. The authors keep it: experience “has a particularly strong effect during early life.” But they make two separate negative claims — that early onset of musical training is not essential (the participant who learned most began music at 24), and that childhood exposure to a tone language is not necessary either.
• What was learned is disputed. Bongiovanni et al. (2023), in this collection, closed the octave test and found adult learning largely collapsing. These two studies are not measuring the same thing, and the field has not settled which design is right.

🌟 Why This Study Matters

How this sits against earlier work in this collection:

• Takeuchi & Hulse (1993) concluded that earlier training attempts were “not strong evidence that AP can be learned by adults” — and admitted that a rigorous comparison between children and adults was “still lacking.” That comparison is what Wong et al. are still asking for, 32 years later.
• Van Hedger et al. (2019) reported two adults reaching AP. Wong et al. cite it critically: those two already scored 60–70% before training, and the study selected participants for high auditory working memory.
• Bongiovanni et al. (2023) reached a more sceptical conclusion from a larger sample, and argued that training a single target frequency is an inappropriate way to measure AP learning at all.

⚠️ Limitations & Future Directions

Study Limitations

Most of these are raised by the authors themselves.

• Three-quarters of enrolled participants did not reach the analysis. The authors take this seriously: they compared the progress of included and excluded participants over the first 6,000 trials and found no difference — and concluded that “the risk of over-estimating the AP learning by investigating the included participants only was minimal.” The statistical support for that conclusion is thin, though: the Bayes factors (0.34–0.62) are only anecdotal evidence for no difference.
• Timbre generalisation failed in accuracy — the gain did not carry to the untrained instrument.
• Relative pitch can never be fully ruled out. The authors say there is no known method that eliminates it completely; they minimised it rather than removing it.
• No control group, by design, on the argument that AP does not appear spontaneously.
• No retention test. Durability after training ends was not measured.
• Four departures from the pre-registration, including merging the tone-language and non-tone-language groups and changing how much training was required to reach the post-test.
• No agreed definition of “true” AP. The authors point out that the literature has no consensus on what performance level counts, which makes the classification of their two best performers arguable by definition.
• Small, unpaid, volunteer sample recruited online during the pandemic. The analysis code was not made public; the data are on the OSF.

Future Research Needed

• Larger sample sizes to confirm findings
• Longitudinal studies tracking retention over months/years
• Comparison of different training methodologies
• Investigation of individual factors predicting success
• Transfer to musical performance contexts

🔗 Related Research

• Complementary findings: Bongiovanni et al. (2023) found adult learning that transfers poorly across octaves (d′ = 0.23) and argued single-frequency paradigms mismeasure AP learning — note they also criticise this group's earlier work (2020b) for using a single timbre and octave
• Historical context: Deutsch et al. (2006) established tone language effects, but assumed critical period was essential
• Earlier pessimism: Levitin & Rogers (2005, p. 29) wrote that "no case exists of an adult successfully acquiring [AP]" — a position now challenged by post-2015 training studies

📖 Access Full Study

📚 Full Citation