Attention
Categorical Perception Task
Also called: categorical perception of speech, phoneme categorization, phoneme identification and discrimination, ba-da continuum, same-different discrimination, AX discrimination, speech continuum labeling, phoneme boundary task
Measure how sharply people sort speech sounds: synthetic syllables that step from ba to da, compared in pairs, then labeled one at a time.
8-11 minUntimed
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Try it here first, with no account.
- Papers, 2021 to 2025
- 240
- Evidence grade
- C
- Registered clinical studies
- 2
What it measures
Measures how sharply people sort speech sounds into categories. Through headphones, synthetic syllables step by step from ba to da: participants hear two and say whether they were the same or different, then label each one as ba or da. The main score is the slope of the labels at the boundary.
Constructs
- Speech sound categorization
- Phonetic discrimination
- Auditory perception of speech
Research uses
- Speech perception research
- Reading and language development research, at the group level
- Cross-language and second-language speech research
How the paradigm works
The Categorical Perception Task measures how sharply people sort speech sounds into categories. Its sounds are a continuum of nine syllables that change in equal steps from ba to da. Listeners usually label the steps with a sharp switch from one syllable to the other, and they tell two steps apart better when the steps fall on the two sides of that switch than when both fall on one side, a pattern Liberman, Harris, Hoffman and Griffith (1957) first described. It needs headphones and normal hearing. The test starts with the sound setup: the nine syllables load, the participant turns the device's volume until a steady tone is comfortable, and a six-trial headphone check (Woods, Siegel, Traer and McDermott, 2017) confirms that headphones are in use. By default a failed check ends the test, and its result is saved either way.
More about the paradigm
The first part is same-different discrimination, the task of Pisoni (1973): the participant presses Play, hears two syllables with half a second of silence between them, and chooses whether they were the same or different. Half the trials play two steps two apart on the continuum, in either order, and half play one of those steps twice: the seven pairs from steps 1 and 3 to steps 7 and 9, each in its four orders, twice, 56 trials. The second part is identification: one syllable at a time, labeled ba or da, each of the nine steps ten times, 90 trials. Discrimination comes first so that the labels do not shape it, as Gerrits and Schouten (2004) ran labeling last. Each part opens with four practice trials on the two ends of the continuum, which show the answer; the test itself gives no feedback, has no time limit and does not adapt to the answers. The syllables are synthetic speech, made by formant synthesis in the platform's own code, not recordings of a person, and they are the same sounds in every language: the onsets of the second and third formants and the burst move in equal steps from ba to da, while the vowel, the pitch, the length and the voicing before the release stay the same. Each plays at a fixed level, 12 dB below the volume-setting tone.
The main score is the labeling slope: a logistic curve is fitted to the share of da labels at each step, and its slope where it crosses half says how sharp the switch is, in percentage points per step, with the boundary step beside it. The same-different part gives the share right for each pair, for the pairs across the fitted boundary and within a category, and the difference between them, the discrimination peak. It also compares discrimination with the accuracy the labels alone would allow, 0.5 plus half the squared difference between the two steps' shares of da labels (Pollack and Pisoni, 1971): discrimination well above that prediction shows that differences within a category were heard. Both measures depend on the task. A two-choice labeling slope reflects how consistent the answers are as well as how sharply the categories are heard (Kapnoula and colleagues, 2017), and its size depends on the continuum's steps (McMurray, 2022). Discrimination looks more categorical in tasks that leave listeners to judge against their own criterion, and less biased tasks show little of the peak (Schouten, Gerrits and van Hessen, 2003; Gerrits and Schouten, 2004). Where the boundary falls depends on the listener's language, as phoneme boundaries do (Lisker and Abramson, 1970; Miyawaki and colleagues, 1975), so results are compared within a language and within studies that use this continuum. Little has been published on how stable these measures are from one session to the next (Kim and colleagues, 2025). In research, weaker labeling and weaker discrimination across the boundary have been reported in groups with dyslexia (Noordenbos and Serniclaes, 2015) and developmental language disorder (Robertson and colleagues, 2009). The test does not measure or screen hearing: scores are research measures and do not diagnose.
Scores
- Labeling Slopescore
- Slope where a logistic fitted to the share of da labels at each step crosses half, in percentage points per step; at most 250, labels that switch completely within one step.
- Higher means a sharper switch from ba to da; inconsistent answers also lower it
- Boundary Stepscore
- Step of the continuum, 1 to 9, where the fitted curve crosses half da labels.
- Depends on the listener's language; informational
- Da Labels by Step%
- Share of da labels at each step, keyed by the step.
- Rises from step 1 to step 9
- End Labels Right%
- Share of step 1 labeled ba and step 9 labeled da.
- Under 0.7 is flagged for review
- Same-Different Accuracy%
- Share of same-different trials answered right, over the seven pairs two steps apart, half of them the same step twice.
- Higher means the steps were told apart better; 0.5 is chance
- Same-Different Accuracy by Pair%
- Share of same-different trials right for each pair, keyed by its two steps.
- Usually highest for the pairs across the boundary
- Across-Boundary Accuracy%
- Share of same-different trials right on the pairs whose two steps lie on either side of the fitted boundary.
- Informational
- Within-Category Accuracy%
- Share of same-different trials right on the other pairs.
- Informational
- Discrimination Peakscore
- Across-boundary accuracy minus within-category accuracy, in percentage points.
- Positive means pairs across the boundary were told apart better, the categorical pattern
- Accuracy Predicted from Labels%
- Mean over the pairs of 0.5 plus half the squared difference between the two steps' shares of da labels (Pollack and Pisoni, 1971).
- Informational
- Observed Minus Predicted%
- Mean over the pairs of the share right less the accuracy predicted from the labels.
- Near 0 means discrimination follows the labels; positive means differences within a category were heard
- Median Answer Timems
- Median time from when the answers opened, after the sounds, to the answer, over both parts.
- Informational; the answer has no time limit
- Anticipationscount
- Answers within 100 ms of the answers opening, which leave the scores.
- Half or more of the answers makes the session invalid
- Completed Trialstrials
- Trials answered in the scored run, both parts.
- Fewer than planned is flagged for review
- Planned Trialstrials
- Trials the run set out to play, both parts.
- Informational
- Headphone Check Passed0/1
- 1 when the headphone check passed (at least 5 of 6 right), 0 when it did not.
- 0 stops the test unless the study set it to continue
- Headphone Check Correctcount
- Headphone check trials answered right, of 6.
- 5 or 6 passes
- Reported Output Delayms
- Output delay the device reported to the browser; empty when the browser does not report it.
- Informational
Use and evidence
240 papers from 2021 to 2025 name this task in their title or abstract, down 30% compared with 2016 to 2020. That ranks it 118 of 495 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 50 |
| 2024 | 38 |
| 2023 | 40 |
| 2022 | 54 |
| 2021 | 58 |
| 2020 | 67 |
| 2019 | 78 |
| 2018 | 67 |
| 2017 | 62 |
| 2016 | 70 |
| 2015 | 74 |
| 2014 | 59 |
| 2013 | 67 |
| 2012 | 72 |
| 2011 | 61 |
| 2010 | 70 |
| 2009 | 63 |
| 2008 | 68 |
| 2007 | 40 |
| 2006 | 48 |
| 2005 | 38 |
| 2004 | 40 |
| 2003 | 22 |
| 2002 | 28 |
| 2001 | 27 |
| 2000 | 16 |
Use grade C
100 to 299 papers from 2021 to 2025.
Evidence grade C
At least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 53
- Registered clinical studies using it as an outcome
- 2
Note on the counts: includes color and face studies.
How the grades work
Use: A is 1,000 or more papers from 2021 to 2025, B 300 to 999, C 100 to 299, D 30 to 99 and E fewer than 30. Evidence: A means at least 25 papers report its test-retest reliability and at least 25 registered clinical studies use it as an outcome; B, at least 8 test-retest papers, and either 5 registered clinical studies or 150 reliability or validity papers; C, at least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies; D, little published measurement evidence found. The counts measure use. They include every paper that names the task in its title or abstract. Grades describe the published research on this task, not ConductCognition's version of it.
Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Sound setup
- The syllables load, the participant turns the device's volume until a steady 1 kHz tone is comfortable, then takes a 6-trial headphone check (Woods and colleagues, 2017). By default a failed check ends the test; the result is saved either way.
- Syllables
- Nine synthetic syllables from ba to da, made by formant synthesis in our own code, not recorded from a person, and the same in every language. The onsets of the second and third formants (900 to 1,650 Hz and 2,150 to 2,800 Hz), their transitions (35 to 45 ms) and the burst move in equal steps; the vowel, the pitch, the 450 ms length and the voicing before the release stay the same.
- Discrimination
- First, same or different: two syllables with 500 ms of silence between them, then the choice of whether they were the same or different. The 7 pairs two steps apart, 1 and 3 up to 7 and 9, each as A then B, B then A, and each step twice, all of it twice: 56 trials, half of them the same step twice.
- Identification
- Then one syllable at a time, labeled ba or da: each of the 9 steps 10 times, in blocks that hold every step once, 90 trials.
- Practice
- 4 trials before each part on the two ends of the continuum, showing the answer for 1,000 ms.
- Level
- Every syllable 12 dB below the volume-setting tone, a level relative to the volume the participant set.
- Score
- The labeling slope: a logistic fitted to the share of da labels at each step, and its slope where it crosses half, in percentage points per step, with the boundary step. Same-different accuracy for each pair, across the boundary and within a category, beside the accuracy the labels alone predict, 0.5 plus half the squared difference between the two steps' shares of da labels. Answers under 100 ms leave the scores.
The syllables are synthetic speech, not the recorded or published continua of other studies, and the level is relative to the volume the participant set, so compare results within studies run the same way. Where the boundary falls depends on the listener's language, so results are compared within a language, and no norm is claimed. A two-choice labeling slope also reflects how consistent the answers are. The test needs normal hearing and headphones, and it does not measure or screen hearing. The scores support research and do not diagnose.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: categorical_perception_discrimination and categorical_perception_identification from the scored run, two buttons after the sound has played, with no time limit. Accuracy is judged against chance on the labels of the two ends, right when step 1 is labeled ba and step 9 da (chance 1 in 2); the labels of the steps between and the same-different answers have no answer a listener must get right, since pairs within a category are often heard as the same. Below chance invalidates and not above chance is flagged; one button on 90% of all answers invalidates when the ends are not above chance; answers under 100 ms invalidate when they are at least half. A headphone check that stopped the test invalidates; a failed check with the test run anyway, ends labeled as expected on fewer than 70% of their trials, labels with no boundary, or a run short of its planned trials is flagged for review.
Why these rules
Accuracy significantly below chance (one-sided exact test, p < .05) cannot come from guessing; it means the response mapping was reversed or the task was not followed (Slick, Sherman and Iverson 1999 treat below-chance forced-choice performance as definite response bias). Chance-level accuracy can be a real result in impaired participants, so it is only flagged. One option on at least 90% of answered trials, when the correct options are balanced and accuracy is not above chance, means the answers did not depend on the items (the task counterpart of long-string responding in surveys; Curran 2016). The answers open once the sound has played and have no time limit, so an answer under 100 ms was pressed before they opened, and only a session made mostly of such answers is invalid. Woods, Siegel, Traer and McDermott (2017) let only listeners who passed the headphone check go on, because a listener on loudspeakers hears the sounds differently; a session that ran after a failed check keeps its scores but not its percentile. The ends of the continuum are the clearest ba and da it has, so ends labeled wrongly mean the sounds or the task were not followed, and the boundary and slope of such labels say little.
No percentile. Scores and quality flags only.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| When the headphone check fails | Stop the test | Stop the test, Continue the test |
| Labels per step | 10 | 6 to 12 |
| Discrimination rounds | 2 | 1 to 3 |
Yearly plans can change settings within these limits. Free and pay-as-you-go studies use the standard settings.
Need a setting this test does not have?
Devices and languages
- Untimed test: runs on desktops, laptops, tablets and phones. Needs headphones. Participant devices
- Languages: English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese.
- Available on: Every plan, including the free plan.
- No percentile. Scores and quality flags only. How norms work
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Categorical Perception Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Liberman, Harris, Hoffman and Griffith, 1957; Pisoni, 1973), with the standard settings (when the headphone check fails: Stop the test; labels per step: 10; discrimination rounds: 2). The main score was labeling slope (slope where a logistic fitted to the share of da labels at each step crosses half, in percentage points per step; at most 250, labels that switch completely within one step).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Liberman, A. M., Harris, K. S., Hoffman, H. S., & Griffith, B. C. (1957). The discrimination of speech sounds within and across phoneme boundaries. Journal of Experimental Psychology, 54(5), 358-368. https://doi.org/10.1037/h0044417The paradigm: labeling and discriminating the steps of a synthetic speech continuum.
- Pollack, I., & Pisoni, D. (1971). On the comparison between identification and discrimination tests in speech perception. Psychonomic Science, 24(6), 299-300. https://doi.org/10.3758/BF03329012The discrimination accuracy that the labels alone predict, for same-different and ABX tests.
- Pisoni, D. B. (1973). Auditory and phonetic memory codes in the discrimination of consonants and vowels. Perception & Psychophysics, 13(2), 253-260. https://doi.org/10.3758/BF03214136Same-different discrimination of a synthetic stop-consonant continuum, pairs two steps apart and ten labels per step.
- Gerrits, E., & Schouten, M. E. H. (2004). Categorical perception depends on the discrimination task. Perception & Psychophysics, 66(3), 363-376. https://doi.org/10.3758/BF03194885How the discrimination task changes how categorical the results look, with labeling run last.
- Schouten, B., Gerrits, E., & van Hessen, A. (2003). The end of categorical perception as we know it. Speech Communication, 41(1), 71-80. https://doi.org/10.1016/S0167-6393(02)00094-8Discrimination that looks categorical when listeners judge against their own criterion.
- Kapnoula, E. C., Winn, M. B., Kong, E. J., Edwards, J., & McMurray, B. (2017). Evaluating the sources and functions of gradiency in phoneme categorization: An individual differences approach. Journal of Experimental Psychology: Human Perception and Performance, 43(9), 1594-1611. https://doi.org/10.1037/xhp0000410Two-choice labeling slopes and crossovers as measures of individual listeners.
- McMurray, B. (2022). The myth of categorical perception. The Journal of the Acoustical Society of America, 152(6), 3819-3842. https://doi.org/10.1121/10.0016614The limits of two-choice slopes and discrimination tasks as evidence of categorical perception.
- Lisker, L., & Abramson, A. S. (1970). The voicing dimension: Some experiments in comparative phonetics. In Proceedings of the 6th International Congress of Phonetic Sciences, Prague 1967 (pp. 563-567). Academia. Phoneme boundaries that differ with the listener's language.
- Miyawaki, K., Strange, W., Verbrugge, R., Liberman, A. M., Jenkins, J. J., & Fujimura, O. (1975). An effect of linguistic experience: The discrimination of [r] and [l] by native speakers of Japanese and English. Perception & Psychophysics, 18(5), 331-340. https://doi.org/10.3758/BF03211209Discrimination of a continuum shaped by the listener's language.
- Kim, H., McMurray, B., Sorensen, E., & Oleson, J. (2025). The consistency of categorization-consistency in speech perception. Psychonomic Bulletin & Review, 32(5), 2246-2258. https://doi.org/10.3758/s13423-025-02700-xHow little is known about the reliability of categorization measures.
- Noordenbos, M. W., & Serniclaes, W. (2015). The categorical perception deficit in dyslexia: A meta-analysis. Scientific Studies of Reading, 19(5), 340-359. https://doi.org/10.1080/10888438.2015.1052455Group differences in categorical perception in dyslexia.
- Klatt, D. H. (1980). Software for a cascade/parallel formant synthesizer. The Journal of the Acoustical Society of America, 67(3), 971-995. https://doi.org/10.1121/1.383940The formant synthesis behind the syllables.
- Woods, K. J. P., Siegel, M. H., Traer, J., & McDermott, J. H. (2017). Headphone screening to facilitate web-based auditory experiments. Attention, Perception, & Psychophysics, 79(7), 2064-2072. https://doi.org/10.3758/s13414-017-1361-2The headphone check before the test.
Source papers
- Kapnoula EC, Winn MB, Kong EJ, Edwards J, McMurray B (2017). Evaluating the sources and functions of gradiency in phoneme categorization: An individual differences approach. Journal of Experimental Psychology: Human Perception and Performance, 43(9):1594-1611.N 131, Adults
- O'Brien GE, McCloy DR, Kubota EC, Yeatman JD (2018). Reading ability and phoneme categorization. Scientific Reports, 8:16842.N 44, 8-12
Questions
How long does the Categorical Perception Task take?
About 8 to 11 minutes, as listed in the test library.
What device does a participant need?
Untimed test: runs on desktops, laptops, tablets and phones. Needs headphones. The browser must be Chrome, Edge, Firefox or Safari.
What scores does it produce?
Labeling Slope, Boundary Step, Da Labels by Step, End Labels Right, Same-Different Accuracy, Same-Different Accuracy by Pair, Across-Boundary Accuracy, Within-Category Accuracy, Discrimination Peak, Accuracy Predicted from Labels, Observed Minus Predicted, Median Answer Time, Anticipations, Completed Trials, Planned Trials, Headphone Check Passed, Headphone Check Correct and Reported Output Delay. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Categorical Perception Task used in research?
240 papers from 2021 to 2025 name it in their title or abstract, down 30% compared with 2016 to 2020. That ranks it 118 of 495 test families in the ConductCognition test census (use grade C). Source: ConductCognition test census 0.22.0, counted 3 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Categorical Perception Task have?
In the published research, 0 papers report its test-retest reliability and 2 registered clinical studies use it as an outcome (evidence grade C in the ConductCognition test census).
Does it report a percentile?
No. Scores and quality flags only.
Which settings can a researcher change?
When the headphone check fails (Stop the test, Continue the test), labels per step (6 to 12) and discrimination rounds (1 to 3). Settings can be changed on yearly plans; free and pay-as-you-go studies use the standard settings.
Which languages can participants use?
Participants can take this test in English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese, with its words in the same language. English is on every plan, and the other languages come with yearly plans. Have a native speaker read the machine-translated text before launch.
Is this a clinical test?
No. ConductCognition is for research use only. It runs, scores and exports the test; clinical interpretation is outside the platform.
Every test: the full test list.
Run this test in your next study
Start free with 10 free participants on any test, then use participant credits or a yearly plan.