What each test measures, how it does it, the auditory science behind it, what it outputs, and — honestly — where it can be wrong.
Browser-based auditory psychophysics & tinnitus phenotyping · not a diagnostic instrument or a substitute for calibrated audiometry
The one caveat that colours everything below. A web browser cannot produce a known sound-pressure level: the actual level at the eardrum depends on the device, the DAC, the headphones and the OS volume, none of which the page can read. So every level here is relative — measured against a comfort ceiling the user sets, or against the listener's own best threshold — not calibrated dB HL. The shapes and relationships (notches, slopes, pitch-vs-loss, before/after changes) are informative; the absolute numbers are not clinical.
Current scope. The present implementation is designed for exploratory, repeated single-subject measurements. It is not validated for population-level inference or clinical decision-making, and one person's result does not validate any mechanism.
Measurement spine. Every session has a versioned ID and an ordered event stream (requested → applied → observed), a freshly reconfirmed safety state, device capabilities, repeated momentary check-ins and bounded Sound Lab observations. A monitored master bus reports mixed digital RMS/peak/clipping in dBFS; operating-system speech is outside it. History remains local unless explicitly exported, emailed or consented to the configured service. None of these engineering values measures acoustic dB SPL at the ear.
0 Signal & level foundations
Shared machinery every test is built on, via the Web Audio API.
Relative level model substrate
Measures
Sound level as attenuation below a user-set comfort ceiling, not absolute SPL.
Method
A tone/noise passes through a gain node; level in "relative dB" maps to gain by g = ceiling · 10(dB − 75)/20 over a 75 dB working range. The ceiling is fixed by a 1 kHz reference the user raises to "loud but comfortable".
Basis
Amplitude scaling: a change of N dB is a factor 10N/20 in amplitude (20·log₁₀). Correct as a relative scale; the absolute anchor is unknown.
Output
All thresholds/loudness are "dB relative" — later re-referenced to the listener's own best.
Limits
Not dB HL/SPL. Device- and headphone-dependent; only meaningful within one session on one setup.
Coloured-noise generators substrate
Measures
—
Method
White via a seeded LCG; pink (−3 dB/oct) by the Paul-Kellet IIR approximation; brown by leaky integration; blue by differencing; speech-shaped by band-passing pink. Used for maskers, therapy and the speech-in-noise masker.
Basis
Perceptually, pink/speech-shaped noise matches natural spectra far better than white, which is harsh and fatiguing.
Limits
Buffer approximations, not laboratory-exact spectra.
1 Pure-tone audiometry
A staged, per-ear threshold search across the standard audiometric frequencies (250 Hz–8 kHz), delivered as a numbered pipeline because the order matters.
1.1 Comfort-ceiling calibration setup
Measures
The loudest level the test is allowed to use.
Method
The listener raises a slider on a 1 kHz tone to "clearly audible, never uncomfortable"; that gain becomes the ceiling, and the 75 dB range sits below it.
Basis
A safe, self-set upper bound in place of a calibrated audiometer output.
Limits
Subjective; the true SPL of the ceiling is unknown, so it anchors a relative scale only.
1.2 Device / channel check verification
Measures
Stereo routing (L/R correct, not reversed or leaking) and the audio sample rate.
Method
A tone panned hard-left then hard-right; the listener names the side. Sample rate is read from the audio context and gates extended highs (Nyquist).
Basis
Per-ear testing is only valid with true stereo separation; a reversed or leaking channel would misattribute an ear.
Limits
Confirms routing, not acoustic isolation — no microphone means real inter-aural attenuation and headphone response are unmeasured.
1.3 Warbled-chirp stimulus stimulus
Method
Each presentation is 3 pulses (~220 ms on / 180 off) of a sine, frequency-modulated ±4% at 5 Hz (a warble), at a fixed level.
Basis
Pulsed/warble tones are standard in clinical audiometry: pulsing aids detection and, critically, distinguishes the probe from tinnitus; warble avoids standing-wave nulls. A fixed level per burst means a slow or late button press cannot inflate the threshold (unlike a rising-ramp method).
1.4 Bayesian adaptive threshold core measure
The heart of the audiometry — a probabilistic (ZEST-inspired) estimator, not a fixed staircase.
Measures
The detection threshold at each (ear, frequency) — and its uncertainty.
Method
Each track holds a posterior over threshold θ on a grid (0–75 dB, 2.5 dB). A logistic psychometric function P(hear | level, θ) (spread ≈6 dB, 3% lapse) updates the posterior after every yes/no. The next chirp is placed at the posterior mean; a tone finishes when the posterior SD is small (or a trial cap). New tones borrow an informed prior from the listener's age, neighbouring frequencies and the other ear.
Basis
Maximum-likelihood / Bayesian adaptive psychophysics (ZEST/QUEST family): posterior-mean placement is efficient and the informed prior exploits the known smoothness of threshold curves. It is ZEST-inspired — placement is at the mean, not by expected-information-gain (the Ψ method).
Output
Threshold (relative dB) plus a 95% credible interval and trial count per frequency — not a bare integer. Simulation: ~6 trials/frequency, ~1.8 dB mean error.
Limits
Relative, not dB HL; no contralateral masking, so a large asymmetry could reflect cross-hearing; posterior width depends on the assumed psychometric shape.
1.5 Shuffled interleave & catch trials bias control
Measures
Response bias / false-alarm rate, alongside the thresholds.
Method
Tones are presented in a randomised order across ear/frequency/level (never the same pitch twice running), with randomised silent gaps. ~12% of trials are silent catch trials; a "heard it" then is a false alarm.
Basis
Signal-detection theory: measuring hits without false alarms confounds sensitivity with a "say-yes" criterion. Randomisation defeats anticipation and rhythmic guessing.
Output
Per-ear false-alarm rate feeds the response-quality flag.
1.6 Extended high frequencies exploratory
Measures
Thresholds at 10 / 12.5 / 14 / 16 kHz (optional).
Basis
Extended-high-frequency thresholds can detect changes not captured by conventional audiometry, but their interpretation depends strongly on transducer calibration and measurement quality.
Limits
Most consumer headphones roll off sharply above ~6–8 kHz, so an apparent EHF loss often reflects the hardware, not the ear — shaded and labelled exploratory, and gated on sample rate.
1.7 The hearing profile (audiogram) visualisation
Method
Thresholds are self-referenced to the listener's own best result (best = 0), plotted with 95%-CI whiskers; red ○ right, blue × left; no-response frequencies break the line. Optional overlays: an illustrative age-related shape (ISO-7029-style) and an estimated tonotopic (Greenwood cochlear-place) axis x = log₁₀(f/165.4 + 0.88)/2.1 — a coordinate estimated from frequency, not a measured cochlear location.
Basis
Self-referencing gives a meaningful "relative to your own best" read without pretending to dB HL. The Greenwood function maps frequency to position along the basilar membrane — the tonotopic view the tinnitus question turns on.
Limits
Called a "hearing profile", not an audiogram; the age curve is shape-only, not a calibrated comparison.
1.8 Reliability & test–retest quality
Measures
How much to trust the run, and its repeatability.
Method
Response quality summarises false alarms + median credible-interval + trials per ear. Test–retest re-runs the thresholds and reports the mean absolute difference vs the first run, overlaying it dashed.
Basis
An instrument should report when it doesn't know; repeatability is more informative than any single number.
Limits
Good/Fair/Poor bands are product heuristics, not validated clinical cut-offs (hence "response quality" and "repeatability", not "reliability").
2 Speech-in-noise
A functional measure — understanding speech in noise — that pure-tone thresholds miss.
Digits-in-noise (adaptive) functional measure
Measures
The signal-to-noise ratio at which the listener can just report spoken digits — a speech-reception-threshold analogue.
Method
Three spoken digits (device speech voice) over speech-shaped noise. SNR follows a 1-up/1-down track (all-three-correct → harder); the noise amplitude is scaled by 10−SNR/20 (speech fixed). A larger 4 dB step shrinks to 2 dB after the first reversal; the threshold is the mean of the reversal SNRs (first discarded).
Basis
The Digits-in-Noise paradigm is calibration-free and correlates with pure-tone audiometry; SNR is 20·log₁₀(RMSspeech/RMSnoise), and scaling the noise is a validated "fix-speech" mixing method. It provides a functional speech-in-noise measure that can reveal difficulties not apparent from pure-tone thresholds alone.
Output
A relative SNR threshold (SRT-equivalent). Simulation: recovers a known SRT to ~0.8 dB.
Limits
The speech voice's RMS is unmeasured, so the SNR is relative (true = value + an unknown constant), and text-to-speech is not the validated recorded corpus.
3 Tinnitus psychoacoustics
Characterising the percept — pitch, quality, loudness, maskability — and phenotyping its subtype. These are measurements of the percept, not a diagnosis.
Pitch match psychoacoustic
Measures
The frequency that best matches the tinnitus.
Method
A slow sweep from 500 Hz (documented low bound) rising over ~60 s; the listener taps at the match. Repeated up to 3× and combined by geometric mean. Each match is then octave-confirmed on a live fine-tuner (oscilloscope + spectrum + ÷2/×2 jumps + A/B against silence).
Basis
Tinnitus pitch matching is notoriously noisy and can exhibit octave errors, so the test explicitly checks octave alternatives; best practice is multiple matches combined (a Bayesian-sequential idea) and explicit octave testing.
Output
Matched pitch (Hz) with the spread across passes as an uncertainty.
Limits
Inherently variable even in clinic; a probe tone can transiently alter the tinnitus during matching.
Sound-quality audition psychoacoustic
Measures
The perceptual character (pure tone / ringing / hiss / buzz / low roar / cricket).
Method
The listener auditions synthesised exemplars and picks the closest.
Basis
Tinnitus is often narrowband noise rather than a pure tone; character informs the phenotype and the masker choice.
Loudness match → sensation level psychoacoustic
Measures
How loud the tinnitus is relative to the listener's own threshold.
Method
A tone at the matched pitch is set to equal the tinnitus loudness; the result is expressed as dB SL (Sensation Level) = matched level − threshold at that pitch.
Basis
A raw level is meaningless uncalibrated; sensation level above threshold is the interpretable quantity, and tinnitus is typically only a few dB SL.
Limits
A loudness-match dB is not a loudness VAS; both are approximate on uncalibrated hardware.
Minimum masking level (MML) psychoacoustic
Measures
The lowest noise level that just covers the tinnitus.
Method
Narrowband noise around the pitch is raised until the tinnitus is no longer heard.
Basis
A standard tinnitus-assessment metric; also useful for setting up a masker.
The lowest stimulation level that produces temporary post-stimulus suppression (residual inhibition).
Method
Triangular-envelope narrowband bursts (0.5 s on / 0.4 off) around the pitch, with an adaptive level search; after each train the listener reports gone / quieter / same / louder.
Basis
Residual inhibition is a well-studied phenomenon; a triangular envelope gives comparable results with more comfort than hard-gated bursts.
Limits
RI is variable and can interact with pitch matching; a single run is exploratory.
Questionnaire & auditory symptom profile structured history
Measures
Laterality, pulsatility, somatic (jaw/neck) modulation, fluctuation/onset, and a middle-ear symptom cluster (flutter, fullness, sound-triggered, swallow/yawn).
Basis
These separate red-flag and pattern categories: pulsatile → possible vascular; unilateral → asymmetry check; somatic → musculoskeletal; the middle-ear cluster records a symptom pattern that may warrant middle-ear / somatosensory assessment. It records patterns; it does not identify a middle-ear pathology.
Output
Prominent flags (e.g. urgent if sudden onset + asymmetric profile) and an ENT/audiology steer.
Phenotype classifier rule-based
Method
Presents observed characteristics → a neutral phenotype descriptor → clinical flags, with any aetiological associations clearly hedged as "possible, not causes".
Basis
A phenomenological questionnaire cannot infer pathology; the literature (and NICE) caution against treating psychoacoustic tinnitus features as a clean map to a cause.
Limits
A measurement summary, explicitly not a diagnosis.
Impact index subjective burden
Measures
How much the tinnitus affects daily life (0–100).
Method
Ten items scored 0–10, averaged ×10, banded Negligible→Severe.
Basis
Psychoacoustic pitch/loudness correlate weakly with distress; a validated-style burden score captures what matters for management.
Limits
Original wording modelled on validated instruments (e.g. the Tinnitus Functional Index), not the copyrighted questionnaire itself — for care, a clinician should administer the validated form.
4 Mechanism-probing experiments
The scientifically interesting layer: relating the tinnitus to the hearing loss, and probing whether external sound changes it. All at comfortable levels only.
Pitch vs hearing-loss geometry derived
Measures
Where the tinnitus pitch sits relative to the loss.
Method
Computes the loss edge (first frequency ≥15 dB worse than best), the octave displacementlog₂(pitch/edge), the high-frequency slope (dB/octave, 1 kHz→top), and both estimated tonotopic positions (Greenwood cochlear-place function — estimated from frequency, not measured on the cochlea).
Basis
Tests the tonotopic/central-gain account: the literature finds pitch usually sits within the loss and often >1 octave above the edge, not at it. So it names a leaning — pitch at the edge → Model A; pitch well above → an emergent/compensatory frequency (Models B/C).
Limits
Correlational and single-subject; suggestive, never confirmatory.
Where, across the neighbourhood of the pitch, external sound quietens or worsens the tinnitus.
Method
Comfortable narrow bands from ~½× to 2× the pitch; after each, the listener reports the change.
Basis
Frequency-specific residual inhibition; suppression near the pitch supports an RI relationship and can guide a masker.
Limits
Order effects and RI carry-over between trials; comfortable levels only — never used to provoke.
Probe & masker (modulation) tool
Method
A live, tweakable sound — tone / narrowband / notched / pink / speech — with AM tremolo, FM warble, beating, notch width and a master low-pass, updated smoothly.
Basis
Lets the listener match the tinnitus character (pulsing, buzzing, warble) beyond pitch, and set up a comfortable masker; a notch removes energy at the pitch (experimental stimulus generation, not a validated therapy).
Prospective personal A/B comparison experiment
Measures
The immediate after-minus-before change in one outcome chosen in advance: tinnitus prominence, perceived loudness, or distress.
Method
Two Sound Lab conditions are compared over three balanced pairs (six periods). A computer-generated bit assigns AB or BA within each pair. Duration and digital gain remain fixed; later allocations and interim results stay concealed. A minimum gap and a user-confirmed settled state guard against immediate carryover.
Reporting
Condition means, usable observations, missing outcomes and early stops are shown descriptively after all six periods. The generated sequence and actual completion state remain in the raw export, following the transparency principles of CENT 2015.
Limits
The sounds are recognisable, so this is not blinded. Six observations cannot separate expectation, time trends, autocorrelation or longer carryover from a real condition effect. A recent tinnitus crossover trial found carryover despite a planned washout, illustrating this risk (Quemar et al., 2026).
Spectral stimulus reactivity experiment
Measures
Whether spectral structure — not loudness — changes the tinnitus.
Method
Controlled "degraded audio" conditions from a harmonic source: A full band → B telephone band (300–3400 Hz) → C band + harmonic peaks → D codec-like (band + peaks + soft distortion + a "boxy" speaker peak). Each is played comfortably; the app computes its spectral descriptors (centroid, flatness, crest, harmonic-to-noise ratio, peak list, a stored 48-bin spectrum) and records the tinnitus change before → after → a 5/15/30/60 s recovery curve. The listener's own recording can be loaded through the codec chain.
Basis
A behavioural test of the "frequency-selective excitation → nonlinear processing → persistent percept" hypothesis (H2→H3): does a spectral signature predict a change, rather than "did a tone help?"
Limits
Hypothesis-explorer, not proof; the codec→tinnitus chain is unestablished. Deliberately capped at comfortable levels — never reproduce a past exposure to re-trigger symptoms.
Sound-sensitivity & recovery log provocation-free
Measures
Self-reported sound pain/discomfort, tinnitus reactivity, trigger and recovery duration.
Method
No sound is played. Reporting pain or uncertainty activates a session interlock that removes the optional research sound stimuli.
Output
A structured log for export, not an autonomic measurement. HRV/pupillometry remain part of the separate native-iPhone integration plan and are not claimed by this browser.
Tinnitus spectrum, RI function & tuning curve research battery
Measures
A frequency-by-frequency tinnitus-likeness spectrum; residual-inhibition depth and duration by masker centre frequency; and the relative masker level needed to just cover the tinnitus.
Method
Randomised likeness ratings across tone/noise frequencies, comfortable 20-second narrow-band RI probes, and a multi-frequency masker adjustment task. The single-pitch match is retained separately.
Limits
Tinnitus matching and masking vary within a person; order/carry-over and uncalibrated transducers limit comparisons. These curves describe this session only.
Frequency discrimination spectral resolution
Measures
The smallest detected pitch difference around a 500 Hz, 1, 2 or 4 kHz reference.
Method
Two-interval forced choice: identify the higher tone. A 2-down/1-up adaptive track reports the geometric mean of the last reversals in cents and hertz.
Limits
Practice, audibility and the headphones affect the estimate; it is not a calibrated laboratory difference limen.
Gap & temporal-modulation detection temporal resolution
Measures
The shortest detected silent gap in a noise band, and the smallest detected amplitude-modulation depth at 4, 8, 16 or 32 Hz.
Method
Gap detection uses an adaptive gap/no-gap track with silent catch trials. Modulation detection uses two intervals (steady vs modulated noise) and a 2-down/1-up adaptive depth track.
Limits
Browser scheduling and transducers constrain precision. Human tinnitus studies of gap detection are mixed; neither result is a tinnitus detector.
Binaural lateralisation ILD / ITD
Measures
Headphone sensitivity to interaural level differences at 4 kHz and interaural time differences at 500 Hz.
Method
Two noise bursts carry equal-and-opposite cues; the listener reports left→right versus right→left. Separate 2-down/1-up tracks report dB ILD and microsecond ITD thresholds.
Limits
This is lateralisation inside headphones, not free-field localisation. Channel leakage, mono settings and headphone processing can invalidate it.
Dichotic stream attention selective attention
Measures
Accuracy while attending to one ear and ignoring a simultaneous competing tone stream in the other.
Method
Five separately routed tone pairs play; the listener counts high beeps only in the cued ear. The result reports overall, attend-left and attend-right accuracy.
Limits
A short tone-stream paradigm, not a validated dichotic-digits test. Ear differences can reflect audibility, routing, attention or chance and do not establish hemispheric dominance.
Mechanistic hypotheses the measurements can inform
The measurements are compatible with several overlapping explanations, which may coexist. The panel does not determine which mechanism is responsible in an individual — these experiments provide behavioural observations relevant to the models, not a way to select between them. They are theoretical frameworks, not competing diagnoses.
AMissing-signal compensation — the brain represents the deprived frequency more. Central.
BNoise amplification (stochastic resonance) — internal gain/noise rises to recover weak signals, and that noise becomes audible. Central.
CPrediction error — degraded evidence mismatches the brain's model, which fills in a percept. Central.
Turning the tests into a defensible, re-analysable record.
Raw session export reproducibility
Method
A JSON export carries per-frequency thresholds with posterior SD and trial counts, catch-trial/false-alarm data, the hardware fingerprint, the full tinnitus profile, reactivity descriptors/spectra/recovery curves, and the raw trial histories from frequency, gap, modulation, binaural and dichotic research tests.
Basis
Exporting observations and posterior state (not just the final numbers) lets a session be re-analysed later with different estimators/priors, or compared against calibrated audiometry — the route from a clever app toward a validation study.
6 References & scientific background
The literature that informed the design, grouped by theme. Peer-reviewed work and theoretical/preprint work are listed separately so the evidence hierarchy stays explicit. Links open external sites (journals, Consensus, arXiv).
Theoretical / preprint literature not peer-reviewed
Where the stochastic-resonance and nonlinear-cochlea hypotheses come from — theoretical and preprint work, kept separate from established clinical/audiological evidence.
Uncalibrated & relative. No true dB HL/SPL; results are device- and headphone-dependent and valid within one session/setup. Trust shapes and relationships, not absolute values.
No masking. There is no contralateral masking, so a marked left/right asymmetry could reflect cross-hearing rather than one ear.
Extended highs are hardware-limited. Apparent 10–16 kHz loss is often the headphones, not the ear.
Tinnitus outputs are measurements, not a diagnosis. Pitch/loudness/character are variable and weakly related to a cause or to distress; the classifier is a phenotype summary.
Comfortable levels only. The experiments never drive the ear hard, and must not be used to reproduce a past exposure or provoke symptoms.
Red flags go to a professional: sudden, one-sided, pulsatile, or asymmetric hearing change, or tinnitus with vertigo — see an audiologist or ENT, not a web page.
entertrainment · Hearing & Tinnitus Panel — a browser-based auditory psychophysics and tinnitus phenotyping instrument. This document describes methods; it is not medical advice, and the panel is not a substitute for calibrated audiometry or clinical assessment.