entertrainment MIT · Web Audio · no backend

Browser-based auditory psychophysics & tinnitus phenotyping

Where does the tinnitus sit, relative to the hearing loss?

A self-contained hearing & tinnitus instrument that runs in any browser — Bayesian adaptive audiometry, speech-in-noise, tinnitus psychoacoustics and exploratory mechanism experiments. Uncalibrated and relative: a research and phenotyping tool, not a diagnostic device.

Launch the panel Read the methods View source

The primary analytical question

Tinnitus is often discussed in terms of reduced input to the ear and compensatory changes in the central auditory system. A simple missing-input model might predict that the tinnitus should fall where hearing sensitivity is reduced — but in practice its pitch does not necessarily land at the audiometric edge. The panel estimates these quantities together and compares them:

hearing profile → loss edge → tinnitus pitch → octave displacement → estimated tonotopic position

Greenwood tonotopic coordinate is estimated from frequency — not a measured cochlear location. The result is correlational and specific to the individual.

What the panel measures

Six connected layers, from peripheral sensitivity to repeated daily-life observation and controlled auditory-processing research tasks.

Hearing profile

Per-ear detection thresholds, 250 Hz–8 kHz (plus optional extended highs), each with an uncertainty estimate, false-alarm rate and test–retest variation.

Bayesian ZEST-inspired · warbled chirps · catch trials

Functional hearing

A digits-in-noise task estimates the signal-to-noise ratio needed to recognise speech — information pure-tone thresholds alone can miss.

1-up/1-down adaptive · relative SRT

Tinnitus phenotype

Pitch (with an explicit octave check), loudness relative to threshold, sound character, minimum masking level and residual inhibition.

geometric-mean pitch match · dB SL · MML / MRIL

Mechanism experiments

Exploratory: pitch-vs-loss geometry, tinnitus spectrum, residual-inhibition function, tuning curve, and spectral stimulus reactivity.

Greenwood place · likeness map · RI / tuning curves

Auditory processing

Frequency discrimination, gap and modulation detection, binaural ILD/ITD lateralisation, and dichotic stream attention—with a provocation-free sound-pain interlock.

adaptive 2AFC · temporal · binaural · dichotic

Longitudinal context

Momentary prominence, loudness, distress, stress, sound tolerance and context remain separate, while bounded Sound Lab sessions capture immediate before/after observations.

local history · safety gate · ordered event trace

Built to be honest about its limits

A browser cannot measure the pressure at the eardrum — device, DAC, OS volume, headphones and coupling all affect the delivered sound. So the panel surfaces what is meaningful on uncalibrated hardware, and hides nothing behind apparently precise numbers.

It can explore

  • an individual hearing profile
  • tinnitus pitch & loudness relationships
  • functional hearing in noise
  • residual inhibition & stimulus reactivity
  • spectral, temporal & binaural processing
  • measurement repeatability

It cannot establish

  • the medical cause of tinnitus
  • cochlear pathology
  • calibrated hearing thresholds (dB HL)
  • a specific neural mechanism
  • treatment efficacy, or a cure
Current scope. Designed for exploratory, repeated single-subject measurements — not validated for population-level inference or clinical decision-making. Comfortable levels only; never raise levels to “test tolerance”. Sudden, one-sided, pulsatile or asymmetric hearing change — or tinnitus with vertigo — needs professional assessment.

Under the hood

Everything runs client-side on the Web Audio API. No account and no automatic upload: audio, results and saved history remain local unless the user explicitly exports, emails, or consents to the configured results service.

  • Bayesian adaptive thresholds (ZEST-inspired). A logistic psychometric posterior over each (ear, frequency), placed at the posterior mean, with informed priors from age, neighbouring frequencies and the other ear — reporting threshold + 95% credible interval + trial count.
  • Bias controls. Warbled pulsed chirps at a fixed level (reaction-time-proof), shuffled interleaving, and silent catch trials for a signal-detection false-alarm rate.
  • Speech-in-noise. Adaptive digits-in-noise with SNR = 20·log₁₀(RMS_speech / RMS_noise); a relative SRT from reversal means.
  • Tonotopy. A Greenwood place estimate x = log₁₀(f/165.4 + 0.88)/2.1 maps pitch and loss edge onto an estimated cochlear coordinate.
  • Spectral reactivity. Controlled degraded-audio conditions with spectral descriptors (centroid, flatness, crest, HNR) and a 0/5/15/30/60 s recovery curve.
  • Prospective personal A/B. Six Sound Lab sessions in three computer-randomised balanced pairs, with a preselected outcome, fixed digital level and duration, carryover guard, concealed interim comparison, and descriptive—not causal—result.
  • Measurement spine. A schema-versioned session joins per-session safety, capabilities, ordered requested/applied/observed events, repeated check-ins and bounded Sound Lab observations. A monitored master bus records mixed digital RMS/peak/clipping in dBFS—not acoustic SPL. History supports JSON/CSV export, merge-import and deletion.

Run it yourself

One self-contained index.html — open it locally, host it as a static file, or deploy to Netlify (the optional serverless function emails results via Microsoft Graph; set the MS_* env vars). No keys live in the repo.

# clone and open — that's it
git clone https://github.com/entertrainmentuk/hearing-tinnitus-panel
open hearing-tinnitus-panel/index.html

# or try it right here
→ Launch the panel   → Scientific methods