Sound generated by the inner ear and measurable in the ear canal. The outer hair cells of the cochlea are motile: they actively amplify the travelling wave, and a by-product of that amplification is acoustic energy propagating back out through the middle ear. Discovered by David Kemp in 1978. Two clinically used forms — transient-evoked (TEOAE, a click stimulus) and distortion-product (DPOAE, two simultaneous tones f1/f2, measuring the 2f1−f2 intermodulation product the cochlea itself creates). OAEs are the basis of newborn hearing screening worldwide, because a present emission means a functioning cochlear amplifier.
The distinction that keeps getting collapsed
Otoacoustic emission is not the same thing as the ear-canal acoustic transfer function, and in ear-biometrics pitches the two are routinely used as if they were.
| Ear-canal acoustic response | Otoacoustic emission | |
|---|---|---|
| Source of the signal | The probe tone, reflected and filtered | The cochlea, generating new sound |
| Physics | Passive geometry: canal length and diameter, cartilage compliance, eardrum and middle-ear impedance | Active: outer-hair-cell motility, a non-linear amplifier |
| What varies between people | Anatomy of the outer/middle ear | Cochlear micromechanics |
| Signal strength | Strong — your own stimulus coming back | Weak — the clinical criterion is 3-6 dB above the noise floor, recovered by time-averaging over many stimulus repetitions |
| Fails when | Fit changes | Hearing loss, noise, age, medication, ear infection |
| Prior art | Dense. Sandia US5787187A, filed 1 Apr 1996, expired. NEC live family since 2016 plus an Apple filing. NEC shipped ear acoustic authentication via crowdfunding in 2020 and has no OEM design win ten years on. | Thin for identity |
The practical consequence for diligence: a company claiming canal response is claiming something with a long, well-published, partly-expired prior-art trail and published accuracy that is not competitive — independent literature reports around 22% FRR at 0.1% FAR, against Face ID’s ~1-in-1,000,000 false-accept rate. A company claiming true OAE identity is claiming something more novel, but is fighting a signal that sits at the noise floor and needs averaging time set by noise statistics — which is not engineerable away, and which degrades exactly where a consumer wearable lives (motion, ambient noise, imperfect fit).
Related but separate: OAE as a channel for attention or eye position
The medial olivocochlear efferent pathway is real and top-down: the brain does modulate cochlear gain. Two claims get built on this and both are weaker than they sound.
Cost floor
Earbud-form OAE detection is already open-sourced at roughly $10 BOM (Nature Biomedical Engineering, 2022, University of Washington). Hearing screening as such is therefore not a differentiator; whatever is defensible has to sit above the sensing layer.
Prior art is denser than “thin for identity” (research pass, 28 Jul 2026)
The table above records OAE prior art as thin for identity. A literature sweep on 28 Jul 2026 shows that is too generous, and the line should be read as superseded.
EarNet (Neural Processing Letters, 2021, 2021 Earnet Teoae Biometric Embeddings) does identity from TEOAE specifically, not canal response: a FaceNet-style embedding network reporting verification EER 0.581% (left) and 0.057% (right), fused identification accuracy 99.92%, and embeddings that generalise to subjects outside the training set, which is an open-set claim. Related TEOAE work records sessions separated by a week or more, and a University of Toronto MASc thesis carries the title Earprint: Transient Evoked Otoacoustic Emission for Biometrics.
Two consequences for diligence.
The liveness argument is published prior art. EarNet’s authors state that TEOAE is immune to replay and falsification because the cochlea has to actively generate the signal. That is the strongest commercial argument available to an OAE identity company, and it has been in print since 2021.
So the honest position is not that OAE identity is unexplored. It is that OAE identity has been demonstrated in controlled academic settings with strong reported numbers, by groups that did not commercialise it, and the unanswered question has moved from “does the signal exist” to “does it survive a consumer earbud in motion, and does anyone pay for it”. Full landscape in Biometrics.
Frontier
Does any published work demonstrate cross-session, open-set identity from a true cochlear emission, as opposed to canal transfer function?Answered 28 Jul 2026: yes, EarNet 2021, EER 0.581%/0.057% with open-set generalisation. The live question is now why no one commercialised it.- What averaging time is needed to pull an identity-bearing OAE out of the noise floor in a moving subject, and does it fit a passive-authentication interaction budget?
- How much of a reported “OAE identity” result is actually earbud fit and session artefact?