Most tinnitus appears to begin when the ear sends the brain less signal than it used to, and the brain's hearing pathways compensate by turning up their own activity until that activity is heard as sound.
That much is broadly agreed. A 2021 review in Physiological Reviews says tinnitus is usually initiated by acquired hearing loss and maintained only alongside distinct changes in auditory and non-auditory brain networks, and that what is necessary and sufficient to produce it remains incompletely understood.
A 2026 Nature Reviews Disease Primers article (the Primer) sets out the current chain: cochlear injury and deafferentation may trigger maladaptive plasticity, increased central gain and thalamocortical dysrhythmia, modulated by limbic and salience networks, with neuroinflammation, somatosensory–auditory coupling and stress possibly contributing to chronicity. Tinnitus and hearing loss gives the short version; below, each link is graded by its evidence.
Where the evidence comes from
Most single-neuron evidence comes from rodents and guinea pigs, which cannot say they hear ringing. The usual test infers tinnitus from the startle reflex: a silent gap before a loud sound normally dampens the startle, and when it fails to, the gap is taken to have been filled by tinnitus. A 2015 review found that laboratories run and interpret this test so differently that studies are hard to compare, and that recent work challenges the "filling-in" interpretation itself.
A 2016 review by Shore, Roberts and Langguth in Nature Reviews Neurology notes that the stronger studies compare identically exposed animals with and without signs of tinnitus, separating tinnitus from hearing loss.
Step one: the ear loses input
Most tinnitus is associated with hearing loss from ageing or noise, according to the 2016 review, and the clue is where the sound sits. When people with sloping high-frequency hearing loss rate tones for similarity to their tinnitus, ratings begin near the edge of normal hearing and rise with the depth of loss, so the tinnitus spans the damaged range. Band-limited noise centred there can suppress tinnitus for 30 to 60 seconds afterwards, an effect called residual inhibition.
Three lines of human evidence tie tinnitus to lost input:
- Silence. In a 2008 study, 53 normal-hearing young adults spent 4 minutes in an empty anechoic chamber, then ticked sounds they had heard from a list; 83% reported at least one, echoing Heller and Bergman's 1953 finding. With a dummy loudspeaker present in a second session, it rose to 92%; the authors concluded suggestion plays only a minor role.
- An earplug. In a 2012 study, 18 normal-hearing volunteers wore an earplug in one ear for 7 days, simulating a mild high-frequency hearing loss. 14 of them reported phantom sounds. In the 11 whose sounds were stable at day 7, they were mostly high-pitched, in the range the plug blocked most. In every case the sound disappeared when the plug came out.
- Deafness from birth. A 2020 review built its model on evidence that tinnitus is rare in people born deaf, common in people who lose hearing later, and potentially suppressed by working cochlear implants. Its authors propose that tinnitus can only develop in a brain that has first learned to hear.
When the damage hides from the hearing test
Some people with tinnitus have a normal audiogram, and one leading explanation is damage a standard test cannot see. In a 2009 mouse study, noise that raised hearing thresholds only temporarily left the hair cells intact but caused acute loss of their nerve terminals and delayed degeneration of the auditory nerve, which the authors suggested could contribute to tinnitus.
In people the picture is mixed. A 2011 study found a smaller wave I in the auditory brainstem response, the part generated by the auditory nerve, in adults with tinnitus and normal audiograms, but a normal wave V from higher up, which the authors read as the brainstem restoring its response to a weaker input. A 2017 study of young adults matched for hearing up to 14 kHz linked tinnitus to more lifetime noise exposure but found no reduction in wave I. This nerve damage is established in animals and contested in humans; hidden hearing loss covers the argument.
Step two: the brain turns up its gain
Neurons tend to hold their average activity steady, so when input falls they become more responsive: homeostatic plasticity, the basis of central gain. A 2006 computational model showed this compensation can produce hyperactivity after outer hair cell loss and can amplify non-auditory inputs. It also predicted that added sound could reverse the hyperactivity, one rationale for sound enrichment such as the sound library; that is a model's prediction, not a trial result.
In animals, three changes accompany behavioural signs of tinnitus, from the cochlear nucleus to the auditory cortex: more spontaneous firing, more burst firing and more synchronised firing. In a 2019 mouse study, noise-induced hearing loss triggered inflammation in the auditory cortex; blocking the inflammatory molecule TNF-α reduced behavioural signs of tinnitus, and infusing it into the auditory cortex of normal-hearing mice produced them. That evidence is from rodents only.
The difficulty is that gain, on its own, does not appear to explain who gets tinnitus. A 2019 review titled "Tinnitus: Does Gain Explain?" notes that dramatic gain changes follow ear damage whether or not tinnitus occurs; a few studies find tinnitus-specific changes at group level, but the limited attempts to classify individuals by gain have not succeeded, and the review leaves open whether gain is necessary or sufficient. Synchrony may be one difference: a 2019 review in Neuron reports that among animals given equivalent cochlear damage, only those that develop tinnitus show synchronised spontaneous firing in brainstem neurons fed by the auditory nerve. And many older people with impaired hearing have no tinnitus, a 2010 review notes, possibly because their inhibitory circuits are better preserved.
A rival 2020 model proposes that added central noise is likely to generate tinnitus, and multiplied central gain is likely to generate hyperacusis, over-sensitivity to everyday sound.
Where the jaw and neck come in
The dorsal cochlear nucleus, in the brainstem, is the first place in the hearing pathway where sound meets touch. Its main output neurons combine input from the cochlea with somatosensory input from the head and neck. In animals, those somatosensory inputs strengthen over days after cochlear damage; animals that develop tinnitus then show more strengthening at these connections, and animals that do not show more weakening.
This is the proposed basis for somatic tinnitus, tinnitus that changes when you clench the jaw or move the neck, which the 2016 review says occurs in up to two thirds of people with tinnitus. In a 2018 study, guinea pigs with noise-induced tinnitus and then 20 people received paired sound and somatosensory stimulation timed to weaken those connections. In the double-blind, sham-controlled human crossover, 28 days of it reduced tinnitus loudness and intrusiveness, and sound alone did not. The human trial was small; bimodal neuromodulation covers later trials.
Brain rhythms: thalamocortical dysrhythmia
In 1999 Llinás and colleagues found increased slow theta rhythm, more strongly coupled with fast rhythms, in people with neurogenic pain, tinnitus, Parkinson's disease or depression, and proposed a shared "thalamocortical dysrhythmia" driven by bursting thalamic cells. A 2015 theoretical update applied it to tinnitus, proposing that lost input slows the resting alpha rhythm towards theta, coupled to fast gamma activity.
The most direct human evidence comes from one person: a 2015 study recorded inside the brain of an awake patient while residual inhibition briefly quietened the tinnitus. Tinnitus-linked slow delta rhythms appeared as anticipated, but extended across almost all of auditory cortex and large parts of temporal, parietal, sensorimotor and limbic cortex.
Scalp recordings are less consistent. A 2016 study noted that EEG and MEG research shows no clear relationship between tinnitus and the power of brain rhythms, found none in its own patients between resting EEG and tinnitus measures, and concluded that resting scalp EEG should not be used as a tinnitus biomarker. The model is influential; the human evidence is inconsistent.
Why loudness and distress are different problems
How loud tinnitus is and how much it bothers someone are only moderately connected. In a 2007 survey of 4,958 people recruited through a national tinnitus association, a self-selected group, loudness and annoyance correlated at 0.45, and about one third of those with very loud tinnitus reported only mild or moderate annoyance. Annoyance was more likely with hyperacusis (odds ratio 4.96), vertigo or dizziness (1.94) and hearing loss (1.71).
One imaging study points the same way. A 2012 MRI study found less grey matter in the ventromedial prefrontal cortex in tinnitus than in controls matched for age and hearing loss, unrelated to distress, anxiety or depression; distress instead tracked the thickness of the anterior insula, a key salience-network region according to the 2016 review. The authors concluded that the systems behind having tinnitus differ from those behind being distressed by it. An earlier model from the same group proposes that limbic regions normally tune the tinnitus signal out, and that chronic tinnitus follows when that fails. Tinnitus and anxiety covers the emotional loop built on top.
The 2016 review cautions that it is not yet known whether these non-auditory changes predispose people to tinnitus or follow from it, or how much they reflect hearing loss, hyperacusis and distress. And a 2024 meta-analysis of 17 resting-state fMRI studies began from "notable inconsistency" between them; pooled, it found increased activity in the insula and in temporal and frontal regions on both sides.
That separation is consistent with the Primer listing tinnitus-focused counselling and cognitive behavioural therapy as first-line treatments.
The newest frame: tinnitus as a prediction
Predictive coding treats perception as the brain's best guess, checked against incoming signals. In 2013 Roberts and colleagues proposed that tinnitus involves a mismatch between what the brain predicts it should hear and what the damaged cochlea delivers, engaging attention. A 2016 model by Sedley, Friston and colleagues calls spontaneous activity in the hearing pathway a "tinnitus precursor", normally ignored as unreliable evidence against the default expectation of silence. Weighted heavily enough, it is heard; focused attention, and the expectation resetting to include tinnitus, keep it going. The model absorbs the others as ways of making the precursor stronger or more trusted.
Tested directly, it has mixed results. A 2019 EEG study found the predicted asymmetry in responses to unexpectedly loud and quiet sounds in 26 people with chronic tinnitus and 15 with recent-onset tinnitus, but not in 26 matched controls, separating groups with an area under the curve of 0.77. A 2023 replication attempt from the same group, with tighter matching, found no significant group differences.
What is solid and what is not
- Solid in humans: tinnitus is tied to hearing loss and usually sits in the damaged range; reduced input can produce a phantom sound.
- Shown in a large survey and one MRI study, mechanism unclear: loudness and distress are partly separate.
- Strong in animals, partial in humans: central hyperactivity and synchrony; somatosensory rewiring in the dorsal cochlear nucleus.
- Strong in animals, contested in humans: hidden nerve damage.
- Plausible, contested in humans: thalamocortical dysrhythmia as a signature; specific limbic circuits; predictive coding.
- Not yet available: an objective test; tinnitus is still measured by self-report, as how tinnitus is measured explains.
What this means if you have tinnitus
The Primer's diagnostic steps include audiometry and screening for somatic factors, and it calls hearing rehabilitation, such as hearing aids where there is hearing loss, a valuable adjunct. How tinnitus is diagnosed sets out the workup, and the treatment comparison lines up the options against their evidence.
Some tinnitus has a different cause. The Primer lists pulse-synchronous tinnitus as a red flag needing vascular imaging (pulsatile tinnitus explains why), and NIDCD advises treating sudden hearing loss as a medical emergency and seeing a doctor immediately. The when to seek care checklist sorts which symptoms need prompt attention.