Notched Sound Therapy for Tinnitus — Play Noise With a Notch at Your Tinnitus Frequency

Take a sound you would listen to anyway, cut out the frequencies around your tinnitus pitch, and listen for a long time. That is the method. The sound can be music — the best-known version is tailor-made notched music training (TMNMT), developed by Christo Pantev’s group at the University of Münster — or broadband noise, which is what this page plays, or the world coming through a hearing aid with a notch programmed into it. “Tailor-made” means the notch is centred on your own measured tinnitus frequency.

The reasoning is a hypothesis, and should be read as one. Most tinnitus follows some hearing loss. The cortical neurons that used to respond to the lost frequencies are proposed to become hyperactive and fire in synchrony without a sound to fire at, which is what you hear as a tone. Neurons tuned to neighbouring frequencies normally suppress their neighbours when active — lateral inhibition. A sound with energy either side of the tinnitus frequency and none at it should drive the neighbours while leaving the tinnitus neurons silent, so their inhibition falls on the overactive region and, with repetition, the map reorganises. Pantev’s group calls this inhibition-induced plasticity.

Keep two things separate. Lateral inhibition is real and well documented. Whether feeding it a notch for weeks changes tinnitus is the empirical question, and the trials answer it less tidily than the theory. It is also a different mechanism from masking — covering the tinnitus with another sound, which is what the plain noise generator does — and from acoustic coordinated-reset stimulation, which plays short tones around the tinnitus frequency rather than removing it; Wegger and colleagues (Frontiers in Neurology, 2017) reviewed that approach across eight studies and 329 patients and judged the evidence insufficient for clinical use.

The short version: the first studies were small and encouraging, the controlled trials since have mostly been null or marginal, and the pooled estimates disagree depending on which trials are counted.

The founding paper is Okamoto, Stracke, Stoll and Pantev (PNAS, 2010). Eight patients listened to their favourite music, notched at their tinnitus frequency, for twelve months, against eight whose music was notched elsewhere as a placebo. The treated group reported lower tinnitus loudness and showed less tinnitus-related auditory-cortex activity on magnetoencephalography. It is careful work, and it is sixteen people. Teismann, Okamoto and Pantev (PLoS One, 2011) then compressed the dose to five days at six hours a day: loudness, distress and cortical activity fell in the ten patients whose tinnitus sat at or below 8 kHz and did not change in the ten above it — the origin of the “below 8 kHz” caveat — and the gains had faded by the four-week follow-up. The group later showed the cortical effect accumulating over three days in nine patients (Stein and colleagues, Clinical Neurophysiology, 2015) and, most relevant to the controls on this page, that notch width does not seem to matter: Wunderlich and colleagues (PLoS One, 2015) compared one-octave, half-octave and quarter-octave notches over three months and found distress fell in every group with no difference between widths — though all three arms received a notch, so that study cannot say whether any beat unnotched music.

The test of that was the group’s own randomised trial, Stein and colleagues (BMC Neurology, 2016), run to CONSORT standards with a registered protocol (Pantev and colleagues, BMC Neurology, 2014): 100 people with tonal tinnitus, two hours a day for three months, against a placebo-notched control. The primary outcomes — the Tinnitus Handicap Questionnaire and visual-analogue ratings of loudness, awareness, distress and handicap — showed no effect. A secondary distress score moved the wrong way in the treatment group, short of significance, and had normalised by follow-up; loudness was lower than control at the one-month follow-up, and a post-hoc analysis found the averaged visual-analogue score lower straight after training. The authors called it one more step, not a result.

Independent trials have leaned the same way. Piromchai and colleagues (Otology & Neurotology, 2021) randomised 75 people to notched music, ordinary music or counselling alone for three months: handicap fell in all three groups with no difference between them. Therdphaothai and colleagues (Journal of International Advanced Otology, 2021) ran a double-blind trial of notched against ordinary music in 104 Thai patients followed for six months and found no significant difference on handicap or loudness, though the notched group’s scores fell further over follow-up. The one recent positive headline is Tong and colleagues (Ear and Hearing, 2023): 120 people randomised to notched music or a two-hour-a-day version of tinnitus retraining therapy, where the notched group did better by about seven handicap points at one month — an interval that just cleared zero — and by about six at three months, an interval that did not. Seven points is roughly the smallest change on that scale a patient notices.

Notched noise, which is what this page plays, has less evidence than notched music. Lugli and colleagues (International Tinnitus Journal, 2009) found loudness fell over months in people given notched broadband noise but not in those given plain noise or a water sound; the abstract does not describe randomisation. The most direct comparison since points the other way: Jin, Choi and Ku (Journal of the American Academy of Audiology, 2021) gave thirty adults notched or plain broadband noise for three hours a day for a year, at the level where the sound just mixes with the tinnitus, and both groups improved by a similar amount — their conclusion was that people who cannot get a pitch match can simply use plain noise. A two-group comparison in 64 adults with normal hearing, Mekki and colleagues (International Archives of Otorhinolaryngology, 2024), set notched sound against cognitive behavioural therapy: both improved, CBT did more for distress, notched sound more for matched loudness; it is not described as randomised.

Notches programmed into hearing aids, the version most people would actually be offered, have been tested properly and have not held up. Daniel Strauss’s group in Saarland reported a twenty-patient evaluation of notched environmental sounds (Clinical Otolaryngology, 2017) and a longer evaluation of a half-octave notch showing a trend towards larger benefit than standard aids (Haab and colleagues, IEEE Journal of Translational Engineering in Health and Medicine, 2019). Since then Marcrum and colleagues (Progress in Brain Research, 2021) found no difference between notched and conventional amplification in 39 adults over twelve weeks; Waechter and colleagues (International Journal of Audiology, 2026) randomised 108 people to conventional amplification, extra gain at the tinnitus pitch or a notch, and found no difference between them, with conventional fitting giving better speech understanding; Santacruz, de Kleine and van Dijk (Audiology Research, 2025) found no difference in an 18-person double-blind crossover; and Quemar and colleagues (European Archives of Oto-Rhino-Laryngology, 2026) found standard amplification beat the notch in a 26-person crossover.

A randomised trial of a notched-sound smartphone app is instructive. In the preliminary report, Adamovich-Zeitlin and colleagues (Otology & Neurotology, 2023), only about a third of people given a free subscription managed the prescribed two hours a day, and handicap fell similarly in the app group and in standard care. The updated analysis, Sturm and colleagues (Otology & Neurotology, 2026), found that being assigned the app went with a larger fall in handicap but that actually using it did not — which the authors read as a psychological effect of being given something to do rather than an acoustic one.

The reviews split as you would expect. Alfonso and colleagues (American Journal of Otolaryngology, 2024) pooled three studies of notched against plain music and found no significant difference in handicap. Tavanai and colleagues (European Archives of Oto-Rhino-Laryngology, 2024) pooled five trials and found no significant effect on handicap at three or six months, with high heterogeneity, though four of five studies reporting a loudness scale found it improved. Jiang and colleagues (European Archives of Oto-Rhino-Laryngology, 2025), which also searched Chinese-language databases and pooled fourteen randomised trials with 793 participants, found notched music beat conventional music by about nine handicap points at three months — an interval running from a meaningful benefit down to almost nothing — and by about one point on a ten-point loudness scale; it drew a letter to the editor in reply. Like Wunderlich, they found notch width was not a moderator.

The fair summary: notched sound is safe as far as anyone has looked, cheap and plausible. In the best-controlled trials it has not clearly outperformed the same sound without a notch, or a hearing aid without one. Where positive results exist they are small — around the threshold of what a patient can feel — mostly in loudness ratings rather than in how much the tinnitus intrudes on life, and seen more often in tonal tinnitus at or below 8 kHz. That is the claim this page supports, and no more.

Every trial began by measuring the tinnitus frequency, because a notch in the wrong place is just noise with a hole in it. The hypothesis depends on the removed band sitting over the neurons you are trying to quieten; an octave away, the tinnitus region is being driven by the sound rather than surrounded by it.

In the clinic that is done by pitch matching: a tone is adjusted until it sounds the same pitch as the tinnitus, repeated, and checked an octave above and below, because people commonly land on the wrong octave first. At home you can approximate it with the tinnitus frequency test on this site, which links back here with the frequency filled in. Do it on different days and keep the value you return to; check the octave above and below; and if your tinnitus is a hiss rather than a tone, distrust any single number, because a broadband sound has no single frequency to notch.

The frequency also sets expectations. Teismann’s trial saw effects only at or below 8 kHz, and much tinnitus sits above that. This page will place a notch up to 16 kHz, but most consumer headphones roll off sharply in the top octave, so the sound either side of a 12 kHz notch may be barely there, and the theory needs healthy neighbours on both sides of the notch, which a high-frequency hearing loss does not supply.

The trials that found anything used a lot of listening: two hours a day for three months in Münster, three hours a day for a year in the Korean noise study, twelve months in the original paper. Teismann’s intense five-day version produced changes that had faded a month later, which is why the group moved to long daily protocols. If notched sound does anything, it does it slowly.

That is also its practical weakness. In the app trial only about a third of people given the app free managed two hours a day. Two hours daily for a season is a large ask, so if you try it, fold it into something you already do for a long stretch — work, a commute, reading — rather than adding a session to the day.

Loudness has one clear rule: at or just below the level of your tinnitus. Jin’s trial used the mixing point, where the sound and the tinnitus just blend. Louder is not better, and hours of loud sound are a hearing risk of their own, which is why this tool starts quiet. Bring it up until the noise sits alongside the tinnitus, and leave it there.

Expect nothing dramatic. In the positive trials the average change was a few points on a handicap scale after months, and control groups often improved by similar amounts, because tinnitus tends to bother people less over time and doing something about it helps in itself. If after a few weeks the sound is pleasant and the tinnitus a little less prominent while it plays, that is a reasonable outcome; if it makes the tinnitus more noticeable, stop.

What this page cannot do is deliver the trial dose: the free tools stop at twenty minutes, enough to hear what a notch sounds like, compare it against plain noise with the A/B button, and decide whether to pursue it. If you do, any free audio editor with a band-stop filter can notch your own music, and an audiologist can say whether a notch is available in a hearing aid you already need.

Before any sound experiment, tinnitus deserves a proper assessment, and the guideline on what that looks like is clear. The 2014 clinical practice guideline on tinnitus by Tunkel and colleagues (Otolaryngology–Head and Neck Surgery) recommends a targeted history and examination at first presentation; a prompt full hearing test when tinnitus is one-sided, has lasted six months or comes with hearing difficulty; and a distinction between tinnitus that is bothersome and tinnitus that is merely present. For persistent, bothersome tinnitus it recommends education, a hearing aid evaluation where hearing loss is documented, and cognitive behavioural therapy; it lists sound therapy as an option; and it recommends against routine medication, supplements such as Ginkgo, melatonin and zinc, and transcranial magnetic stimulation.

CBT sits at the top because of the size of its evidence base. The Cochrane review by Fuller and colleagues (2020) pooled 28 trials with 2,733 participants and found CBT reduced the impact of tinnitus on quality of life against no treatment by roughly eleven points on the Tinnitus Handicap Inventory, where seven is the smallest change patients notice; against ordinary audiological care the difference was smaller but held at moderate certainty, with serious adverse effects essentially absent. CBT does not change the sound; it changes what the sound costs you.

Sound therapy as a category, where notched sound lives, has thinner evidence than people assume. The Cochrane review of sound therapy by Hobson, Chisholm and El Refaie (2012) found six trials and no significant change in loudness or severity from sound devices compared with education, relaxation, counselling or retraining, and no side effects. Its successor, the Cochrane review by Sereda and colleagues (2018), found eight trials, no usable data for devices against a waiting list or placebo, and no difference between hearing aids and sound generators, though both were associated with a clinically meaningful fall in severity at low certainty. The Cochrane review of hearing aids for tinnitus by Hoare and colleagues (2014) found one randomised trial and no evidence to support or refute hearing aids as a routine tinnitus intervention — which does not stop them being the sensible first step when hearing loss is present.

So the order is: doctor or audiologist, then hearing aids if you need them, then CBT if the tinnitus is distressing, and sound — masking, enrichment, notched or otherwise — alongside those, not instead of them.

The noise is the same white and pink noise as the noise generator, generated in your browser from a four-second loop, and the notch is a cascade of eight filters rather than one clean cut.

The notched-music trials filtered recordings offline with steep band-stop filters that remove essentially everything inside the band and nothing outside it. A browser filters live audio with the blocks it has, and the relevant one is a second-order notch: deep at its centre, but narrow and V-shaped. To approximate a wide, flat-floored notch, this page spaces eight of them across the middle half of the band you select, makes each a quarter of that band wide, and corrects for how digital filters distort near the top of the audible range so the notch keeps its shape up to 16 kHz. Modelled at the two common sample rates: the width at which the noise is halved in power is the width you chose; across the central half of the band the noise is at least about 28 dB down — very quiet, not silent; and the outer quarter on each side is only a few decibels down, so the shoulders are rounded rather than vertical. The notch is deepest at the centre, where your pitch match puts the tinnitus.

Two other limits. The loop repeats every four seconds, inaudibly in noise, but it is a loop. And nothing after the filter is calibrated: your volume, device, headphones and room all shape what reaches your ear, and most consumer headphones roll off steeply in the top octave, so a notch centred above about 12 kHz may have little sound on its upper side.

The A/B button is the most useful control. It crossfades between the filtered and unfiltered paths of the same noise, so the only thing that changes is the notch. If you cannot hear a difference at your matched frequency, the match is off, the headphones are not producing that band, or the notch is above where your hearing reaches — worth knowing before spending weeks on it.

This site is mostly about audio for sleep, focus and calm; the tinnitus pages are a side door. Everything here is free, runs in the browser, and treats nothing.

• Tinnitus frequency test — an uncalibrated pitch match that hands its result to this page.
• Noise generator — the five noise colours without a notch, for plain masking. Jin’s trial is a fair argument that for many people this is as good.
• Binaural beats and tinnitus — what the two small trials of binaural beats in tinnitus found.
• Hearing range sweep — probe the top of your hearing; not a hearing test.
• The main generator — binaural, isochronic or pure tones over a white, pink or brown noise bed, capped at twenty minutes.

The mobile app is a different product: 44 presets, 34 ambient sounds and a Frequency Lab with a custom tone generator, sweep designer and mixer, with no session cap. It has no notch filter and does no pitch matching.