BRENNEN

Industries / Health / Medical devices / Hearing aids

Sound processing

Sound processing is what the chip in a hearing aid does to incoming sound: amplifying soft sounds more than loud ones, favoring speech over noise and stopping the whistle of feedback.

7

Milestones achieved, from 1971 to 2025.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1971

    Microphones that favor the front

    A directional microphone picks up more sound from in front of the wearer than from behind, which helps when the talker faces the listener. Hearing aids with directional microphones were first marketed in the US in 1971, but their use declined after 1980, because in many places, such as echoing rooms or when speech and noise come from the same direction, they give no benefit. Phonak revived the idea in 1992 with AudioZoom, which let the wearer switch between directional and all-round pickup.

    Lessona feature that helps only in some places needs a way to tell where it is, which led makers to classify the listening scene automatically, Harry Levitt wrote in 2007.

    pmc.ncbi.nlm.nih.gov

    Credit

    Phonak (AudioZoom)

  2. 1973

    Soft sounds lifted most

    Edgar Villchur's 1973 paper described multiband compression: in each of several frequency bands, soft sounds are amplified a lot and loud sounds little, to fit speech into the narrow range between what a damaged ear can hear and what it finds too loud. He published instead of patenting, and tested a three-channel portable aid, built by Mead Killion, on profoundly deaf people for a year. In 1987 AT&T passed the rights to a Bell Laboratories aid that used two-channel compression to the ReSound Corporation, which sold it with "immediate success", Levitt wrote. Villchur's compression "became the ReSound hearing aid", Killion said in 2011, and "now it is almost impossible to find a hearing aid that doesn't incorporate his compression."

    hearingreview.compmc.ncbi.nlm.nih.gov

    Credit

    Edgar Villchur; Bell Laboratories; ReSound

  3. 1989

    A digital aid withdrawn

    Nicolet's Project Phoenix, run by Kurt Hecox from 1984, built what Levitt credits as the first commercial digital hearing aid, on custom chips: a processor worn on the body and wired to the ears, and later a behind-the-ear model. It did not sell, and Nicolet left the hearing aid market in 1989.

    LessonPhoenix showed that a wearable digital aid could work, but the main obstacle to a viable product was shrinking the size and power use of digital chips, while analog chips were smaller and used less power still, Levitt wrote.

    pmc.ncbi.nlm.nih.gov

    Credit

    Nicolet Corporation; Kurt Hecox

  4. 1996

    Digital aids reach the ear

    Widex's Senso, launched in 1996, was the first commercially successful digital hearing aid; Oticon's DigiFocus followed, built on a platform Oticon had distributed to research centers the year before. Digital processing allowed feedback cancellation, which removes the whistle when amplified sound leaks back to the microphone without simply cutting gain; Levitt judged it the digital change with "the greatest impact", because it allows more gain and open, unsealed fittings.

    Lessonearly comparisons found little or no measured benefit over analog aids, and Ruth Bentler showed a halo effect: wearers rated an aid better when told it was digital.

    pmc.ncbi.nlm.nih.gov

    Credit

    Widex; Oticon

  5. 2008

    Noise reduction eases listening

    Digital noise reduction turns down frequency bands that seem to hold steady noise. In a double-blind study of 25 wearers of one commercial aid, Ruth Bentler and colleagues found that it made no difference to speech scores in the laboratory, but listeners rated listening easier with it on and rated noise more aversive with it off.

    Lessonnoise reduction makes noisy places less tiring without, on its own, letting wearers understand more words.

    europepmc.org

    Credit

    Ruth Bentler and colleagues

  6. 2024

    Neural networks on chips

    A deep neural network (DNN) is software trained on many examples, here of speech mixed with noise, to separate the two. Oticon More, reported in January 2021, ran a DNN on board, trained on 12 million recorded sound scenes by Oticon's count. In 2024 Phonak's Audéo Sphere Infinio added a second chip, DEEPSONIC, only for DNN noise removal. Phonak's own tests on a manikin gave a 5.9 dB signal-to-noise benefit over the open ear, 2.6 to 3.7 dB more than three rival premium aids. The independent HearAdvisor lab scored Sphere 4.4 of 5 for background noise in its standard test and 4.7 with the DNN mode on, the best it had measured; the mode at first ran the battery down in about 7 hours, by Phonak's account.

    Lessonthe network cannot clean sound that bypasses the aid; Brian Moore, who wore a DNN aid for 15 months, rated it 8 of 10 with a sealed fitting and 5 of 10 with an open one, because noise leaks through the vent.

    eenewseurope.comhearingreview.comhearingtracker.compmc.ncbi.nlm.nih.gov

    Credit

    Oticon; Phonak (Sonova); HearAdvisor

  7. 2025

    Machines judge intelligibility

    The Clarity Project, a group of UK universities funded by UK Research and Innovation, runs open contests for hearing aid processing and for software that predicts how many words a listener with hearing loss will understand. In its 2023 contest, five teams beat the baseline on simulated scenes but did much worse on scenes built from real-room measurements. In the third prediction contest (2025), the best system missed listeners' word scores on single sentences by 25.0 percentage points on average (root-mean-square error) and correlated 0.80 with them; HASPI, the standard hearing-aid speech perception index of J. M. Kates and K. H. Arehart, ranked 19th, at 29.5 points and 0.70.

    Lessonmethods tuned on simulated rooms can fail in real ones, and even the best predictor is far from exact for a single sentence.

    arxiv.orgclaritychallenge.orgclaritychallenge.org

    Credit

    The Clarity Project; J. M. Kates and K. H. Arehart (HASPI)

Watch for

  1. A smaller noise-removal chip

    Phonak says its Audéo EON Sphere, on a new HYPERSONIC chip, improves the signal-to-noise ratio by 10.5 dB over all-round microphone pickup (9.7 dB at default strength), in a device 25% smaller and 20% lighter, and that it beat four other brands' aids for speech in loud noise in Phonak's own technical measurements.

    Achieved whena lab with no tie to Sonova measures the improvement over all-round pickup within about 1 dB of Phonak's figure in its own scenes.

    phonak.com

    Claimed by

    Phonak (Sonova), technical paper by Ulloa Sanchez, Kepp, Preuss and Raufer, August 2026

  2. An AI lead that shrank

    Fortell, a US company that fits its prescription aids at its own locations, says its "Spatial AI" gave a 17 dB improvement in signal-to-noise ratio against 1.9 dB for a control aid, and that wearers understood 80% of words at 0 dB signal-to-noise ratio against 31% with "the leading AI-powered hearing aid", in its own blinded studies. HearAdvisor ranked Fortell AI 2.0 above Phonak's EON Sphere in September 2026, but when it retested both in October 2026 with the same gain (tuned to NAL-NL2) and equally sealing domes, Fortell's lead in loud scenes fell from 2.7 to 0.2 points on HearAdvisor's scale.

    Achieved whena peer-reviewed or independent study with gain and ear tips matched shows a speech-in-noise advantage over current premium aids.

    fortell.comhearadvisor.com

    Claimed by

    Fortell, 2026

Open

  1. 01

    Delay in each noise mode

    A public table, for current flagship prescription aids, of processing delay (the milliseconds between sound reaching the microphone and leaving the receiver) with each noise-removal mode, DNN modes included, switched off and on, all measured in a test box with no wearer.

    The brief
    Credit · unclaimed
  2. 02

    Programmed versus delivered compression

    Measure the compression ratios that current aids actually deliver, with running speech and with test tones, and compare them with the ratios their own fitting software shows, for the same audiograms, across at least five makers. A compression ratio of 2, for example, means a 10 dB rise in input gives a 5 dB rise in output.

    The brief
    Credit · unclaimed

Open 01 · the brief

Delay in each noise mode

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public table, for current flagship prescription aids, of processing delay (the milliseconds between sound reaching the microphone and leaving the receiver) with each noise-removal mode, DNN modes included, switched off and on, all measured in a test box with no wearer.

Why it is still open

The longest published series is student measurements from Joshua Alexander's hearing aid class at Purdue, 2009 to 2015, all under 10 ms; Wayne Staab wrote in 2023 that he had found no more recent data. HearAdvisor's published method scores five areas, none of them delay, and Phonak's EON paper gives no delay figure. A 2026 review (Hanegan and colleagues) found that only 10 of 46 studies of AI for hearing devices described a model that fits a 5 to 10 ms delay budget, and only 2 of the 16 that used deep neural networks. Delays this short still matter: Küçük and Çetinkaya (2026) found test scores differed between 0.5 ms and 2.7 ms of delay in 20 hearing aid users.

hearinghealthmatters.orgcanadianaudiologist.capmc.ncbi.nlm.nih.goveuropepmc.org

Built with what exists

A hearing aid test box that measures delay, such as the Frye FONIX 8000 or Audioscan Verifit 2, with a 2cc coupler; the makers' fitting software, through a university audiology clinic; a recorded noisy scene played on a loop so that each noise mode engages; Python to measure delay by lining up input and output signals.

  1. First step

    Measure the delay of two DNN-equipped aids with their DNN mode off and on in a university test box.

  2. Done looks like

    At least 15 current models from at least 5 makers, delay at four frequencies in each mode, each measured three times and published with the method, and a second lab matching three models within 0.5 ms.

Open 02 · the brief

Programmed versus delivered compression

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

Measure the compression ratios that current aids actually deliver, with running speech and with test tones, and compare them with the ratios their own fitting software shows, for the same audiograms, across at least five makers. A compression ratio of 2, for example, means a 10 dB rise in input gives a 5 dB rise in output.

Why it is still open

Brian Moore wrote in 2026 that "in my experience, the obtained CRs are less than the CRs programmed using the manufacturer's software", with test signals and with running speech alike, so soft sounds cannot be made audible without loud ones becoming uncomfortable; he cited no measurement across makers. HearAdvisor and Kruger and colleagues report how closely gain matches targets at three speech levels, not the ratios. A search of Europe PMC in November 2026 for measured effective compression ratios found studies of compression's effects but no comparison of current makers.

pmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

Built with what exists

A hearing aid test box or acoustic manikin; the International Speech Test Signal played from 50 to 80 dB SPL in 5 dB steps, plus tone and noise sweeps; the makers' fitting software and its exported settings; the standard audiograms N3 and S2; Python to compute delivered ratios from the output levels.

  1. First step

    Program one aid to N3, export the compression ratios its software displays, and measure its input-output curve with speech at seven levels.

  2. Done looks like

    For at least 5 makers and 3 audiograms, a published table of programmed against delivered compression ratio in each frequency band, for both speech and tones, with raw recordings and code, and a second lab matching two of the aids within 0.3 in ratio.