Medical devices & diagnostics · Hearing technology
AudioFocus
A Y Combinator company built to lift the voices at your table out of the roar of a crowded room. Seven years on, the public record holds one granted patent, an archived waitlist, and no sign that a device reached a buyer.
A show of hands
On November 9, 2019, the East Bay chapter of the Hearing Loss Association of America met to hear a young scientist talk about Bluetooth. Shariq Mobin had finished a doctorate at UC Berkeley in auditory neuroscience and machine learning, and had started a company. Partway through he asked the room: how many of you actually use Bluetooth with your hearing aids?
About a quarter of the hands went up.
Then he asked what does not work. The answers came back as a list of places. Restaurants. Cafes. Parties. Musical events. Meetings. Television. Driving in cars or trucks. Every one of them a room with more than one voice in it.
The gap it aimed at
The numbers under the pitch are federal. About 37.5 million American adults, roughly 15 percent, report some trouble hearing, according to the National Institute on Deafness and Other Communication Disorders. About 28.8 million could benefit from hearing aids. Among adults 70 and older who could benefit, fewer than one in three has ever used them.
The reason people give for leaving them in a drawer is noise. A 2023 paper in Scientific Reports opened on the same point: over 80 percent of people who would benefit from use of hearing aids do not wear them.
Conventional noise reduction leans on direction, and the company argued that the assumption breaks in the room that matters most. Its site made the objection twice over: directional systems assume the noise sits on one side, and machine learning systems that enhance speech and suppress everything else have nothing to grip in a cafe, where the noise is itself speech.
A neuroscientist and a hardware architect
Mobin studied deep learning models of sound and the science of hearing loss at UC Berkeley and Google Brain, per the company team page. His co-founder, Reza Kassayan, came from EarLens, where the page says he was hardware architect of a light driven hearing aid.
The Berkeley licensing office lists AudioFocus among its inventions, dates the founding to 2019, and names Shariq Aman Mobin and Bruno Olshausen as inventors. Whether that invention was ever licensed to the company has not been announced.
Floyd
The technical bet was on data the company could make itself. Training a network to pull one voice from a crowd needs recordings of that crowd as the listener hears it, from both ears, in thousands of rooms, and the company said no such corpus exists. So it built a simulator, generating each room twice, once loud and once quiet, then training the model to map the first onto the second.
A prototype, and then quiet
In July 2023 Mobin posted to Hacker News. The company had a behind the ear prototype running, he wrote, several excited patients, and a pilot study in the works with a professor in San Francisco. He asked for volunteers and for investor introductions. A commenter who listened to the posted sample said the separated voices sounded thin, hollow and a little garbled.
Eight months later the patent issued. Application 18/386,165 was filed on November 1, 2023, claiming a provisional filed exactly one year earlier, and granted on March 19, 2024 to AudioFocus, Inc of Oakland, California, with Mobin as sole inventor. It covers the simulator rather than the earpiece.
After that the record thins. The waitlist page, last captured in August 2024, offered 10 percent off a purchase. The technology page has been captured several times since, and the Internet Archive content digests for it are identical from July 18, 2024 through March 15, 2026.
The market did not wait. In August 2024 Phonak, a Sonova brand, launched the Audeo Sphere Infinio around a chip dedicated to running a separation network in real time. The same year the FDA authorized Apple software that turns AirPods Pro into an over the counter hearing aid, the first such authorization, granted through the De Novo pathway on a 118 subject study.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| US patent 11,937,073 | Filed Nov 1, 2023 as application 18/386,165, claiming provisional 63/421,425 of Nov 1, 2022. Granted Mar 19, 2024 to AudioFocus, Inc, Oakland, CA. Sole inventor Shariq Mobin. The claims cover simulating rooms to curate synthetic training data, not the hearing aid itself. | Public record |
| Which category it is in | 21 CFR 800.30 defines the term: “A hearing aid is any wearable device designed for, offered for the purpose of, or represented as aiding persons with or compensating for, impaired hearing.” FDA guidance draws the other line: “PSAPs are not intended to compensate for hearing impairment.” AudioFocus markets to patients and the hearing impaired, so what it described is a regulated hearing aid, not a personal sound amplification product. | Public record |
| What an OTC sale requires | Sold direct to adults 18 and over for perceived mild to moderate loss, without a licensed person, it falls under 21 CFR 800.30: output at or below 111 dB SPL, or 117 with input controlled compression; distortion plus noise at or below 5 percent; self generated noise at or below 32 dBA; latency at or below 15 ms; bandwidth to 250 Hz and 5 kHz; 10 mm minimum distance from the eardrum; a user adjustable volume control; OTC labeling. | Public record |
| FDA footprint | Air conduction hearing aids are Class I and exempt from premarket notification (21 CFR 874.3300); wireless ones are Class II with special controls and also exempt (874.3305), so no 510(k) is expected. Registration and listing are required of a distributing manufacturer, and no AudioFocus entry appears, searched Sep 23, 2026. | Not found |
| Clinical evidence | No registered trial and no PubMed indexed paper. The pilot study the founder described in July 2023 has produced no public protocol, site or result. | Not found |
| Public funding and filings | No NIH or NSF award. No Form D, no EDGAR company match and no full text hits for the name, searched Sep 23, 2026. | Not found |
| Market claim | The company says 37M US adults have hearing loss and only 8M use hearing aids. NIDCD reports 37.5 million with some trouble hearing and 28.8 million who could benefit. The direction matches; the 8M figure is not sourced on the company pages. | Company-stated |
Read plainly: the problem is well posed and the science behind it is live, but the evidence that AudioFocus turned it into a product stops at a prototype and a patent on a training method. That patent protects how the model was taught, not what a patient would wear. Everything downstream rests on company statements, most of them more than two years old.
What to watch
- An FDA establishment registration and device listing, the first hard signal that a hearing aid is being distributed in the United States.
- Which lane the company picks. Over the counter under 21 CFR 800.30 puts a 15 ms latency ceiling on a separation network running in the ear; the prescription lane under 21 CFR 801.422 avoids those electroacoustic limits but puts a licensed person between device and buyer.
- Any result from the San Francisco pilot study, now three years without a public protocol or outcome.
In their words
“We just got our behind-the-ear (BTE) hardware prototype running and already have several excited patients.”
Shariq Mobin, founder and CEO, Hacker News, July 2023 · Founder
“It’s a big deal because untreated hearing loss is associated with social isolation & depression and while 37M people have hearing loss in the United States, only 8M use hearing aids.”
Shariq Mobin, Hacker News, July 2023 · Founder
“Clinically, a person might have ‘normal’ hearing but they can’t hear in background noise. That person deserves care too.”
AudioFocus website, archived 2025 · Company-stated
“To solve this we created a ray tracing auralization framework called Floyd. Floyd takes a 3D model of a room and patients’ head recordings to simulate millions of echoes.”
AudioFocus technology page, archived 2024 and 2026 · Company-stated
“Dr. Mobin admits that solving the problem of hearing speech in noisy places is a challenge, and his research is in the early stages.”
Kathy Fairbanks, Patch, December 2019 · Independent
“The company designs state-of-the-art deep learning algorithms that focus on sounds using binaural cues, just like the human brain.”
UC Berkeley IPIRA listing for AudioFocus · Public record
“We have enjoyed helping people hear better, but have made the decision to complete a voluntary market withdrawal of our Hearing Systems as our business direction changes, effective June 8th, 2023.”
Andrew Song, CEO of Whisper, reported by Hearing Tracker, June 2023 · Independent
Related companies
Sources
- Public recordUS Patent 11,937,073
- Public record21 CFR 800.30, over-the-counter hearing aid controls
- Public record21 CFR 874.3300 and 874.3305, hearing aid classifications
- Public recordRegulatory Requirements for Hearing Aid Devices and Personal Sound Amplification Products
- Public recordQuick Statistics About Hearing
- Public recordFDA authorizes first over-the-counter hearing aid software
- Public recordAudioFocus invention listing, inventors and founding year
- Public recordCapture index and content digests for audiofocus.io
- IndependentHearing Aids That Improve Hearing of Speech in Noisy Environments
- IndependentWhisper AI announces voluntary market withdrawal of Whisper Hearing Aid System
- IndependentRestoring speech intelligibility for hearing aid users with deep learning
- FounderHearing aids that work in noisy places, founder post and comments
- CompanyAudioFocus home, technology, team and waitlist pages
- CompetitorPhonak elevates sound quality with new Infinio portfolio
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
