Why they built it
Difficulty hearing in noisy environments like restaurants was the number-one complaint of more than 300 million hearing aid users globally, because conventional aids amplify everything instead of isolating nearby voices. Founder Shariq Mobin, a UC Berkeley auditory neuroscience PhD and former Google Brain engineer, thought the brain's use of echo statistics to judge distance could be replicated in machine learning software to spatially filter sound. The team saw a large opportunity in the gap between 37 million US adults with hearing loss and only 8 million using hearing aids.
Why it didn't work out
AudioFocus demonstrated 2-3x improvements on standardized QuickSIN and ANL hearing tests but never moved beyond a behind-the-ear research prototype built around a credit-card-sized Variscite VAR-SOM-MX8 board. The 10ms latency limit, tiny power budget, and miniaturization required for a consumer wearable needed custom silicon far beyond what a four-person team with $393K in total funding could build. The company quietly wound down around 2023 when Mobin moved to a new employer, with no paying customers and no shutdown announcement.
What went wrong
Most important first.
- 1
Embedded-compute wall blocked miniaturization Technology
Mobin publicly named 10ms latency, small power budget, and real-time performance constraints in 2020, while the prototype ran on a research-grade Variscite board and he estimated in-ear form factor was 5-10 years away.
- 2
Severe undercapitalization for a hardware-AI medical device Capital
Total documented funding was $393K with no Series A documented, versus $5-20M in engineering and fabrication costs alone to design the custom chip needed for on-device inference.
- 3
Tiny team lacked medical-device commercialization coverage Team
At peak the company had four employees including hardware architect Dr. Reza Kassayan, and Mobin's neuroscience and Google Brain background did not extend to chip architecture, FDA strategy, or audiologist channel development.
- 4
Grant-funded research pace replaced product velocity Other
The NIA/Johns Hopkins A2 Pilot Grant and collaborations with Stanford, Johns Hopkins and University of the Pacific produced clinical validation, but after four years traction was still several excited patients in an ongoing San Francisco pilot study.
- 5
Prescription-only regulation blocked direct sales Regulation
Until the FDA created the OTC category in August 2022, hearing aids were Class II devices requiring audiologist fitting and prescription, forcing reliance on an established clinical channel.
- 6
Entrenched incumbents held chip and channel advantages Competition
Phonak, Oticon, Starkey, Widex and Signia collectively controlled roughly 90% of the market with proprietary chips, miniaturization expertise, FDA relationships, and audiologist distribution networks.
What could be different today
- Hearing-aid-class edge AI chipsets like the Qualcomm QCC5100 series and Syntiant NDP120 support on-device neural inference within sub-10ms latency and ~1mW power budgets, versus the credit-card-sized Variscite VAR-SOM-MX8 board AudioFocus required.
- FDA's August 2022 OTC hearing aid ruling removed the audiologist prescription and fitting requirement for mild-to-moderate loss, AudioFocus's target population.
- FDA's 2024 clearance of Apple's AirPods Pro 2 Hearing Aid feature establishes that software-defined hearing assistance on consumer earbuds and smartphone compute can clear the regulatory path.
- Pre-trained voice separation models like Meta's Demucs and open Conv-TasNet implementations plus synthetic datasets like Microsoft's CRUSE replace building a custom acoustic ray-tracing training pipeline from scratch.
- The OTC market launched in October 2022 with Jabra Enhance and Eargo shows consumer willingness to pay $200-$1,500 for self-fitted devices targeting the 29 million US adults with diagnosed loss but no aid.
