Fortell Announces AI 2.0 Noise-Separation Update for Its AI Hearing Aids

The company started from a problem hearing aids have long failed to solve. Conventional hearing aids raise the volume of the frequency bands a hearing test shows a person struggles with, but the clatter of plates and laughter from the next table sit in the same bands as human speech, so the device cannot tell them apart and turns both up. It is like a single radio volume knob making a loud room louder. On its explainer page, Fortell says its chip first splits the sound reaching the microphones into voices and background noise, then amplifies only the voices while reducing the noise. The company argues that other AI hearing aids on the market are closer to a switch that recognizes a scene and flips settings to a restaurant or car mode, without changing the sound itself.

Doing that inside the ear required computing power. According to the company, Fortell's custom AI chip, manufactured by TSMC, handles 235 times the compute of a conventional hearing aid processor and performs up to 100 billion operations per second while keeping the time from sound in to sound out under 10 milliseconds. The company stresses that number because a longer delay makes the speaker's lips and voice fall out of sync and conversation feel awkward. With AI left on, the battery lasts more than 24 hours, and a portable charging case adds three more charges for over 72 hours of use. A tap on the right device toggles AI mode, the left device adjusts volume, and an iPhone app lets the wearer set how much background noise to let through.

The company's clinical research page, which METAL reviewed, lists two studies. In a November 2025 speech intelligibility study, the company reported eliminating up to 98% of background noise and improving signal-to-noise ratio by 17 dB, while the leading AI hearing aid used as a comparison managed 1.9 dB under the same conditions. In challenging noise, wearers understood 80% of words with Fortell versus 31% with the comparison device. In a May 2026 preference study, 95% of participants said Fortell made it easier to follow speech than the top five competing products, and Fortell was rated better in 99 of 100 listening scenarios. The company says the studies were blinded and randomized, were run with NYU Langone, Mount Sinai and Weill Cornell, and that the preference study was published in the International Journal of Audiology.

The AI 2.0 announcement also disclosed how the device improved, down to the method. The company collected new training audio in restaurants, cafés, construction sites, trains, buses, cars and reverberant offices and living rooms, retrained its neural network on it, and revised several signal-processing algorithms. For validation it placed the same hearing aids on a KEMAR acoustic manikin shaped like a human head and played 40 background-noise scenes such as cafés and train stations plus 40 scenes with three people talking at once. It applied four metrics to each of two programs, All Voices, which keeps voices from every direction, and Front Voices, which emphasizes the voice in front, for 16 comparisons in all. Every one of the 16 improved on average and 15 were statistically significant. In background-noise conditions, the signal-to-noise gain in All Voices rose from 8.9 dB to 10.6 dB and in Front Voices from 9.1 dB to 10.8 dB.

The limits the company wrote down itself are also clear. The AI 2.0 results are laboratory figures measured on a manikin, not a clinical trial that split real wearers between AI 1.0 and AI 2.0. One intelligibility metric in Front Voices mode fell short of statistical significance, and the device is a prescription product for adults with hearing loss that is currently sold only in the United States. According to reports, since launching the product in December 2025 Fortell has sold only through its own audiology clinic in Manhattan and by appointment in Greenwich, Connecticut and Palm Beach, Florida, with more clinics planned in Boston, San Francisco and Palo Alto. The biggest constraint on expansion, it says, is hiring audiologists to staff them.

The founder's motive came before the technology. According to reports, co-founder and CEO Matthew de Jonge first imagined a better hearing aid more than a decade ago after watching his grandparents withdraw from people because they could not follow conversations, and he said, "In 2015, it felt like science fiction." He saw the possibility again when the pandemic pushed meetings onto Zoom and he watched AI erase background noise. "Zoom is the easiest version of the problem, and the hearing aid is the gnarliest," he said. He founded the company in early 2021 with former colleagues Andrew Casper and Igor Lovchinsky, and reports say it has raised $163 million from investors including Founders Fund, Thrive Capital and Valor Equity Partners at a $740 million valuation.

Hearing loss does not end with the ears. The company explains that as people age, the signals the ear sends to the brain become sparse and off-key, and the harder the brain works to interpret them, the more processing capacity it burns. That is the sequence in which someone exhausted by following conversation in loud rooms starts avoiding gatherings, and the avoidance turns into isolation. Early customers named in reports include the comedian Steve Martin and KKR chairman Henry Kravis, and de Jonge said, "This should not be a status symbol. We want this to be available to as many people as possible." METAL has reported on the outlook for camera-equipped AirPods becoming Siri's eyes and on Google Research's AI that finds depression-related markers in wearable data, and the trend is the same: the device in the ear is becoming AI's next home.

What Fortell has shown is that the AI hearing aid has become a measurable object rather than a marketing phrase. An announcement that discloses which sounds were gathered for retraining, which manikin and how many scenes were used, and which metric failed to reach significance is rare in the hearing aid industry. While the large model companies argue over benchmark scores, AI worn on the body is judged by how many more words it lets you catch from the person across the restaurant table.