UMI launches /sign for real-time Deaf–hearing communication on Bittensor

Universal Motion Intelligence said Friday it launched a Bittensor AI network for motion-to-meaning intelligence and introduced /sign, a real-time two-way communication app between Deaf and hearing people that uses a standard camera.

Most AI still stalls on bodies in motion. UMI is putting Deaf–hearing communication on an open Bittensor leaderboard so the best sign models win in public — a rare accessibility bet that treats gesture, face, and timing as first-class signal.

On Friday, 2 October 2026, Universal Motion Intelligence said it launched an AI network on Bittensor and introduced /sign, its first application, for real-time communication between Deaf and hearing people. The GlobeNewswire page is titled “UMI Launches Bittensor AI Network for Motion-to-Meaning Intelligence, Starting With Sign Language.” The line under the headline says Subnet 78 opens a decentralized model competition as UMI introduces /sign. The page stamps October 02, 2026, 08:08 ET, which is 8:08 a.m. Eastern. The dateline is San Francisco. The source line is Universal Motion Intelligence. Bittensor, as the release describes it, is a decentralized machine-learning protocol with its own blockchain. A protocol, here, is the shared set of rules the network runs on. A blockchain is the public record of what that network does. Subnet 78 is the slice of the network the company says it operates. Those lines are the wire’s.

What the release says is still hard for machines. They have become better at words and at pictures. Understanding movement, expression, timing, and physical context is still harder. UMI says it is building an open intelligence layer around that gap. The uses it names run from communication to facial understanding, human intent, activity recognition, and embodied AI. Embodied AI, in that list, means software that has to deal with a body moving in the world, not only text on a screen. Motion-to-meaning, the company’s phrase, means combining what a camera sees — the hands and the body, the face, the language, and the timing — so a machine can treat the movement as meaning. Those lines are UMI’s.

What /sign is supposed to do. The release calls it the first commercial application, built for real-time two-way communication between Deaf and hearing people. The system combines 3D visual capture, skeletal and gesture modeling, facial expression, contextual language understanding, multilingual voice, and live inference. 3D visual capture means the camera is reading the shape of the person signing, not only a flat picture. Skeletal modeling is a map of the joints the software follows. A gesture is a hand shape or a movement that carries a sign. Contextual language means the words around that sign, so the system is not translating one handshape on its own. Inference, here, is the model producing the translation while the conversation is happening. The release says /sign translates signed communication into spoken language, and spoken communication into captions and signing. It uses a standard camera. It does not use gloves, wearables, or specialized sensors. A wearable is a device a person puts on, such as a glove with sensors in it. Those lines are the release’s. The release does not print a measured accuracy.

How the contest is supposed to work. Independent AI teams can deploy models and compete on measurable performance. UMI operates Subnet 78. The release says that subnet was formerly Vocence, a voice-AI subnet, before the company expanded from voice into motion. A miner, on this network, is a team that contributes a model. A validator scores those submissions against defined benchmarks. A benchmark is a public test with a score. The release calls the setup King of the Hill: the best models rise, and every score is public. The stated design is that a model’s performance can be judged in the open, rather than inside one company’s closed system. Those lines are UMI’s. The release does not print a current score, a prize, or the name of a model that is winning.

Koyuki Nakamori, co-founder and head of AI at UMI, is quoted on the release. “Sign language is one of the hardest motion problems in AI, because meaning travels through hands, body, face and timing all at once,” Nakamori said. “That’s why we built /sign on open competition. It’s a King of the Hill dynamic: the best models rise, and every score is public.” The release says Nakamori was formerly head of AI at the OpenTensor Foundation, which the release calls the core organization behind Bittensor. That quotation is Nakamori’s, in the release.

The hearing figures, as UMI cites the World Health Organization. The release says the WHO estimates that more than 1.5 billion people currently live with some degree of hearing loss, including around 430 million people with disabling hearing loss that requires rehabilitation. Disabling, in that phrase as the release uses it, means the loss is serious enough that the person needs rehabilitation. By 2050, the release says, nearly 2.5 billion people are projected to have some degree of hearing loss, and more than 700 million would require hearing rehabilitation. The release also says the global economic cost of unaddressed hearing loss is estimated at almost US$1 trillion a year. A trillion is a thousand billion. Those figures are the WHO’s, as UMI cites them. They are not a count of people using /sign. The release points to WHO pages for those figures, and for two more: around 95.1 million children aged 5 to 19 live with hearing loss, and nearly 80 percent of people with disabling hearing loss live in low- and middle-income countries. Nearly 80 percent is about 4 in 5. Those lines are the release’s citations. The release does not say /sign has been measured in those populations.

What the release says U.S. law already requires, and what it says /sign is not. Accessible communication, it says, is a legal requirement in many settings. Under the Americans with Disabilities Act, state and local governments and businesses that serve the public must ensure effective communication with people who have communication disabilities, and must provide appropriate auxiliary aids and services when necessary. An auxiliary aid, in that sentence, is the extra help that makes the conversation work. The release says that, depending on the context, those aids may include a qualified sign-language interpreter, video remote interpreting, real-time captioning, and accessible communication technologies. Video remote interpreting is an interpreter on a screen. The release also says healthcare organizations may have effective-communication duties under Section 504 of the Rehabilitation Act and Section 1557 of the Affordable Care Act. UMI says it is developing /sign as one technology inside that broader accessibility ecosystem, not as a replacement for a qualified interpreter where one is required or appropriate. Those lines are the release’s. The release does not say a court or a regulator has ruled that the app meets those duties. It does not say the Food and Drug Administration has cleared /sign.

The rest of the product family, as the release lists it. UMI says the same approach applies to other motion problems. The roadmap names /face for facial understanding, /intent for body language and human intent, /flow for crowd movement, /move for embodied AI, /motion for human activity recognition, and /trust for behavioral credibility, plus developer tools through /sdk and /api. An SDK, a software development kit, is the package another developer uses to build on the system. An API is the door a program calls when it wants a result. The release calls this a roadmap. It does not say the later names are on sale on Friday. The company site, umi.vision, labels /sign as live, /face and /intent as next, and /flow, /move, /motion, /trust, /sdk, and /api as planned. Those status words are the site’s. The site also says six pieces — vision, motion, expression, language, voice, and inference — can run as one engine, live, in under a second, and that the live pass travels over encrypted streams. “Under a second” and “encrypted” are the site’s lines. The release says real-time and live inference. It does not print a measured delay.

When a person can try it, as the release states it. Early access requests, clinical pilot inquiries, and launch partnerships are open now at umi.vision. The web app is rolling out in the coming week. A TestFlight beta opens on October 15. TestFlight is Apple’s program for trying an iPhone app before it is on the public store. The App Store launch is November 15. A clinical pilot inquiry is a request to ask about a pilot. The release announces that the inquiry line is open. It does not publish results from a clinical trial, and it does not name a hospital that has already run one.

Who the about box says the company is. Universal Motion Intelligence says it builds the intelligence layer between human movement and machine understanding, translating motion, gesture, expression, and context into meaning a machine can act on. It says /sign is the first product, for real-time communication between Deaf and hearing people, developed through open model competition on Bittensor Subnet 78. The media contact is Sam Allcock at sam@digital24.com. The about box points readers to umi.vision. The company site says the corporation develops the models, the intellectual property, and the products, and that /sign is its first commercial platform. The release does not print a price, a funding round, or a customer.

The picture is the launch graphic for the announcement. A dark panel on the left carries a magenta stripe, the UMI mark, the words UMI /sign, the line “motion → meaning,” and “Bittensor SN78.” The right side is stacked indigo and violet bands. White type reads “Real-time sign language,” “Deaf ↔ hearing,” and “standard camera · no gloves,” with “open model competition” underneath. Three thin rings, pink, orange, and blue, sit on the word language. The frame does not print a calendar date. It does not show a person signing, and it does not show a score from Subnet 78.

In plain terms, Universal Motion Intelligence said on Friday, from San Francisco, that Subnet 78 on Bittensor is a public contest for models that turn human movement into meaning, and that the first app is /sign. The company says a standard camera, without gloves or extra sensors, can turn signing into speech and speech into captions and signing, using the hands, the body, the face, and the timing together. Teams submit models, and validators score them in the open. The hearing figures are World Health Organization numbers as UMI cites them. The company says /sign sits among other accessibility tools and is not a stand-in for a qualified interpreter where one is required. Early access is open at umi.vision, with a web app in the coming week, a TestFlight beta on October 15, and an App Store date of November 15. The release does not print a clinical result, and it does not print a regulatory clearance.