People in a completely locked-in state are awake and aware but cannot move any muscle, including the ones that move their eyes. That leaves even eye-tracking devices, which many people with amyotrophic lateral sclerosis (ALS) use to spell out words, with nothing to follow. On Tuesday in Washington, Daniel Rubin, a critical care neurologist at Massachusetts General Hospital (MGH), will report that a person in that state communicated independently for more than a year through a brain implant that decodes intended movement.
Rubin is one of three speakers at a Society for Neuroscience press briefing at Neuroscience 2026, the society’s annual meeting, which runs from 14 to 18 November. The briefing covers brain-computer interfaces, which record brain signals and turn them into words or commands, for people who have lost the ability to speak or move. Maitreyee Wairagkar of UC Davis will present a synthetic voice that captures changes in pitch, and Chethan Pandarinath of Georgia Tech and Emory University will present a portable system that decodes brain activity into written words, even on an iPhone. All three groups belong to BrainGate, an academic consortium that has been testing implanted brain-computer interfaces in people with paralysis since 2004.
“The ability to decode and convert brain activity into spoken or written words has life-changing potential for people with paralysis,” said Cynthia Chestek, a biomedical engineer at the University of Michigan and the briefing’s moderator. “The work presented today brings brain-to-voice and brain-to-text technologies closer to reaching people outside of clinical trials, which is extremely exciting.”
UC Davis: a voice with pitch
In June 2025 a UC Davis team, with Wairagkar as first author, reported in Nature that an implant could turn a paralysed man’s attempted speech into a synthesised voice within a fortieth of a second. The man, who has ALS, has four electrode arrays, 256 electrodes in total, in the part of the brain that controls speech. He could change the voice’s intonation to ask a question and sing short melodies. Listeners understood more than half of his words, against a few per cent without the implant.
The team has since built a new decoder, the software that turns brain signals into sound, called Brain2voice 2.0. Evaluated on recordings from the 2025 study, its voice was transcribed by untrained listeners with a word error rate, the share of words they got wrong, of 5.24 per cent. The best previous result on the same recordings was 43.75 per cent, so the error rate fell roughly eightfold. The study is a preprint, posted before peer review and revised on 13 November, and will be presented as a poster at NeurIPS, the machine learning conference, at its Sydney venue on 8 December. The University of California has a patent pending on the decoder and lists it for licensing.
The team’s brain-to-text system is already in near-daily use at home. In June it reported in Nature Medicine that Casey Harrell, who has ALS, used it for more than 3,800 hours over nearly two years, operating it on his own on a near-daily basis and averaging 56 words a minute.
Earlier MGH work
Rubin’s group published two earlier studies this year. In March it reported in Nature Neuroscience a system that maps the letters of a QWERTY keyboard onto attempted finger movements. Two BrainGate participants, one with ALS and one with a spinal cord injury in the neck, used it at home after calibrating it with as few as 30 sentences. One reached 22 words a minute with a word error rate of 1.6 per cent.
In Cell Reports, the group decoded attempted speech from a person with ALS who had long been unable to speak, was locked in and depended on a ventilator. Sentence accuracy fell short of results in people who can still make speech sounds.
Earlier Emory work
An Emory-led team, with Pandarinath as senior author and co-authors from MGH and UC Davis, has decoded speech in people paralysed by a stroke in the brainstem. In a preprint first posted in February and revised on 10 November, the team decoded sentences that two participants mouthed without sound, using as little as a single 64-electrode array. Median word error rates on a 125,000-word vocabulary were 19.4 and 10.6 per cent, within the 1 to 24 per cent range reported for implants in people with ALS. With standard adjustments, decoding stayed stable for more than two years in one participant.
Pooling data across people
The three groups also work together. In July, researchers from all three, with BrainGate colleagues, posted a preprint describing a decoder trained jointly on data from six people with speech implants. For every participant it decoded speech more accurately than models trained on that person alone, cutting word errors by more than half on average. After adjusting it with fewer than 200 sentences from a new user, it got fewer than 7 words in 100 wrong.
Signal: Why it matters
The companies building these implants work with these labs. Paradromics’ Connect-One study of its Connexus implant is led by David Brandman of UC Davis Health, with Rubin’s team at Mass General Brigham as one of its other two sites. The MGH Translational Research Center has clinical research support agreements with Neuralink, Precision Neuro, Synchron, Medtronic and others. UC Davis neuroscientist Sergey Stavisky is an inventor on Stanford intellectual property licensed to Blackrock Neurotech and Neuralink. The University of California is offering Brain2voice 2.0 for licence.
Leigh Hochberg, who leads the BrainGate trial, said in March that by advancing the science “we make it that much easier for industry to create the final form of implantable medical devices for our patients.”
In October Neuralink said, in a company post that has not been peer-reviewed, that a participant in its speech study reached 75 words a minute of decoded text with a 5.8 per cent word error rate in their first week. Shorter setup and phone-based use, if they hold up, would make these implants easier to use at home.
Not disclosed
What Wairagkar’s talk adds to the 2025 results. How many people used the iPhone system, and what speed and accuracy the systems in the MGH and Emory talks reached. Whether the iPhone system decodes on the phone itself. Whether the participants in the earlier MGH and Emory studies are the people in Tuesday’s talks. Whether Brain2voice 2.0 has been run live with a participant.
Forward Radar: What to watch
What the three teams show at the briefing on 17 November, from 11:30am to 12:30pm Eastern time.
Whether the MGH and Emory results appear as papers or preprints.
The Brain2voice 2.0 poster at NeurIPS on 8 December, and whether the decoder is licensed.
Sources
Primary:
- Society for Neuroscience via EurekAlert: Brain-computer interfaces: A new voice for people with paralysis, 14 November 2026
- EurekAlert: Neuroscience 2026 newsroom, Washington, 14 to 18 November 2026
- StatPearls, NCBI Bookshelf: Locked-in syndrome
- Wairagkar et al., Brain2voice 2.0: High-performance voice synthesis brain-computer interface, bioRxiv, version 2, 13 November 2026
- NeurIPS 2026: Brain2voice 2.0 poster, Sydney, 8 December 2026
- University of California: Brain2voice 2.0 technology licensing listing, UC Case 2026-909-0
- Wairagkar et al., An instantaneous voice-synthesis neuroprosthesis, Nature, June 2025
- UC Davis Health: First-of-its-kind technology helps man with ALS speak in real time, June 2025
- UC Davis Health: Brain-computer interface enables independent, accurate communication for man living with ALS, 15 June 2026
- Mass General Brigham via EurekAlert: Brain computer interface enables rapid communication for two people with paralysis, 16 March 2026
- Jude et al., Decoding intended speech with an intracortical brain-computer interface in a person with long-standing anarthria and locked-in syndrome, Cell Reports, 2026
- Nason-Tomaszewski et al., Restoring intended speech in people with dysarthria from pontine stroke using an intracortical brain-computer interface, medRxiv, version 2, 10 November 2026
- Fogg et al., A generalizable speech neuroprosthesis, bioRxiv, July 2026
- Paradromics via Business Wire: Paradromics achieves real-time speech with Connexus brain-computer interface, 14 September 2026
Cross-reference to prior InsideBCI coverage:
- 16 June 2026: A man with ALS has used his speech brain implant at home for nearly two years, mostly on his own
- 16 March 2026: BrainGate BCI restores typing at 22 words per minute for paralysed patients
- 16 September 2026: Paradromics’s first Connexus participant used a brain implant to speak her own words
- 29 October 2026: Neuralink says an ALS patient reached 75 words a minute with its speech implant