RealityRewi.red

Updated: a 58-year-old singer with ALS sang a scale through a brain implant

The singing result now has a patient, an age and 384 electrodes behind it. The locked-in patient and the iPhone decoder are still one sentence each.

Reality Admin//Nov 16, 2026//5 MIN READ
Updated: a 58-year-old singer with ALS sang a scale through a brain implant

This post was first published on 16 November 2026 as a developing story, before the research was presented. It was updated on 17 November 2026 with the findings as presented and reported since. One of the three results now has real detail. The other two do not, and the gaps are marked.

The patient who sang is a 58-year-old woman with ALS. She is a trained singer, known in the study as T21, and she could no longer speak intelligibly when she enrolled. A team at the University of California, Davis implanted 384 electrodes in her brain. With them she controlled the pitch of a synthetic voice well enough to sound like she was asking a question, or sounding happy, and then she sang a musical scale. Afterwards, Maitreyee Wairagkar told the New York Times, the former singer cried tears of joy.

That is the most concrete thing to come out of the Society for Neuroscience's brain-computer interface session at Neuroscience 2026 in Washington. The other two results, a locked-in patient and a decoder that runs on a phone, are where they were on Monday: one sentence each in a press release.

What we know now

The detail on the singing comes from Carolyn Y. Johnson's report in the New York Times, published on 14 November and syndicated since, and from MassLive's coverage on 17 November. Both call the result preliminary. The Times is blunter: "the finding is preliminary and unpublished." There is no paper and no preprint for it yet, so there is no number for how close to the notes she got. Wairagkar's case for why pitch matters: "They're not just for essential communication, they're expressive too; you can show more of your personality."

A correction to the original post. On Monday this post matched the singing announcement to the brain2voice 2.0 preprint and asked whether the singer was a new participant or the man from the group's 2025 Nature paper. The answer is that the session covered two separate experiments. T21 is a new participant, and her singing is not in the preprint. The man is Casey Harrell, a climate activist who was 45 when four arrays with 256 electrodes were implanted in 2023. His recordings are behind the other half of the talk.

That half is about intelligibility. In the 2025 Nature study Harrell spoke through a synthetic copy of his own voice with a 10-millisecond delay, and, as Scientific American reported then, could hum notes in three pitches. The Times notes the catch: listeners could accurately transcribe "only a bit more than half of his spoken words." In the new work, described in a preprint, the team reanalysed the same dataset and could "increase the intelligibility of words by up to 94%," according to the Times. That fits the brain2voice 2.0 preprint this post cited on Monday, which reported listener word errors falling from 43.75% to 5.24%. So the preprint match holds for this part. Harrell, for what it's worth, has used the system to produce nearly 2 million words over two years.

Outside researchers quoted by the Times liked it. "I think this is really clinically meaningful improvement and not just impressive machine learning," said Christian Herff of Maastricht University. Cindy Chestek of the University of Michigan went further: "This system, for the first time, is good enough." The Times describes Chestek as not involved in the new work. She is also the moderator of the session that presented it and the source of the pre-written reaction in the society's release. That is not a conflict, but it is worth knowing who is doing the cheering.

Mariska Vansteensel of University Medical Center Utrecht gave the Times the necessary caveats. The intelligibility gains come from data already collected, so the decoder still has to prove itself in real time and in people with different degrees of speech loss. And for any of this to leave the lab, she said, the hardware has to be fully implantable. Today's participants are wired to a computer through ports in the skull.

Still unanswered

The locked-in result from Daniel Rubin at Massachusetts General Hospital is still the one to watch, and it is still a single sentence: a person in a completely locked-in state "was able to communicate independently for more than a year" with an interface that decodes intended movement. No coverage this post could find adds anything. How the participant communicates, how fast, how accurately and what "independently" covers are all open. So is whether this answers the old argument. The 2022 Nature Communications case was real communication from a completely locked-in man, but it was slow and needed recalibration through the day.

The phone decoder from Chethan Pandarinath at Georgia Tech and Emory is in the same state. The society says it decodes accurately and reliably "even on an iPhone." Nothing published since says whether it is the FETTUCCINE system from his group's November preprint, and nothing says whether an implant user has run it live. The preprint's own numbers, a word error rate as low as 4.39%, come from recorded data from one person replayed through the model.

On peer review the answer is simple: none of the three presented results has been through it. The singing is unpublished. The intelligibility gains and FETTUCCINE are preprints. The locked-in work has no public paper this post could find. The brain2voice preprint's disclosure section also still stands: several authors hold patent applications licensed to a neurotech startup, and some have consulted for or worked at Neuralink.

The press conference ran from 11:30am to 12:30pm Eastern today. As of this update no reporting from inside the room on the Rubin or Pandarinath talks had appeared, and this post won't guess at it.

A woman who used to sing for a living sang a scale again and cried. It is the best moment of the session, and it is the result with the least data behind it.

Sources: Society for Neuroscience release (EurekAlert), FETTUCCINE preprint (arXiv), Brain2voice 2.0 preprint (bioRxiv), Scientific American, Nature Communications.

Sources: The New York Times (via The Star), MassLive (via Yahoo News), Society for Neuroscience release (Medical Xpress).

// RELATED_SIGNALS

More from the feed