Neuralink Restores Voice for ALS Patient
5 min read
When Terry speaks, even his relatives barely understand him; the nerve disease ALS has taken his speech. In a Neuralink video, a brain implant now outputs his intended sentences in his own voice. The company has not yet published metrics from the underlying study, while university research groups disclose their figures.
Key takeaways
- An ALS patient outputs intended sentences in his own voice. Neuralink presents Terry in the video “Speaking With The Mind” as the third participant in its speech study; output runs with Grok Voice from SpaceXAI.
- Public metrics for the VOICE study are missing. Neuralink has so far published neither accuracy nor vocabulary size nor latency. The company cites a conversation pace of 140 words per minute as its goal.
- Other research groups publish long-term data. ALS patient Casey Harrell used his UC Davis speech implant for more than 3,800 hours at home and averages 56 words per minute.
Terry turns thought into speech
On September 18, 2026, Neuralink released the video “Speaking With The Mind” (in the video). It shows ALS patient Terry, who in 2024 received a diagnosis of ALS with bulbar onset, so symptoms begin with speaking and swallowing. His speech is so impaired that his family can barely understand him.
According to the video, Terry is the third participant in Neuralink’s speech study. His implant sits in the speech motor area and uses around 3,000 channels. He first trained the algorithm by mouthing speech, then by merely thinking the words. Neuralink says the sentences come out in his own natural voice. According to Neuralink, output runs with Grok Voice from SpaceXAI.
Terry’s wife says in the video that her husband had just become a partner in his firm. Because of the speech disorder he could barely communicate with clients and colleagues anymore. He finally went onto long-term disability and ended his career. In the closing scene she sees the words he is thinking appear on screen and hears him say “I love you.”
What is a brain-computer interface? A brain-computer interface is an implant that picks up signals from the brain and makes them usable for computers. In speech applications it sits in the speech motor area, the region that controls the movements for speaking. Software decodes these signals and outputs the words the user is thinking as text or as spoken language.
Video “Speaking With The Mind”. Source: Neuralink (YouTube)
What the VOICE study is designed to measure
The VOICE study is registered as an early feasibility study under identifier NCT07224256 (in the registry). According to the registry, it began on October 3, 2025; six participants are planned. The sole study site is the University of Texas Southwestern Medical Center in Dallas; the principal investigator is Nader Pouratian.
As primary endpoints, the registry lists safety measures only, namely the share of participants with device-related and procedure-related adverse events after twelve months. Speech performance, measured in correctly recognized words per minute, is a secondary endpoint. In May 2025, Neuralink received FDA Breakthrough Device designation for restoring communication in severe speech impairment.
In late January 2026, Neuralink named conversational speed of 140 words per minute as the goal of the VOICE study (on the company site). At the same time, the company reported 21 participants in its studies worldwide.
Clinical metrics such as accuracy, vocabulary size, or latency have not yet been published by Neuralink for the VOICE study, and Terry’s implantation date is also unknown (according to Tesla North). A PubMed search for speech publications with a Neuralink affiliation returned zero hits on September 28, 2026. Below the video, Neuralink notes that the devices are investigational and are not approved by the FDA or other authorities.
Long-term data from other research groups
In mid-June 2026, the UC Davis research group published results in Nature Medicine. An ALS patient used his speech implant at home for just under two years and more than 3,800 hours, with no researchers present. According to the clinic, the participant is Casey Harrell (on the clinic website).
On average, Harrell reaches 56 words per minute. When he attempts to speak, the system decodes his speech with more than 99 percent word accuracy.
On September 14, 2026, Paradromics reported real-time communication via its own Connexus implant (at Paradromics). The first participant, a woman in her sixties from Michigan with a progressive motor neuron disease, used the system for speech and text communication in real time.
Researchers at Stanford Medicine were already able to decode inner speech from the motor cortex in August 2025, that is, words that are only thought and not spoken. At the same time, they developed a password system that blocks any decoding of inner speech until the user first thinks a password (at Stanford Medicine).
In the related Cell study, the word error rate for real-time decoding of inner speech with a vocabulary of 125,000 words ranged from 26 to 54 percent (in the study).
Rules for neural data
In early November 2025, UNESCO member states adopted the first global recommendation on the ethics of neurotechnology (at UNESCO). UNESCO warns that neural data could reveal thoughts, feelings and reactions and be shared without consent.
At the end of September 2025, the MIND Act was introduced in the US Senate to task the FTC (Federal Trade Commission) with an inquiry into how neural data is handled (in Congress). The bill has since been pending in the Commerce Committee.
People affected in Germany and Europe
In Germany, according to the ALS Center at Charité, 6,000 to 8,000 people live with ALS. About 2,000 new diagnoses are added each year (at the ALS Center at Charité). In 30 to 40 percent of those affected, symptoms begin with speaking and swallowing.
With INTRECOM, an EU project is developing an implant meant to capture what a person with speech loss wants to say (at INTRECOM). The clinical study is planned at UMC Utrecht and at University Hospital Graz; so far, UMC Utrecht is seeking participants.
For Terry’s wife, being able to communicate again means the world, as she says in the video. Whether the film becomes a therapy that also reaches people with ALS in Germany hinges on data covering accuracy, long-term safety and the protection of brain signals. Neuralink has yet to provide precisely this data.
Frequently asked questions
Are the systems already approved?
Neuralink notes under the video that the devices are in trials and are not approved by the FDA or other authorities. Participants’ experiences may not reflect all participants or future outcomes.
How many people in Germany live with ALS?
According to the ALS Center at Charité, 6,000 to 8,000 people in Germany live with ALS, with about 2,000 new diagnoses each year. In 30 to 40 percent of cases, symptoms begin with speaking and swallowing.
What rules apply to neural data?
UNESCO member states have adopted the first global recommendation on the ethics of neurotechnology. UNESCO warns that neural data could reveal thoughts, feelings and reactions and be shared without consent. In the U.S. Senate, the MIND Act, which would task the FTC with an investigation into the handling of neural data, is before the Commerce Committee.
Read more on MyBusinessFuture
MyBusinessFutureAI in medical devices: the new EU obligationsMyBusinessFutureBitkom: 90 percent of companies want a GDPR reformMyBusinessFutureFor manufacturers who want more – How GTIA membership strengthens the entire partner networkMore from the MBF Media Network
Digital ChiefsAccessibility belongs in the architecture of IT leadershipSecurityTodayDeepfake CEO fraud: How executives can fend off the voice cloneImage source: AI-generated (September 2026)
