Neuralink Reports Early Speech Restoration Data: An ALS Participant Reached 75 Words per Minute at 5.8 Percent Word Error Rate
Early Neuralink speech restoration data: We are enabling speech to text and voice at conversational rates in participants with ALS. One participant achieved 75 WPM at 5.8% WER on a 125k-word vocabulary in their first week.
Context
If the numbers hold up beyond one participant, a brain implant decoding attempted speech at conversational speed would give people who lost their voice to ALS a practical way to talk again.
Neuralink builds brain computer interfaces: an implanted device records neural activity, and software models translate that activity into actions such as moving a cursor or, in this case, producing text and speech. ALS (amyotrophic lateral sclerosis) is a progressive disease that can take away a person's ability to move and speak. Speech decoding is one of the hardest targets for these systems because it needs both speed and a large vocabulary.
Deep dive
Neuralink posted what it called early speech restoration data on Wednesday evening. According to the company, it is enabling speech to text and voice at conversational rates in participants with ALS.
The headline figure comes from a single participant. Neuralink says this person reached 75 words per minute with a 5.8 percent word error rate on a vocabulary of 125,000 words, and that this happened in their first week of using the system. Word error rate measures the share of words the decoder gets wrong, so a lower number is better, and a large vocabulary makes the task much harder than decoding from a short list of commands.
These are company self reported results shared in a post, not a peer reviewed paper. The post does not say how many participants are in the speech work, how the 75 words per minute was measured, or how stable the results are over time. It also mentions voice output alongside text, without describing how the voice is produced.
The announcement drew wide attention on X, passing 860,000 views by Thursday morning, and Elon Musk reposted it. Until Neuralink publishes fuller data, the figures should be read as early and unverified by outside researchers.
What to expect next
The key thing to watch is whether Neuralink publishes the full method and results, including how many participants are involved and how accuracy holds up over weeks and months. Independent researchers will want to compare the 75 words per minute and 5.8 percent error rate with published academic speech decoding studies under the same conditions. Also watch for regulatory steps or expanded trials that would move speech restoration from early data toward a product patients can use.