Using their own voice to train artificial intelligence, new brain-computer interface technology allows ALS patients to "speak" again

The University of California, Davis Health Center has developed a new brain-computer interface (BCI) that can convert brain signals into speech with an accuracy rate of up to 97%, the highest among similar systems currently available. The team implanted sensors in the brain of a man whose language ability was severely impaired due to amyotrophic lateral sclerosis (ALS, commonly known as Lou Gehrig's disease). After starting the system, the man was able to say what he wanted to say within a few minutes. The relevant research was published in the New England Journal of Medicine on the 15th.

Designed to restore speech to people who are unable to speak due to neurological diseases, the brain-computer interface interprets the user's brain signals when they speak and converts them into text that is read aloud by a computer.

The research team recruited a 45-year-old ALS patient to participate in the clinical trial. The patient had symptoms such as weakness in his arms and legs and difficulty understanding speech.

In July 2023, the team implanted a dedicated BCI device in the patient. Four microelectrode arrays were placed in the left precentral gyrus, an area of ​​the brain responsible for coordinating speech. These arrays can record brain activity from 256 cortical electrodes.

The patient used the system in both a prompted and impromptu conversation environment. In both cases, speech decoding was done in real time, with the decoded words quickly displayed on the screen and the system constantly updated to ensure it was operating accurately. The research team also trained the voice synthesized by the software using samples of the patient's previous voice, so that when the computer "read" aloud the words the patient was thinking of, it sounded exactly like the patient's voice before the disease.

Tests showed that in the first speech data training, the system took 30 minutes to achieve 99.6% accuracy with a vocabulary of 50 words. In the second training, the potential vocabulary increased to 125,000 words. With only 1.4 hours of additional training data, the BCI achieved 90.2% accuracy with this greatly expanded vocabulary. After continuing to collect data, the BCI's accuracy remained at 97.5%.

What ALS "freezes" is actually the nerve cells that control the movement of the whole body, so patients will gradually lose the ability to stand and walk, and will also lose control of the muscles used to speak. Brain-computer interface technology has been constantly changing in the past two years, giving ALS patients the opportunity to "speak" again. In addition, brain-computer interfaces also have important uses in restoring movement and stimulating nerves. People are looking forward to seeing this technology move from the laboratory to clinical applications, and even the emergence of consumer-grade brain-computer products. But before that, the formulation of norms and standards for the collection of EEG information in this field needs to be extremely cautious.