AI gives the voice back to the paralyzed by reading their “thoughts”: “Almost 3 out of 4 words have been deciphered”

2023-08-27 19:39:07

This is a serious reason for hope for paralyzed people deprived of speech. Two American scientific teams have developed devices capable of reading and deciphering brain activity related to language with unprecedented precision and responsiveness. Their promising work has been published in two published studies Wednesday, August 23 in the leading scientific journal Nature.

“It is now possible to imagine a future where we can give a paralyzed person the possibility of having a fluid conversation by allowing him to say everything with sufficient precision to be perfectly understood”, predicts Francis Willett, neuroscientist at the prestigious Stanford University (California) and creator of one of these two direct neural interfaces, a device that connects the brain to a computer.

The first device can convert electrical signals from the brain into text, at the rate of 62 words per minute. The second interprets this cerebral activity in 78 words per minute and transforms it into speech emitted by a synthetic voice. A natural human conversation is defined by the use of 160 words per minute and the rates of interpretation errors are still high, but the progress made is colossal.

Eighteen years later, Ann speaks again

They are especially concrete for several patients who participated in these cutting-edge studies. Victim of Charcot’s disease and its progressive degeneration, the American Pat Bennett was transplanted at the age of 67 with a device made up of silicone electrodes and implanted in the region of the brain that controls speech. In parallel, the Stanford team trained an AI with machine learning algorithms so that the software recognizes and interprets signals specific to the patient’s brain.

Two configurations were tested: with a vocabulary of 125,000 words and with a sample of 50 words. “Nearly three words out of four were deciphered correctly,” explained Francis Willett during a press conference. “For those who have lost speech, it means they could stay connected to the outside world, maybe even continue to work and maintain family and friendship relationships,” he hopes.

For Ann, a 40-year-old participating in a study at the University of California, San Francisco, the advances made possible by AI materialized in the reconstruction of her voice lost for eighteen years after a stroke. The teams of neurosurgeon Edward Chang used a less invasive implant placed on the surface of the cortex of the brain. An artificial intelligence trained by them then read their “thoughts”, while another converted them into an avatar that mimicked facial expressions and spoke with an artificial voice.

Even more impressive: the researchers took advantage of the voice-mimicking capabilities of the AI ​​and used footage from her wedding to reproduce Ann’s voice as closely as possible. “However, this technology should be made wireless,” the patient suggested to the scientists. Miniaturization, testing on different profiles and industrialization… several other steps remain to be taken before a massive distribution of these devices that break the silence.