It is expected that people who have lost their ability to speak "speak": brain-computer interface experiments decode human brain speech signals in real time for the first time

Chengdu Business Daily Red Star News

2024-05-17 17:33Posted on the official account of Red Star News of Chengdu Business Daily, Sichuan

In its latest issue, Nature Human Behavior publishes a new study on brain-computer interface (BMI) technology.

▲The "superior gyrus" of the human brain has not been explored in previous studies of speech decoding

Unlike Musk's brain-computer project, the brain science research team at the California Institute of Technology has developed a device that can decode speech information inside the human brain, but it needs to be trained before it can be used. Although the technology is still in its early stages and is only suitable for a few words, it is expected that in the future, it will be used in clinical applications to allow those who have lost the ability to speak with ideas.

The researchers implanted tiny electrode arrays in the brains of two spinal cord injury patients and placed these devices in an area of the brain called the supramarginal gyrus, which is associated with silent speech and determining whether words rhyme or not, and has not been explored in previous studies of speech decoding.

By giving participants a series of tasks to "imagine speaking," the researchers were able to capture and interpret speech signals in their brains in real time. According to Sylvia Malcesotti, a neuroscientist at the University of Geneva in Switzerland, figuring out the best place to implant a brain-computer interface in the brain is one of the key challenges in decoding internal speech.

This technology is not capable of reading the human mind. Each user's BCI device needs to be trained individually and is only effective when people are focused on a specific word. Despite its limitations, this research undoubtedly opens up a new path for future brain-computer interface technology and the application of artificial intelligence in the medical field.

The researchers first rehabilitated two quadriplegic participants for two weeks before starting to collect data. They trained the brain-computer interface device to recognize brain patterns that occur when participants say or think certain words mentally by using the words "battlefield," "cowboy," "anaconda," "spoon," "swim," and "phone," as well as two nonsensical words, "nifzig" and "bindip."

This training session lasted only about 15 minutes. Participants were asked to mentally "speak" the words displayed on the screen and repeat the process for each word to reinforce data collection. The brain-computer interface device then combines measurements of participants' brain activity with computer models to predict their internal speech in real time.

For the first participant, the BCI device was able to capture the unique neural signals of all words and accurately identify 79% of the words. However, for the second participant, the decoding accuracy was only 23%.

Christian Huff, a computational neuroscientist at Maastricht University in the Netherlands, believes that this result may be due to differences in how different people process internal speech. In addition, "previous studies have shown that the ability to control brain-computer interfaces varies depending on the ability to perform imaginative tasks." The study also found that 82%-85% of neurons that were internally speech-active were similarly active when participants vocalized, but some neurons were only active during internal-speech and could respond differently to specific words in different tasks.

Red Star News reporter Fan Xu Intern Gold

Edited by Pan Li and edited by Wei Kongming

View original image 20K

•