Brain Implants and Artificial Intelligence Turn Thoughts Into Intelligible Speech, Researchers Prove
Researchers demonstrate the ability to turn thoughts into speech — even mimicking the user's voice.
Researchers from Radboud University and the University Medical Center Utrecht (UMC Utrecht) have demonstrated the ability to turn brain signals into audible speech using a brain-machine interface and an artificial intelligence (AI) system — offering a claimed accuracy of 92 to 100 percent.
"Ultimately, we hope to make this technology available to patients in a locked-in state, who are paralyzed and unable to communicate," lead author Julia Berezutskaya explains. "These people lose the ability to move their muscles, and thus to speak. By developing a brain-computer interface, we can analyze brain activity and give them a voice again."
The team's work focused on non-paralyzed test participants fitted with temporary brain implants, who were given a list of words to read out loud while their brain activity was monitored. By analyzing these recordings, the team was able to create a direct map between activity and speech — feeding the data into AI models to turn the thoughts directly into speech.
"That means we weren't just able to guess what people were saying," Berezutskaya explains, "but we could immediately transform those words into intelligible, understandable sounds. In addition, the reconstructed speech even sounded like the original speaker in their tone of voice and manner of speaking."
In tests, the system proved able to decode individual words with a 92 to 100 percent accuracy, offering a chance level of eight percent. It's the ability of the system to reconstruct readings from sensorimotor brain activity sensors into intelligible speech which holds hope for those who have lost their own ability to speak — though the research comes with caveats.
"For now, there's still a number of limitations," Berezutskaya admits. "In these experiments, we asked participants to say twelve words out loud, and those were the words we tried to detect. In general, predicting individual words is less complicated than predicting entire sentences. In the future, large language models that are used in AI research can be beneficial.
"Our goal is to predict full sentences and paragraphs of what people are trying to say based on their brain activity alone. To get there, we'll need more experiments, more advanced implants, larger datasets and advanced AI models. All these processes will still take a number of years, but it looks like we're heading in the right direction."
The team's work has been accepted for publication in the Journal of Neural Engineering, with an online copy of the accepted manuscript available under open-access terms.
