A team led by Professor Shu Kai from the Department of Neurosurgery at Tongji Hospital, affiliated with Huazhong University of Science and Technology, in collaboration with the university's research team, has made a significant breakthrough in thought-to-speech technology.

For the first time, they have used magnetoencephalography (MEG) combined with AI algorithms to construct a Chinese MEG speech dataset and achieve non-invasive decoding of Chinese vocabulary. This milestone marks a crucial step in enabling speech-impaired individuals in China to communicate using Chinese.

"In simple terms, this technology falls within the realm of speech brain-computer interfaces, aiming to read 'intended speech' from brain signals and output it through external speech devices," explained Professor Shu. "It is particularly beneficial for patients who have lost their speech abilities due to conditions like strokes or amyotrophic lateral sclerosis (ALS)."

The challenge lies in accurately identifying corresponding vocabulary from complex, variable brain magnetic signals. To address this issue, the research team developed an intelligent algorithm using a "dual-advisor" approach, in which both "text" and "synthetic speech" guide the AI model's learning.

"The 'dual-advisor' method means that during AI model training, the model learns the relationship between MEG signals and corresponding text (semantic features) while also referencing the synthetic speech of the vocabulary (acoustic features)," Shu elaborated.

Experiments demonstrated that this intelligent algorithm achieved a decoding accuracy of 46.21 percent in classifying 48 vocabulary words. Additionally, the research highlighted the critical roles of the temporal lobe and sensorimotor cortex in Chinese speech generation, paving the way for future studies of neural mechanisms.