Researchers in Russia have developed a speech decoding system that analyzes brain activity to interpret imagined speech. The announcement came via the press service of the National Research University Higher School of Economics and MGMSU Evdokimov.
Millions around the globe experience partial or complete loss of speech due to congenital conditions, brain injuries, and strokes. For some, assistive technologies help with lip reading or articulation reading, but people with paralysis may still face a daily communication barrier.
The team behind the project includes scientists from the Higher School of Economics and MGMSU Evdokimov who created a system that translates internal speech from neural signals. The approach relies on a machine learning model and is notably noninvasive because it does not require a large array of electrodes in the brain. In the study, researchers used a small set of electrodes ranging from five to nine contacts and drilled only small skull openings rather than removing substantial bone.
For training the neural network, participants read aloud a set of six sentences, each repeated 30 to 60 times with breaks in between. The sentences were designed with varied linguistic patterns and included consonant-heavy phrases, for example, a sentence like Shura walks wide in wide trousers. Each sentence contained 26 words. While the subject recited the sentences, the brain’s electrical activity was captured through the implanted electrodes.
Using these brain signals, the algorithm learned to map neural patterns to spoken content. The trained network demonstrated the ability to predict words with 55 percent accuracy for the first participant and 70 percent accuracy for the second participant. These predictions were based on data from a single six-contact sEEG probe and, in other cases, from an eight-contact ECoG strip.
The researchers intend to push this technology further with the goal of improving communication options for people affected by speech disorders. The early results show promise for expanding how imagined speech can be translated into real words, potentially enabling more natural and efficient ways to express thoughts for those who cannot speak aloud.