Does communication require voice or any form of movement? It turns out that brain waves can be enough, as they can be transformed into words by means of advanced AI algorithms. Students from our Faculty's Neuron Research Club are working on such a solution.

The key element of the 'EEG2Text' project is electroencephalography (EEG). It is a completely safe and non-invasive diagnostic method for detecting the brain's bioelectric activity. It uses electrodes placed on the scalp to record signals from neurons. In medicine, it is used, among others, for diagnosing epilepsy or sleep disorders.

However, brain waves can also be used for communication. Students from the Neuron Research Club at our Faculty are working on such a solution, along with Tymon Drop, this year's high school graduate. The 'EEG2Text’ project is still in its very early stage of implementation, and yet it has qualified for the national final of the global program Red Bull Basement 2026.

How to catch an electric whisper?

The students' idea involves using artificial intelligence to process data obtained during EEG, which would enable a certain degree of communication. Such a solution, not requiring the use of voice or any gestures, would allow individuals struggling, for example, with complete paralysis to interact with their surroundings.

“In our research, we use 32-electrode caps that enable quite advanced analyses. The greatest challenge in the entire project is the extremely low ratio of received neuronal signals to the accompanying electrical noise,” says Kamil Wróbel, the leader of the AI section at Neuron.

It turns out that our brain generates a constant electrical noise, resulting from physiological processes, emotions, or external stimuli. The signal corresponding to a specific thought about a word is extremely subtle and difficult to catch in this flood of information.

“Moreover, the human mind rarely focuses on one thing. Even if we try to 'think' of a specific word, reflections about, for example, a planned lunch or events of the past day may appear in the background. The task of artificial intelligence in this project is to filter out this informational chaos and spot patterns that the human eye is unable to catch on the chart,” explains Kamil Wróbel.

Thoughts are your fingerprints

A fascinating, yet problematic element of the project is that EEG signals are unique to each individual – much like fingerprints. The way our brain “encodes” concepts is unique, and this is due to differences in anatomical structure and individual cognitive processes.

“As a result, it is not possible to create a single, universal AI model that would immediately work for everyone and catch the electrical signal passing through neurons from noise. It is necessary to adjust and calibrate it properly. Each new user would need to undergo a brief process of “learning” the system, in which they would contemplate specific words, enabling the algorithm to align with the specifics of their brain waves,” explains Kamil Wróbel.

Another issue to solve is the way we shape our thoughts. Researchers around the world distinguish several approaches that are currently the subject of discussions. They are:

• inner speech: when we “hear” words in our head,
• imagined speech: imagining the movement of the tongue and facial muscles necessary to pronounce a word,
• visual imagery: imagining a specific object, e.g. a suitcase.

“The problem is that each of us can do it differently. Moreover, while it is easy to visualize 'a suitcase', abstract concepts already require a different approach, probably based on imagined articulation. The coherence of instructions for the test subjects is crucial here, so as not to introduce an additional error into the already complex data,” explains Kamil Wróbel.

Two paths lead to success

The “EEG2Text” project could therefore be developed in two ways. The first path – the most ambitious one – is a direct translation of thoughts into words. The team has already tested models on publicly available datasets, achieving promising results of 70-80% accuracy with a dictionary covering numbers, letters, and basic objects.

“In this case, however, we must deal with the issue of data leakage, which undermines the credibility of many published scientific results regarding EEG signal interpretation,” explains Kamil.

The second path is more pragmatic. It uses the fact that the EEG signal generated when we imagine movement, e.g. of the right or left hand, is much stronger and easier to read than the signal of pure thought about a word.

“Therefore, the system could work as a virtual keyboard, where the user ‘clicks’ letters or selects directions using motion gestures, supported by AI autocomplete, enabling efficient sentence construction. Even if it is a slower method, for people deprived of other forms of contact with the world, it would be an invaluable communication bridge,” emphasizes Kamil Wróbel.

The “EEG2Text” idea qualified for the national stage of the Red Bull Basement 2026 program and although it didn't win the main prize, the Neuron team is not planning to stop working. The students want not only to create a working prototype, but above all to conduct thorough research.

“Currently, the field of EEG suffers from the lack of a large amount of high-quality data. We would like to conduct research on a larger scale to gather the largest possible database of EEG signals, which will allow better generalization of AI models,” says Kamil Wróbel.

While the project focuses on textual communication, the potential of the technologies researched at Neuron is much broader. Recognition of emotions, control of prostheses movement, or modern biometric security systems are just some of the potential evolution directions.

```