The Australian startup Fluent, founded at the University of Melbourne, is working on a new type of brain-computer interface (BCI) aimed at restoring communication for people with severe speech impairments, including those suffering from motor neuron diseases and multiple sclerosis.
Unlike the most well-known invasive BCI systems, which require implanting electrodes directly into the brain through neurosurgery, Fluent’s device is intended to be placed under the skin but above the skull. This approach falls under the category of “sub-scalp” interfaces and is positioned as a compromise between the safety of non-invasive methods and the precision of deep implants.
According to the project’s co-founder and biomedical engineer Tim Mahoney, the signals obtained this way could be comparable in informativeness to data traditionally gathered only through intracranial electrode implantation. If this is confirmed in future trials, the technology could lower the barrier to widespread adoption of BCI systems.
The device sits above the motor cortex—the brain region responsible for controlling muscles, including those involved in speech articulation. When a person attempts to speak, neurons generate distinct patterns of electrical activity. The developers compare these signals to unique “QR codes,” where each movement of the tongue, lips, or jaw corresponds to a unique neural pattern.
The idea is that even in patients who have lost the ability to speak, attempts to utter words persist, and the brain continues to produce corresponding signals. The device captures a sequence of these patterns and transmits them to a machine learning system, which reconstructs the intended utterance and converts it into text or synthesized speech.
To train the model, the team gathered a large dataset of brain activity related to speech. Experiments used caps with 144 electrodes placed over the motor cortex. Participants either spoke words, mimicked speech, or mentally envisioned phrases.
As a result, they compiled one of the largest English-language databases of neural signals associated with speech. In collaboration with Japanese researchers, the machine learning system demonstrated the ability to identify the correct phrase from a set of 128 options with up to 96% accuracy.
The authors emphasize that this concerns preliminary trials, not a finished clinical product. Any brain-computer interface requires extensive regulatory review, comparable to medical device certification, before it can be brought to market.
At the same time, the developers note that Australia could serve as a convenient venue for clinical trials due to its regulatory environment and tax incentives for research and development. This could potentially accelerate the transition of the technology from lab tests to practical medical applications.
Competition in the BCI field is intensifying: trials and commercialization of various systems are already underway in China and the United States, including fully implantable solutions for cursor control and speech decoding. Against this backdrop, Fluent is betting not on maximum signal precision but on accessibility and reduced medical risks.
The developers openly admit that their approach lags behind invasive systems in signal quality, but they plan to compensate for this through artificial intelligence methods, including contextual speech restoration using language models.
If such systems prove effective, they could carve out a distinct niche—mass-market BCI devices with low risk profiles, focused not on neurosurgical precision but on practical rehabilitation and everyday use.
About the Author
