Scientists Are Closer to Turning Brain Signals Into Speech

A group of researchers has published a wide-ranging review of "brain-to-language decoding" — the science of reading brain activity and turning it into words, typed text, or a synthetic voice. The paper rounds up years of studies up to September 2026 across two very different setups: invasive recording, where electrodes sit inside the skull, and non-invasive recording, where sensors sit outside the head like an EEG cap. The invasive version gets better signals; the non-invasive version is safer but noisier.

The review groups the work into three tasks: articulated speech (actually trying to talk), inner speech (silently imagining words), and perceived speech (hearing someone else). Modern systems have moved well past recognising a handful of words. They can now generate full text, stream personalised speech in someone's own voice, and even animate a face to match. For people locked in by paralysis or ALS, that is the difference between a slow letter-by-letter spelling board and something closer to conversation.

The catch is practicality. The best results usually require brain surgery, and even then the systems need regular recalibration, user feedback, and careful control — decoding accuracy alone is not enough. Sessions can be tiring, and long-term studies show that staying accurate day after day is its own hard problem. Cost and clinical access are unresolved too.

The authors end with a five-level roadmap: from simple commands, to language, to meaning, to imagined scenarios, and finally to two-way exchange where a person both sends and receives thoughts. That last level is speculative. For now, think of this as a decade-scale medical project rather than a gadget heading to your phone — one that could one day restore a voice, and raise new questions about who gets to read your mind.