In a major leap forward for neurotechnology and assistive AI, researchers have developed a brain implant that enables a paralyzed woman to communicate nearly in real time, using a synthetic version of her own voice. The system, known as a streaming speech neuroprosthetic, captures electrical activity from the brain?s speech centers and translates it into text and spoken words in about one second?a dramatic improvement over previous methods, which often took 20 seconds or more.
Published in Nature Neuroscience, the study describes how a paper-thin electrode array implanted on the surface of the brain decodes neural signals responsible for controlling speech muscles like the lips, tongue, and vocal cords. The woman, referred to as Ann, lost the ability to speak after a stroke disrupted her brain?s connection to these muscles. While her body couldn?t produce sound, her brain still attempted to send the commands?as ?silent speech.?
An AI model trained on over 23,000 imagined sentences was able to recognize these signals and output speech using recordings of Ann?s pre-injury voice, giving her back not only communication, but a sense of identity and embodiment.
On average, the system can process 47 words per minute, with trials reaching even higher speeds. Though not perfect, it allows for natural, flowing conversation?a game-changer for individuals living with speech loss.
The core innovation is in the AI?s streaming architecture, which continuously processes brain signals in small chunks, minimizing lag and allowing for dynamic interaction. Unlike earlier neuroprosthetics that struggled with delay and limited vocabulary, this model adapts over time and shows potential to scale.
The team also demonstrated that the AI could function with different types of implants, including invasive and non-invasive systems, and could generalize across data collected from other individuals. This suggests broader applicability for people with various types of speech impairments.
Future improvements aim to increase decoding accuracy, expand vocabulary, and even incorporate emotional tone, pitch, and inflection into the synthetic voice.
For now, Ann?s progress is already profound. ?Hearing her own voice in near-real time increased her sense of embodiment,? said lead researcher Gopala Anumanchipalli from the University of California, Berkeley. ?This is more than speech. It?s about restoring presence.?