
For patients with amyotrophic lateral sclerosis (ALS), deteriorating speech can make conversations a challenge. New research into brain-computer interfaces is providing insight into how advanced computer technology might one day translate speech impulses into spoken words.
Researchers at the University of California, Davis, have built systems that identify neural activity associated with attempted speech.
In one experiment, a man with ALS and little-to-no speech capacity generated synthesised speech via an implant. In another, a woman with ALS who was once a singer was able to modulate pitch and generate a vocal scale through a brain-computer interface.
The results show progress in developing a technology that could imitate other aspects of human speech as well as words themselves.
How The Technology Works
The systems rely on electrodes implanted in the brain to record neural activity that correlates with attempts to speak. Software takes those signals and makes a computer-synthesised output.
Some brain-computer interfaces translate brain signals into text, and then into speech. Recent brain-to-speech systems seek to produce speech directly from neural signals, thereby retaining properties such as pitch, timbre, intonation, and rhythm.
There's no cure for ALS, nor will the technology restore your ability to speak like you used to. It simply circumvents certain damaged neural routes that control speech, so a computer can produce the voice for you.
They explored this strategy in a 2025 pilot study of an instant voice-synthesis neuroprosthesis that decoded activity in a user's brain to produce the person's own synthesised voice in real time.
Improving Speech Intelligibility
In the same UC Davis study, scientists demonstrated a brain-to-voice system in a man with ALS. The device produced synthesised speech from neural activity as he attempted to speak, allowing him to communicate through a computer.
Researchers subsequently reported improvements to the decoding algorithms that increased the intelligibility of synthesised speech to as much as 94% when previously collected neural data were reanalysed.
The result signals a major step in decoding speech-related brain activity. However, the improved performance has not yet established that the system can achieve the same level of intelligibility during real-time conversations.
The findings were reported for presentation at the Society for Neuroscience conference in Washington, DC.
A Singer Explores Pitch and Expression
In the same study by UC Davis researchers, a brain-to-voice system was used to help a man with ALS speak. The system translated neural activity into words when he tried to talk, using a computer to synthesise speech.
Later, the researchers said they improved their decoding algorithms, raising the intelligibility of the speech they generated from reprocessed neural recordings to 94%.
The result is a step forward in understanding brain signals linked to speech, but further research is needed to demonstrate that the system can achieve comparable intelligibility in real-world conversations.
The results were announced at the Society for Neuroscience meeting in Washington, DC.
Challenges Remain
At the end of the day, it is still at the research stage and, as with any brain surgery, calls for invasive brain surgery. The systems also require equipment outside the head, and researchers still need to assess their safety, accuracy, and performance across different subjects.
The observed improvements in the most recent analysis were based on data from previous studies. Additional tests are needed to determine whether similar performance can be observed in live speech.
The researchers will need to determine whether the technology works for people with varying levels of speech impairment and whether it can be implemented in usable systems.
Even at this early stage, the results suggest a future in which brain-computer interfaces can enable ALS patients to speak more naturally and regain some of the rich expression often lost when information is transmitted via text.











