Silent Speech Let One ALS Participant Control Smart Devices Without Timing Cues

A man with amyotrophic lateral sclerosis used an implanted brain-computer interface to operate smart devices by silently miming command words, without waiting for a screen or audio cue. The peer-reviewed study, published November 11 in Communications Biology, reports median decoding accuracy of 97.1% across 18 experimental sessions.

The work tackles a practical obstacle in assistive brain-computer interfaces: a person should be able to issue a command when they choose, not only inside a window set by a computer. Removing the external timing signal could make environmental control more private and less dependent on a caregiver or display.

How the system worked

The participant was enrolled in the Johns Hopkins CortiCom clinical trial and had two 64-channel electrocorticography grids placed on the brain surface. ECoG records electrical activity beneath the skull but above brain tissue. The implant covered cortical regions associated with speech and upper-limb movement.

When he silently mimed a command, the system first searched for a characteristic change in beta-band activity to detect the speech attempt. It then analyzed high-gamma activity around that event and classified the signal. The setup separated a 14-command smart-device vocabulary from a six-command navigation vocabulary.

Examples included light, music, television, camera and temperature, plus navigation actions such as up, left, enter and exit. Some interface states required repeated activation words, such as silently miming “NAVI” twice. This was device control through a small set of discrete words, not unrestricted thought reading.

What the numbers show

Across 18 real-time sessions, median classification accuracy for the 14 device commands was 97.1%, compared with a 7.14% chance level. The system delivered a median 11.86 correct decodes per minute. The researchers also reported median online false-positive and false-negative rates of zero for event detection, although medians can hide session-to-session variation.

The key advance is not a larger vocabulary. It is letting the participant initiate a silent command at a self-chosen moment while the system detects when decoding should begin.

Earlier work with the same participant found that speech-based commands could remain stable for months without recalibration. The new paper focuses on silently mimed speech and combines event detection with command classification. It complements NewTqnia’s earlier coverage of a brain implant that decoded speech and gestures together.

Why self-paced control matters

Many laboratory systems tell a participant exactly when to attempt a word. That simplifies analysis because the decoder knows where the relevant signal should appear. Everyday use is harder: the system must distinguish an intentional command from rest, unrelated movement and ordinary inner thought.

This study addressed only part of that challenge. The participant intentionally moved his speech articulators without producing audible speech. The method did not decode free-form inner speech or infer arbitrary private thoughts. Repeated activation commands also served as a safeguard against accidental device changes.

For someone whose movement and speech are progressively limited by ALS, reliable environmental control could reduce dependence on others for small but frequent actions. Lights, entertainment, cameras and temperature controls are modest examples, yet they map directly to daily autonomy.

The next valuable evidence would be continuous home operation. Researchers need to show that the decoder remains available for long periods, ignores unintended activity, recovers safely from errors and can be configured around each user’s abilities. Until then, this is a strong feasibility signal, not a ready consumer system.