People with dysfunctional vocal cords might regain their voice function thanks to a device invented by a team of University of California Los Angeles (UCLA) engineers.

According to the researchers, the soft, thin, stretchy bioelectric device, which measures just over 1 square inch, can be attached to the skin of the throat.

Source: UCLA

There, the device detects movement in the wearer's larynx muscles and translates those signals into audible speech via machine-learning technology — with an estimated 95% rate of accuracy.

The patch-like device is comprised of a self-powered sensing component — one that detects and converts signals produced by muscle movements into high-fidelity, analyzable electrical signals, which are then translated into speech signals via a machine-learning algorithm. The device also features an actuation component, which turns the speech signals into the intended voice expression.

The researchers added that the sensing and actuation components both contain two layers — a layer of biocompatible silicone compound polydimethylsiloxane, or PDMS, featuring elastic properties, and a magnetic induction layer composed of copper induction coils. Between the two components, the researchers explained, is a fifth layer featuring PDMS combined with micromagnets that produce a magnetic field.

Further, the team used a soft magnetoelastic sensing mechanism to detect changes in the magnetic field caused by mechanical forces — specifically, the movement of the laryngeal muscles. The induction coils within the magnetoelastic layers help to produce the high-fidelity electrical signals for sensing applications.

During lab tests of the device — which measures 1.2 inches on each side, is just 0.06 inch thick and weighs roughly 7 grams — the patch technology was applied to eight healthy adults to collect data on laryngeal muscle movement. A machine-learning algorithm was then used to create a correlation between the resulting signals and certain words. Then, the team chose a corresponding output voice signal using the device's actuation component.

The device’s accuracy was demonstrated by having the participants pronounce five sentences — such as "Hi, Rachel, how are you doing today?" and "I love you!" — both out loud and voicelessly.

Based on their findings, the overall prediction accuracy of the model was about 94.68%, with the participants' voice signal amplified by the actuation component, thereby demonstrating that the sensing mechanism was able to recognize their laryngeal movement signal and subsequently match the corresponding sentence the participants intended to say.

The device is detailed in the article “Speaking without vocal folds using a machine-learning-assisted wearable sensing-actuation system,” which appears in the journal Nature Communications.