The human voice, with all its power to mesmerise audiences, woo loved ones and irritate neighbours, is a delicate thing. Each person’s distinctive sound is produced when air from the lungs causes the vocal cords, folds of muscle tissue in the larynx, to vibrate. These vocal cords can easily get damaged by stress, infections, or overuse. It is not just overzealous performers who are at risk of straining their voiceboxes—according to a 2005 study, 30% of the population will experience a voice disorder at some point in their life.

In a study published in the journal Nature Communications this week, a group of bioengineering researchers from the University of California, Los Angeles, have put forward an attractive solution. They have designed and tested a soft patch that can be stuck onto a person’s neck, where it will pick up muscle movements and, with the help of machine-learning algorithms that process the signals, translate them into audible speech.

Even though the team’s device is an early prototype, it has the potential to offer a substantial improvement on current alternatives. When a person loses their voice today, the easiest fix is to resort to typing, texting, or writing notes to communicate. Typing can be slow and inconvenient, says Jun Chen, the paper’s lead author, and writing legible notes is only possible in good lighting. More sophisticated solutions such as electrolarynxes, external devices held against the throat to produce the vibrations necessary for speech, can require special training to use, and surgical interventions are often off-puttingly invasive. A patch would theoretically be able to clear all these hurdles.

The mechanism behind Dr Chen’s device is a principle known as the magnetoelastic effect. When magnetic nanoparticles are embedded into soft materials like elastic or silicone polymers, their magnetic properties can change as the material is stretched. That’s because each deformation causes the particles either to rotate or move relative to one another, changing the magnetisation of the material. When embedded in a patch with a hem of copper coils generating a background magnetic field, the movements of the particles can be accurately captured as variations against this background.

When the throat muscles move under the silicone patch, the resulting magnetic-field variations can also be converted into electrical signals. In a test with eight participants, the researchers captured the signals arising as the subjects spoke and lip-synched five different sentences (including: “Hope your experiments are going well!”, “Merry Christmas!” and “I love you!”). They then trained a machine-learning model to recognise the distinct shapes of the electrical signals associated with each sentence. This algorithm was then able to predict which of the five sentences the participants spoke—whether aloud or in silence—with more than 90% accuracy.

The design of the patch brings additional benefits. In addition to stretchiness, the patch remains sticky on sweaty skin, and can be used continuously for 40 minutes without heating up.

There is a way to go yet. For now the device can only recognise the five phrases it was trained on. Plus, individual differences in vocal folds means the algorithm has to be personalised to each user. To make it practical at scale, the researchers will need to collect a lot more data. ■