The Brainy Necklace expresses the unspoken words of its wearer

There may soon be a powerful new tool for people who can’t speak. British scientists have developed a necklace that detects words spoken quietly by the wearer and converts them into audible synthetic speech.

Created at the University of Cambridge, the device is intended for people who have lost their speech due to injury, throat surgery or conditions such as stroke, cerebral palsy or Parkinson’s disease. It could also be used by people speaking in noisy environments, to amplify the sound of their voice without also amplifying background noise.

The necklace itself takes the form of a strip of bamboo fiber fabric that wraps around the wearer’s neck. It incorporates a strain sensor—located at the front—made of two copper electrodes flanking a flexible layer of electrically conductive graphene ink. The latter is screen-printed onto the fabric.

In the current prototype, the sensor is connected by cable to a power source and a computer. A commercial model could be completely autonomous.

Close-up view of the collar strain sensor

As the user silently pronounces different words, tiny vibrations in their throat cause the woven fabric to stretch and contract on a microscopic scale. In doing so, an orderly network of tiny cracks open and close in the overlying graphene layer.

This opening and closing action causes a fluctuation in the electrical resistance of the graphene, which manifests itself as fluctuations in an electric current flowing through the graphene from one electrode to the other.

By analyzing the telltale strength and pattern of these fluctuations, specially trained machine learning algorithms on the computer are able to determine what word is being spoken. That word is then audibly pronounced by a speech synthesizer.

The algorithms used by the choker (pictured) were trained on the speech of several volunteers with different accents – all the volunteers said certain key words that the software learned to identify, and then used to identify other words

Thanks to the high sensitivity of the strain sensor, the algorithms can identify each word relatively easily. The system is thus able to maintain a speech decoding accuracy of 95.25% while using 90% less computing energy than other so-called silent voice interfaces.

“By combining the ultra-high sensitivity of the sensors with highly efficient machine learning, we have developed a device that we believe could help many people who have difficulty with speech,” says lead scientist Dr Luigi Occhipinti.

An article about the research was recently published in the journal Flexible electronics npj.

And if you think the choker necklace is the first wearable silent voice interface… well, it’s not. Previous examples include a voice recognition collar, headphones, and glasses.

Source: University of Cambridge