Wearable graphene-based strain sensor can detect and broadcast silently spoken words

A wearable “smart” neckband for speech recognition has the potential to redefine the field of silent speech interface (SSI), researchers say, thanks to integrated ultra-sensitive textile strain sensor technology.

When verbal communication is hampered, for example in places with a lot of background noise or when a person suffers from a speech disorder, SSI systems are a cutting-edge solution, allowing verbal communication without vocalization. It is therefore a type of electronic lip reading using human-computer interaction.

In new research, led by the University of Cambridge, an overlying structured graphene layer is applied to an integrated textile strain sensor for robust speech recognition performance, even in noisy environments.

Worn around the neck, the smart collar captures micro-movements in the throat, which are then picked up by the strain sensor as an electrical signal and transmitted to brain-inspired computer models for processing and speech recognition. It can even pick up silently spoken words and play them back, which could help someone who is unable to speak after larynx surgery, for example.

The unique structure of the smart collar features ordered cracks on graphene-coated textiles. The structured graphene layer significantly improves the sensitivity of the strain sensor. It can dynamically respond to micro-movements of the throat, enabling the capture of information-rich speech signals. These signals are then processed by a computationally efficient neural network, with a record accuracy of 95.25% in speech decoding.

The results, reported in the journal Flexible electronics npjoffer a promising, non-invasive solution for practical and portable SSI systems, paving the way for transparent, natural and silent communication in various contexts.

The proposed SSI system is robust and capable of decoding a wide range of words, while quickly adapting to new users and vocabularies. The technology was recently presented in a live demonstration at the IEEE Biosensors 2024 conference, attracting the attention of more than 180 attendees at the event.

The research was led by Dr. Luigi G. Occhipinti, research director of intelligent electronics, biosystems and AI, and head of the Occhipinti Group in the Electrical Engineering Division of the Department of Engineering and the Cambridge Graphene Center.

Dr Occhipinti said: “Our user-friendly smart necklace demonstrates a remarkable ability to perform effectively in real-world scenarios with users of different genders, geographic and ethnic backgrounds, on novel and potentially ambiguous words of varying lengths and familiarity, and at varying reading speeds.

“Our SSI system works with high accuracy and high computational efficiency to distinguish speech from different users from different types of noise, caused by sensor imperfections, external environment and/or users’ body movements while wearing the device.

“In addition, the manufacturing method of our ultrasensitive textile strain sensor technology is biocompatible, simple, inexpensive and scalable. It is also adaptable to prolonged use and can withstand more than 10,000 stretch-release cycles while maintaining stable and reliable electrical functionality.

“Simply put, the synergy between our sensor design and neural network optimization sets a new standard in wearable silent voice communication technologies, delivering a comfortable smart neckband with revolutionary potential.”