An international team of researchers - including members from the University of Cambridge, Beihang University, Beijing Tsinghua Changgung Hospital, Tsinghua University, and other institutions - has developed a wearable, comfortable and washable device called Revoice that could help people regain the ability to communicate naturally and fluently following a stroke, without the need for invasive brain implants.
Schematic of a textile-based strain-sensing choker. Two channels are aligned with the carotid artery and center of throat, respectively. Each channel consists of a two-terminal crack-based resistive strain sensor surrounded by a polyurethane acrylate (PUA) stress isolation layer. The top right SEM image shows the spontaneous ordered crack structure of the graphene coating. Image from: Nature Communications
Wearable silent speech systems hold significant potential for restoring communication in patients with speech impairments, but seamless, coherent speech has so far remained elusive and clinical effectiveness unproven. The Revoice system (also referred to as an intelligent throat, IT) integrates throat muscle vibration sensing and carotid pulse monitoring with large language model (LLM) processing to support fluent, emotionally expressive communication in real time. Ultrasensitive textile strain sensors embedded in a soft choker capture high-quality signals from the neck area and feed them into a token-level decoding pipeline, enabling continuous, delay-free speech reconstruction. In tests with five stroke patients with dysarthria, the system achieved low word and sentence error rates and a marked increase in user satisfaction, suggesting a promising non-invasive route to restore more natural communication.
At the core of Revoice is a graphene-based textile strain sensor that translates minute neck and throat movements into robust electrical signals. The choker incorporates two sensing channels at the front and side of the neck to track both laryngeal muscle activity and carotid artery pulsation, ensuring rich coverage of the mechanical events associated with silent or impaired speech. A thin graphene layer is screen-printed directly onto a stretchable textile substrate; when strained, this conductive layer develops ordered microscopic cracks along the textile’s stress concentration lines. These controlled cracks modulate the electrical resistance of the graphene network in response to very small deformations, allowing the device to pick up the subtle skin vibrations produced when users mouth words or attempt to speak.
To preserve signal quality while the device is being worn in everyday conditions, each graphene sensing region is bordered by a rigid isolation frame with a higher Young’s modulus than the surrounding fabric. This mechanical design confines most external stretching and bending to the isolation structure, so that only a very small fraction of macroscopic strain is transmitted into the active sensing area. As a result, the internal graphene–textile interface remains soft, elastic, and closely coupled to the skin, while the electrical signal remains largely free from motion artifacts caused by donning, doffing, or natural head movement. Comparative tests of devices with and without this isolation layer demonstrate that the isolation architecture markedly suppresses strain transfer between channels and reduces crosstalk, which is critical for reliable decoding of throat-specific motion.
The graphene sensor itself is engineered to be highly anisotropic, responding primarily to strain along the axis corresponding to throat expansion and contraction. Under uniaxial stretching in the physiologically relevant frequency range, the textile-based graphene strain gauge shows a linear response and a large resistance change even for strains as low as about 0.1%, with a gauge factor on the order of 100. This combination of low detection threshold, strong amplification, and directional selectivity enables the system to resolve the tiny, rapid mechanical signatures of silent speech that conventional textile strain sensors would likely miss.
The graphene layer is produced using a printable ink derived from graphite, ethyl cellulose, and isopropyl alcohol, processed through high-pressure homogenization and subsequent centrifugation to remove unexfoliated particles. This yields a stable graphene dispersion suitable for screen printing onto elastic fabrics at scale. Once printed and integrated into the choker, the resulting sensors are not only sensitive but also robust, durable, and washable, retaining performance after repeated stretching and laundering cycles. The textile base ensures comfort, breathability, and good skin conformity, which are essential for long-term clinical or home use by stroke patients and others with neurological speech impairments.
By coupling this advanced graphene-textile sensing platform with embedded LLM agents, Revoice forms a complete pipeline from throat mechanics to meaningful language output. One AI agent reconstructs words from fragmented, low-level signals, while another uses emotional cues and contextual information to expand short, effortful phrases into full, natural-sounding sentences that better reflect the speaker’s intent. The result is a wearable, non-invasive system that can turn a few mouthed words and physiological signals into fluent, emotionally nuanced speech, offering a realistic path toward restoring independence, dignity, and social participation for people living with dysarthria and related conditions.
Dysarthria affects around half of stroke survivors and often leaves people knowing exactly what they want to say but physically unable to articulate it clearly, which can be exhausting and deeply frustrating for both patients and their families. In one trial example, a participant’s mouthed phrase “We go hospital” was automatically expanded by Revoice into “Even though it’s getting a bit late, I’m still feeling uncomfortable. Can we go to the hospital now?”, using heart‑rate signals and contextual cues such as time of day to infer urgency and emotional state. As project lead Luigi Occhipinti has emphasized, the aim is to give people their independence back, because restoring communication is fundamental to preserving dignity and supporting recovery after stroke.
