Voice disorders, affecting nearly 30% of the population at some point in their lives, significantly impact communication and quality of life. These disorders stem from various causes, including vocal fold polyps, keratosis, paralysis, nodules, adductor spasmodic dysphonia, and post-laryngeal cancer surgery. Current solutions like electrolarynges or tracheoesophageal puncture are often inconvenient, uncomfortable, or invasive. This necessitates the development of a comfortable, non-invasive, wearable device to assist communication during voice disorder recovery. Existing research utilizing materials such as PVDF, gold nanowires, and graphene for flexible loudspeakers and throat sensors shows promise but faces limitations in terms of material stretchability, power requirements, and omnidirectional movement detection. The inherent limitations of these materials in capturing the complex, three-dimensional movements of multiple laryngeal muscle groups, coupled with issues like lack of water resistance and potential temperature rise during use, highlight the need for a more advanced solution. This research addresses these shortcomings by introducing a novel wearable sensing-actuation system based on soft magnetoelasticity.

Prior work on medical devices for voice disorders has explored flexible loudspeakers and wearable throat sensors. Piezoelectric materials like PVDF, while offering mechanical-to-electrical signal conversion, have limited material selection and pose safety concerns due to required driving voltages. Gold nanowires and graphene, known for conductivity and flexibility, are suitable for resistive sensors but usually need external power sources and lack stretchability, hindering the accurate detection of complex throat movements during phonation. These existing technologies primarily detect vertical throat movements and fail to capture the critical parallel deformations essential for speech production, particularly crucial for patients with voice disorders who cannot use their vocal folds. The lack of stretchability affects comfort and adhesiveness, while issues like water resistance and temperature rise further compromise functionality and usability. This study aims to overcome these limitations by presenting a superior technology leveraging soft magnetoelasticity.

This study developed a lightweight (7.2 g), flexible, and adhesive wearable sensing-actuation system attached to the throat. The device comprises symmetrical sensing and actuation components, each consisting of a PDMS layer and a serpentine copper coil. The key magneto-mechanical coupling (MC) layer, made of magnetoelastic material (mixed PDMS and micromagnets), is designed with a kirigami structure to enhance sensitivity and stretchability. The MC layer converts laryngeal muscle movements into magnetic field variations, which are transformed into electrical signals by the copper coils via electromagnetic induction. This process is self-powered, requiring only additional circuits for signal processing and filtering. The system's design captures omnidirectional laryngeal muscle movements due to the kirigami structure which allows the device to expand and contract in all three axes, capturing the complex three-dimensional muscle fiber movement patterns. The kirigami structure also enhances stretchability (up to 164%), ensuring optimal adherence to the throat for precise movement detection and user comfort. Intrinsic waterproofness, resulting from the magnetic field's insensitivity to water, ensures durability and functionality even with perspiration. The electrical signals are fed to a pre-trained machine learning model using a two-step process involving feature extraction (PCA) and classification (multi-class SVC). The model classifies the signals representing different sentences, which are then used to select corresponding pre-recorded voice signals to be played by the actuation component. The device underwent extensive characterization, including testing for pressure sensitivity, response time, signal-to-noise ratio (SNR), sound pressure level (SPL), and water resistance. Eight participants tested the system, providing data to train and evaluate the machine learning algorithm. The device's performance was tested under various conditions, including different body movements, perspiration, and varied conversation angles.

The developed wearable sensing-actuation system exhibited several key features: a lightweight design (7.2 g), high stretchability (164%), skin-like modulus (7.83 × 10⁶ Pa), high SNR (17.5), a fast response time (40 ms), and water resistance. The device effectively captured three-dimensional laryngeal muscle movements, producing distinguishable signals for different phonation patterns and remaining unaffected by various body movements. The machine learning algorithm achieved a high accuracy of 94.68% in classifying semantical meaning of the muscle movement signals and selecting the corresponding voice signal for output. The device produced an SPL exceeding normal speaking levels across the human hearing range. Notably, the device maintained consistent performance under sweaty conditions, demonstrating its suitability for real-world applications. The acoustic performance tests indicated an SPL above the normal speaking threshold (40-60 dB) across the human hearing range, even at a distance of 1 meter. The device's resonance point shifted towards higher frequencies with increased strain, showing adaptability for diverse usage scenarios. The water resistance tests confirmed consistent performance even after prolonged water immersion. The machine learning model's accuracy exceeded 93% across all participants, with an overall average accuracy of 94.68%. The device showed stable SPL and temperature over 40 minutes of continuous use.

This study successfully demonstrated a novel wearable sensing-actuation system for assisted speaking without vocal folds. The combination of soft magnetoelastic materials, a kirigami-structured design, and a sophisticated machine learning algorithm enabled accurate detection of laryngeal muscle movements and their translation into intelligible speech. The device's key advantages – lightweight design, high stretchability, skin-like modulus, water resistance, high SNR, fast response time, and high machine-learning accuracy – address critical limitations of existing technologies. The results showcase a significant advancement in assistive technologies for individuals with voice disorders, providing a comfortable and effective means of communication during their recovery. The high accuracy of the machine learning model (94.68%) and the device's robust performance under real-world conditions support its potential for clinical translation.

This research presents a novel wearable system for assisted speaking that effectively bypasses the need for functional vocal folds. The device’s unique combination of advanced materials, a sophisticated kirigami design, and machine learning provides a viable solution for patients with voice disorders. Future work could focus on miniaturization, integration with other assistive technologies, and broader clinical trials to validate its efficacy and safety in a larger patient population.

While the study demonstrated high accuracy and robustness in a controlled setting, further research is needed to evaluate long-term reliability and durability in real-world conditions. The sample size of eight participants, while providing positive initial results, is relatively small and needs expansion for broader generalizability. The current system requires pre-recorded voice signals; future iterations could explore real-time voice synthesis for more natural communication. The study focused on specific sentence recognition. Future research could explore the system’s ability to recognize and translate a wider range of spoken language.