Talking without vocal cords? A new AI-supported technology may help

Voice disorders occur across all age groups and demographics. Research has shown that almost 30% of individuals will experience a voice disorder at least once in their lifetime. The most common is temporary functional dysphonia, with the extreme form being voice loss (aphonia), which can occur, for example, after surgeries for laryngeal cancer. In therapeutic approaches such as surgical interventions and voice therapy, voice recovery can extend from 3 months to up to 1 year, with some invasive techniques also requiring a significant period of mandatory postoperative voice rest.

Nearly 95% accuracy in converting signals to audible speech

The activation system, which is based on soft magnetoelasticity, enables assisted speaking without relying on vocal cord vibration. It effectively captures the external movements of the laryngeal muscles and converts them into highly reliable and analyzable electrical signals. These can then be converted back into speech signals using machine learning algorithms with very high accuracy. The device is trained using machine learning to recognize which muscle movements correspond to which words. It weighs about 7 grams and is only 1.5 mm thick. Using double-sided biocompatible tape, it can easily adhere to the throat near the vocal cords and be reused by reapplying the tape as needed.

The first non-invasive procedure

Existing solutions for patients with voice disorders/loss, such as handheld electrolarynx devices or procedures using tracheoesophageal puncture, are still considered invasive and do not entirely ensure adequate communication with surroundings, thus contributing to a reduced quality of life for patients. The new plaster-like device is another example of the successful integration of artificial intelligence into clinical practice. In the future, the research team at UCLA plans to continue developing this technology, particularly by expanding its vocabulary using machine learning, and testing it in people with speech disorders.

(jas)

Source: Che, Z., Wan, X., Xu, J. et al. Speaking without vocal folds using a machine-learning-assisted wearable sensing-actuation system. Nat Commun 2024 Mar 12; 15 (1): 1873, doi: 10.1038/s41467-024-45915-7.