Source: singularityhub.com
Author: Aaron Saenz

You may have a family member or friend who had throat cancer. Maybe you’ve seen a TV show where a heavy smoker needs an artificial larynx pressed to their neck to speak. Either way, you know the voice that they have to use: robotic, monotone, raspy. It works, but it can leave users unable to express themselves well. Researchers at the University of Witwatersrand in South Africa have developed a new system that measures mouth movement to produce a more “human-like” voice. Their work was on display at the recent International Conference on Biomedical and Pharmaceutical Engineering. By placing a device called a palatometer under the tongue, users can try to speak as normal and have their words synthesized on a speaker. The South African artificial larynx can provide inflection, ending the dreaded monotone and providing the means to indicate you are asking a question. With proper calibration, researchers claim greater than 94% accuracy. That’s good news to those who want to regain a normal speaking voice.

The palatometer, which measures tongue/mouth movements with 118+ pressure sensors, is an older device developed at BYU and produced by Complete Speech. It is most often used by speech therapists in instructing their patients and retails for around $200-$300. You can see a brief presentation of the palatometer after the break. University of Witwatersrand’s innovation comes in developing a selective way of using the mouth movement data to generate toned speech. After cataloging tongue motions, and using predictive-analysis, the team has taught their system to recognize around 50 words with high accuracy. About 18% of the time, however, the new artificial larynx has to skip words it can’t recognize. It also has a 0.3 second delay, leaving users appearing something like a poorly dubbed Godzilla movie. Still, because the palatometer is not an implant (users simply place it in the mouth), it can be easily upgraded as needed. Future versions are likely to improve the accuracy, the vocabulary, and the speed.