Giving voice to a new artificial larynx
Researchers are creating a system that uses a retainer-like mouthpiece designed for use in speech therapy to help those without a voice box speak less robotically.
A new type of artificial larynx could mean better-sounding speech for those who've had their larynx removed due to laryngeal cancer or other ailments.
Researchers at the University of the Witwatersrand in Johannesburg, South Africa, have come up with a system that tracks mouth movements to determine what word is being formed and then uses a speech synthesizer to audibly produce the correct word.
"All of the currently available devices produce such bad sound--it either sounds robotic or has a gruff speaking voice," Megan Russell, a Ph.D. candidate at the university, told Technology Review. "We felt the tech was there for an artificial synthesized voice solution."
Russell and her colleagues created the software for the system, which is being shown off this week at the International Conference on Biomedical and Pharmaceutical Engineering in Singapore. For their research, the team is training a retainer-like mouthpiece already utilized in speech therapy to recognize words mouthed by people without a larynx.
The mouthpiece, called a SmartPalate, is made by Utah-based company Complete Speech. It uses 118 embedded sensors to track tongue-to-palate and lip closure contacts.
A microprocessor input/output device worn around the user's neck or placed on the desktop connects the SmartPalate to a personal computer, and software produces real-time, onscreen visual feedback that shows those with speech impediments how to reposition their tongues.
The system being developed in South Africa, according to Technology Review, would translate mouth movements into words to be reproduced on a small sound synthesizer that could be kept in a pocket.
Russell has trained her software to recognize 50 common English words by saying each one multiple times with the SmartPalate in her mouth. The information picked up by the sensors can be represented on a graph and put into a database, and each time the wearer configures his or her mouth to form a word, the contact patterns are compared against the data to identify the right word.
Russell says the system identifies correct words 94.14 percent of the time, although this doesn't include words that the system classifies as "unknown" and chooses to skip.