Researchers at Cornell University in the US have developed a new technology that allows silent communication through sonar glasses. The glasses use tiny speakers and microphones to read words the wearer says silently, allowing them to perform a variety of tasks without the need for physical input.
The technology, developed by Cornell University PhD student Ruidong Zhang, is an improvement on a similar project that used a wireless headset, whereas previous models relied on a camera.
The glasses use a silent speech recognition interface called EchoSpeech, which uses sonar to sense mouth movements and a deep learning algorithm to analyze echo signatures in real time, according to igeekphone. This allowed the system to identify the words the wearer was saying silently with about 95 percent accuracy.
One of the most exciting prospects for the technology is that it could be used for people with speech impairments to silently input conversations into a speech synthesizer and then speak words out loud. Glasses could also be used to control music playback in a quiet library or dictate messages at a noisy concert.
The technology is small and low-power, and doesn’t invade privacy because no data leaves the user’s phone. That way, there are no privacy concerns. Glasses are very easy to wear and are more practical and feasible than other silent speech recognition technologies available.
According to the researchers, the system only needs a few minutes of training data to learn the user’s speech patterns. Once that is done, it can send and receive sound waves to the user’s face, sense mouth movements, and analyze the echo signatures using deep learning algorithms. The system is currently able to recognize 31 isolated commands and a string of consecutive numbers, with an error rate of less than 10 percent.
The current version of the system offers about 10 hours of battery life and can communicate wirelessly with a user’s smartphone via Bluetooth. The smartphone is responsible for processing and predicting all the data and transmitting the results to a few “action keys” that allow it to play music, interact with smart devices or activate voice assistants.
Cornell University’s Intelligent Computer Interface Future Interaction (SciFi) Lab is using a Cornell grant program to explore the possibility of commercializing the technology.