Cornell University researchers have developed a wearable infrared smart cam that can detect voice commands, even when the speaker isn’t making any sound. The camera measures the neck and face starting under the chin.
Cheng Zhang, Cornell University assistant professor of information science, and Ruidong Zhang, Cornell University doctoral student, have created the wearable camera and given it the name “SpeeChin”. It can recognize 54 English and 44 Chinese silent voice commands.
Zhang says voice control is a good option if your hands are busy or you don’t want your smart devices to interact with you. Voice control is not socially acceptable or efficient if you’re in noisy places or in meetings. Silent speech is the solution.
SpeeChin uses an infrared camera mounted on the neck that captures the movement of the chin. This allows it to detect words even if there is no sound. It is much more discreet than forward-mounted cameras and virtually eliminates privacy concerns.
Gizmodo reported that SpeeChin was tested with 20 students. 10 of them spoke simple phrases, including numbers and common voice assistant commands. 10 also spoke simple words and phrases in Mandarin Chinese. After the logic and camera were trained, it was capable of recognizing commands in English with 90.5% accuracy, and Chinese with 91.6% accuracy.
However, these high marks could not be achieved if participants were seated stationary. Variations in gait and head movement caused the recognizability of the recognition to drop once they were asked to move. This severely reduces the range of places the SpeeChin device can be used.
This SpeeChin iteration may have a problem with a stationary wearer, but theoretically it is possible to solve this issue by either longer training with the logic, or an improved infrared cam. This early version of the SpeeChin is clearly not mature enough to see any changes in subsequent iterations.