If you’re a long-time reader of Health Tech Insider, you know that one of my favorite topics is the user interface for wearable Health Tech devices. While there’s no one solution, some approaches have clear limitations. For example, you can’t fit a full keyboard on your wrist (at least, not one that is practical to use). Most wearable displays are too small to be used as a touchscreen in anything other than a limited way. Non-contact gesture detection may be part of a solution. But as Apple’s Siri and Amazon’s Echo have demonstrated, spoken commands may be a convenient and powerful component of a wearable user interface.
One problem is that speech recognition systems are power hungry, and wearable devices need to manage power carefully. A smartphone with a speech recognition function can consume about 1 watt of power, which puts a serious drain on batteries. Researchers at MIT have created a clever chip to address this problem. The chip draws only 10% as much power, or less. Part of the chip’s secret is that it goes to sleep, with only a small portion of the circuits remaining active to detect whether someone is speaking. If it hears speech, it wakes up the rest of the system. This detection system is fairly complex. The researchers found out that simpler systems generated too many false positives and wasted more power.
Another power drain is accessing data that is stored in other system components. As a result, the researchers worked to maximize the processing the chip could perform while accessing local storage as little as possible. The result is a chip that requires much less power. This in turn could support smaller form factors, such as putting speech recognition into earbuds or the frames of smart glasses. The technology could help make it easier and more natural to interact with wearable devices.