While the idea scares privacy advocates, developers of smart speakers are trying to get their devices up without hearing user wake-up words. We know that Google is already working on the technology, but now that people at Carnegie Mellon University are following suit, researchers have developed a machine learning model that uses the power of artificial intelligence to determine exactly where someone’s voice comes from.
At first glance, this may not seem very important, but researchers are planning ahead for a future in which IoT devices will become “more and more dense” throughout the home and office environment. Imagine you have an Alexa-driven smart TV, speakers and smartphones all in one place. Speaking a wake-up word is likely to activate command mode for each of these devices, which is almost unnecessary.
In this case, you might want to speak to a particular smart device, not to the device in the room, which is what this study does. Unlike other methods, this instructionless wake-up method does not require facial recognition technology. Of course, the accuracy of the system will depend heavily on the ability of ordinary users to adapt to the concept. For example, when asking questions of family or friends, users need to make sure they don’t accidentally speak loudly in the direction of their tracker. For now, however, the technology won’t even be available to the public any time soon.