Technologies that at first seem like magic soon become so commonplace that we no longer even think about how amazing they are – or wonder how they operate. Our digital home assistants – Amazon’s Alexa, Google Home and their emerging rivals – have been with us long enough that we take for granted they can understand our unique voices, commands and questions, even with other noise occurring around them. But we should stop to consider the technology underpinning these modern marvels: signal processing. It’s one of the most important and growing fields in the technology profession, and becoming a signal processing engineer will make you part of a vital network of men and women who make our digital assistants – and future technologies – possible.
One of the reasons we may not think too much about how digital home assistants recognize who we are and what we’re saying is that computers and smartphones have been doing this for a long time now (think Apple’s Siri, for instance). But what most people don’t realize is it’s far more difficult for devices like Alexa – with whom we communicate at a distance with a degree of sound interference – to do this than a computer or phone, with which we are more up-close and personal. They must process not only speech, but also echoes and conflicting noise that deteriorate accuracy, including the TV, kitchen appliances and other people talking. That’s where signal processing comes in.
Alexa and Google Home need to use “Far-Field Speech Recognition,” itself the product of two different methods of signal processing. The first of these types is multichannel speech processing, which uses multiple microphones to allow signals to be enhanced from some directions and suppressed from others. This excludes any interfering noise and focuses on the direction from which your voice is coming (that’s why Alexa flashes a blue light in whichever direction from which she hears your command). The second is multi-condition training, which trains the device to find the command within all the sound waveforms. Google engineers even created a room simulator to generate such signals from various spaces that are small, big, noisy and echo-prone, among other things. By “training” digital assistants in this way, devices themselves are therefore able to work out which sounds they should pay attention to, and answer. Bringing these two methods together are the signal processing engineers, who use machine learning to program home assistants to decipher voice commands.
The use of signal processing and machine learning in digital assistants – though little understood by the average consumer of these products – represents a huge technological leap that would have been impossible just a few years ago. At the very least, the role of signal processing in making these increasingly essential gadgets possible is good food for thought. But if the idea of a radical and constantly developing technology excites you, then the career of signal processing engineer just might be worth exploring. Just as a civil engineer can point to a bridge or tunnel and say “I helped build that,” you may one day be able to give a command to Alexa or Google Home, turn to your friends and say "I helped make that possible."