How Google & Other Search Engines Zone in On Your Voice
Error rates for Google’s voice search have reportedly decreased from 25% to 8% within the last few years
USA Today’s Edward C. Baig interviewed Travis Trekell from Google to find out exactly how Google’s voice search works. Joining Travis inside the rarely seen anechoic chamber within Google’s headquarters in Mountain View, Ed had the chance to record sample phrases such as “What is the temperature in Tokyo?” in order to test the effectiveness of Google’s voice search.
Google’s Algorithm is Designed to Remove External Noise
Google’s anechoic chamber has foam coated walls, a bouncy trampoline like floor, and is suspended off of the ground. The whole room is designed around minimising external noise and isolating the sound of your voice.
After your voice has been recorded, Google can then overlay background noises – sounds of cars, people talking, or music, which the algorithm can then seek to filter out. Having the lab available helps Google to measure the effectiveness of devices and apps that depend on their ability to comprehend a person’s voice.
The Wider Applications
As the applications for voice search, and the competition to create the best voice recognition software between tech companies like Apple, Windows, and Google is growing, there is an ever increasing motivate to create a search app that can reliably and accurately understand a person’s voice.
Linne Ha (known as the “voice hunter” at Google) has been working towards developing a universal model of each of the world’s most commonly spoken languages, so that Google’s can better understand a wider variety of languages.
The end goal will be an app that can carry out a full conversation with a person; a conversation that sounds and feels completely natural.