Google makes voice search faster and more accurate

Google's Aparna Chennapragada shows how Now on Tap will handle your hardest Skrillex queries. Screenshot: Cult of Android

Google’s Now on Tap isn’t the only upgrade for voice search. Screenshot: Cult of Android

Google voice searches just got faster and more accurate, thanks to new acoustic models that provide better voice recognition, especially in noisy environments.

“New acoustic models are now used for voice searches and commands in the Google app (on Android and iOS), and for dictation on Android devices,” wrote the Google Speech Team on the Google Research Blog today. “In addition to requiring much lower computational resources, the new models are more accurate, robust to noise, and faster to respond to voice search queries — so give it a try, and happy (voice) searching!”

In our quick tests on Google’s Android and iOS apps, performance is noticeably snappier — and at least as accurate as ever. The updated Google app is far faster, and typically returns far more useful results, than Apple’s Siri.

Voice recognition has made astonishing leaps in the past decade, and will only grow in importance as more and more smartphones and connected devices add always-on functionality. As the virtual assistants powered by voice become better at hearing, understanding and even predicting our requests, they will become woven ever more deeply into our day-to-day routines.

While Siri might be the most recognizable AI assistant around, Android handsets and the amazing Amazon Echo have led the way with the type of always-on functionality that is only now coming to Apple handsets with the iPhone 6s.

The details behind Google’s updated acoustic models is pretty deep for a anyone without a computer science degree. The search company has been using deep neural networks since 2012, and used connectionist temporal classification and sequence discriminative training to create the new acoustic models, according to the Google blog post.

You can take the deep dive into the science on Google’s site, or watch a machine learn how to say a sentence in the video below, but the takeaway is that the new acoustic models are more accurate, especially in the type of noisy environments that can render voice recognition mostly useless.

If you haven’t tried Google’s apps for search or dictation, give them a shot now. They’re easy to use and even more incredible than ever. And when Google Now on Tap brings even deeper contextual awareness later this year, you can expect an even greater leap.

https://www.youtube.com/watch?v=5_9Soz3D41g&feature=youtu.be