Google Claims Great Leaps Forward in Voice Search
Last year Google added a machine-learning technique called deep neural networks to its voice search. The speech recognition error rate decreased by 20 percent.
Since then, Google’s voice team has kept tweaking the service, resulting in another 25 percent increase in speech recognition accuracy. That’s according to Google VP of Search and Assist Johanna Wright, who spoke at the announcement of Google’s new flagship phone and operating system in San Francisco today.
The difference, Wright elaborated offstage, is that over the past two years, Google has applied the same sort of rigor to voice search as it has long dedicated to search quality.
It’s not one big change, but a series of refinements, similar to the way Google modifies its search engine algorithms on a regular basis, Wright said.
Google Voice Search is integrated across the Android operating system — even more so on the new Nexus 5 — and available as an app for iOS and Windows 8. It’s also provided on the Web.
Sure, these are just a couple of braggy growth stats with no absolute numbers. But since voice search is becoming such a key way to navigate mobile phones, the improvements do add up.