Google has cut down its speech recognition error rate by over 30% in the last five years

Google, much like most other players in the industry, has made AI one of its absolute priorities. One of the areas where it has been applied is speech recognition, where Google has made significant progress over the past few years…

Since 2012, the neural nets Google adopted to help improve its speech recognition system have made its word error rate drop by more than 30 percent. This claim comes from Google’s Jeff Dean, first tweeted out by Mashable’s Karissa Bell, who attended the firm’s AI Frontiers Conference in Santa Clara on Thursday (also confirmed to VentureBeat by a Google spokesperson).

The “addition of neural nets” helped profusely, of course, although the percentage drop has not been accompanied by a precise word error rate — the last time the company publicly gave a figure was in 2015, when CEO Sundar Pichai claimed that it was 8%.

Back in September, Microsoft announced that its researcher had scored a 6.3 percent mark, while a month earlier Apple told Backchannel’s Steven Levy that Siri’s “error rate has been cut by a factor of two in all the languages, more than a factor of two in many cases”.