Google has achieved human parity in its speech recognition technology, according to the latest entry in Kleiner Perkins’ annual Internet Trends report, delivered by the firm’s Mary Meeker at last week’s Code Conference in California.
Defining an accuracy rate of 95 percent as the ‘Threshold for Human Accuracy’, the report indicates that Google has now met and slightly exceeded that level, citing data from Google. The achievement reflects a 20 percent improvement in accuracy since 2013.
The report also suggests that voice is an increasingly popular modality for online searches, accounting for 20 percent of all mobile queries last year.
Clearly, there is a growing market opportunity here for any tech companies that are able to replicate human-like conversation capabilities in AI, especially as a growing number of connected devices emerge with the IoT. And Google is not alone in pursuing it: IBM announced earlier this year that it had reached a word error rate of 5.5 percent with its speech recognition technology, besting the 6.9 percent error rate boasted by Microsoft last autumn. Amazon, meanwhile, continues to train its voice-based Alexa platform on an increasingly wide data set. All of which suggests that consumers will soon be having a lot of in-depth conversations with their devices.