In order for voice assistant softwares to be good and useful, there are many things it needs to do, although we guess the most obvious is that it has to be able to listen to the user and understand what they are saying. If the user’s commands aren’t being picked up on, then there’s really not much point.
It turns out that Google’s speech recognition efforts are chugging along just fine to the point where it is apparently at the same level as humans. This is according to Mary Meeker’s annual Internet Trends Report (via 9to5Google) where apparently Google’s machine-learning voice recognition system has achieved a 95% word accuracy rating for the English language.
This is said to be the same threshold for human accuracy, which means that as far as recognizing words from users in English, Google’s speech recognition is pretty spot on. The report notes that Google has improved upon its efforts by nearly 20% since 2013, and all of this was previously touted by Google’s CEO Sundar Pichai during Google I/O 2017 where he noted that computers these days are getting better at understanding voice input.
As 9to5Google points out, Google’s success at developing the technology has allowed them to create Google Home using just two microphones to pick up speech versus eight, and how it can support multiple users and recognize up to 6 of them at any given time.
Filed in . Read more about Google.