A research paper reveals just how Google Inc(NASDAQ:GOOG) is capable of fiving users accurate and speedy query returns with its new voice search technology. It boils down to massive amounts of data.
How voice search technology listens to provides with information is a complex endeavor, however, Google has put effort into explaining what mechanisms make its voice search app work in a new research paper that it had posted earlier today.
Basically, it comes up to a lot of data.
As per Google, more data improves all Web services. This may seem evident, but for better speech detection, it does not only mean a lot of data, it also means how that data is structured. Google’s voice search technology mainly uses data from anonymized queries on Google’s site to get the data it requires.
Ciprian Chelba, a Google research scientist mentioned in a blog post about the research that the language model is the module of a speech reorganizer that assigns odds to the next word in a sentence given the previous ones. For example, if the previous words are ‘New York’, the model would allocate a higher probability to ‘pizza’ over ‘granola’.
While executing its voice search evaluations, Google scientists incorporated up to 230 billion words from a random sample of anonymized queries from Google’s site that did not prompt spelling correction.
Chelba has concluded in the blog post that with such an enormous data set, word error rate could be reduced by 6 to 10 %, and for systems with an even wider variety of operating points, word error diminution can be between 17 to 52%.
Google’s latest voice search technology app looks to be directly competing with Apple’s Siri. The new app was launched on Tuesday. The software went out as a part of an update to Google’s search app for iOS. It offers contextual, spoken outcomes for voice queries and serves up web searches for everything else.