If you feel like Google’s voice recognition is working a little better than it used to, that’s because it is! Today Google reported that their widely-used voice search capability is now being handled by a new engine that recognizes and anticipates words with a much higher degree of accuracy.
Google says that these advancements are thanks to the development of more effective neural network acoustic models through the use of “Connectionist Temporal Classification (CTC) and sequence discriminative training techniques.”
On the Google Research Blog, members of the Google Speech Team including Haşim Sak, Andrew Senior, Kanishka Rao, Françoise Beaufays and Johan Schalkwyk wrote that “These models are a special extension of recurrent neural networks (RNNs) that are more accurate, especially in noisy environments, and they are blazingly fast!”
In short, Google is now even better at correctly recognizing your speech, especially in situations with a lot of background noise, and the delay between speech and interpretation has been made even shorter.
This is the second time in 2015 that Google has announced improvements in their voice search capabilities. The recurrent neural networks that Google is using for their voice models are widely used in the approach to artificial intelligence called “deep learning.”
Andrew Ng, who worked on the Google Brain, predicted in 2014 that “50% of queries will be on speech or images” within five years. With speech recognition becoming an increasingly popular choice of human-computer interaction, it’s no surprise that Google is pulling out all the stops when it comes to developing better voice search functionality.
This new update is already in effect for all Android and iOS devices, so if you haven’t said “OK Google” in a while, now’s a good time experience how effective the company’s voice recognition technology has become.