Google updates their voice search software, making speaking with the search engine far faster and more accurate than before. The Google Speech Team have announced that they’re improving their neural network acoustic models using two new sets of features. One part of this equation is Connectionist Temporal Classification (CTC), while the rest consists of sequence discriminative training techniques. Go on a try a Google voice search with “OK Google” now – you’ll notice the quickness. And the accuracy. And the ability to be heard in slightly more noisy environments.
This new technology developed by the Google Speech Team uses the entire sentence you speak – not just individual word fragments – to identify what you’re saying. Using Recurrent Neural Networks (RNN) technology, Google is able to hear the sounds your voice makes in context.
“RNNs have feedback loops in their topology, allowing them to model temporal dependencies,” says Google Speech Team‘s Haşim Sak et. all, “when the user speaks /u/ in the previous example, their articulatory apparatus is coming from a /j/ sound and from an /m/ sound before. Try saying it out loud – “museum” – it flows very naturally in one breath, and RNNs can capture that.”
You’ll be able to see this voice recognition and search system available right this minute in multiple places. The new acoustic models are used in voice searches and commands in the Google app (the place where you search for things) on iOS and Android devices of all sorts.
This new model is also in place for dictation on Android devices. Tap in and let us know how it goes!