Siri is all set to sound a lot more natural in iOS 11, ditching its robotic voice for a more human-sounding one. Apple has decided to share how exactly it managed to do this in a new whitepaper which offers some very interesting insights into the work that goes on behind-the-scenes.
Anyone who’s used Siri in the years since its launch would have noticed that the AI doesn’t quite sound like the warm digital assistant that Apple intends it to be. The voice has always sounded a bit clipped and artificial, perhaps even putting people off from using it.
Those users might re-consider Siri now that Apple has taken advantage of deep learning to make it sound more natural and smoother. The company even talks about letting the tool’s personality shine through. It apparently interviewed hundreds of candidates before settling on one to be the new female voice behind Siri.
The process behind the revamped voice basically boils down to Apple engineers using a new on-device hybrid unit selection system based on deep learning. The technology combines unit selection and parametric synthesis techniques, with the former providing high-quality recordings and the latter being used to predict which unit should be selected.
The remarkable difference can be heard in a table which compares Siri’s voice on iOS 9, iOS 10, and iOS 11. Anyone interested in hearing Siri’s voice changes for themselves can go here and scroll down to the very last chart. The new-and-improved version of the AI will hit iPhones and iPads in September.