Apple gets Siri-ous about voice recognition
iOS 8 could see debut of deep neural nets to rival Microsoft and Google
Siri is about to get a high-tech upgrade in iOS 8 to improve its voice recognition abilities, according to a report in Wired.
Apple looks set to ditch Nuance’s technology for an in-house solution and has poached one of Microsoft’s top scientists to head its team.
The move comes just weeks after Microsoft upped the robo-speech ante, demonstrating live two-way translation between German and English within its Skype videophone service and promising a consumer version by the end of the year.
The shift in-house should also see Apple adopt the latest deep neural network algorithms, which Microsoft has found can boost accuracy by 20% or more.
Deep thought
All modern speech recognition systems use neural networks – software composed of virtual neurons that can be trained to match an input to an output by varying the links between them, a bit like the way real brains work.
The latest iteration of this mimicry reproduces the layers of neurons found in human brains to create a deep neural network. Microsoft’s Cortana speech recognition system embedded in Windows Phones has multiple layers that enables it to model a wider range of speech, including accents. Now that technology might be coming to Apple’s Siri virtual assistant, too.
The risk for Apple is that it could experience another fiasco on the scale of 2012’s epic Apple Maps fail, where ditching Google Maps for inferior home-brew navigation infuriated thousands of iPhone users.
If Siri suddenly develops cloth ears this autumn, deep neural nets could simply be another deep PR hole for Tim Cook to dig himself out of.
READ MORE: Apple iOS 8 preview
[Wired]
Mark Harris
Freelance tech writer 
