Ever since Apple launched Siri with the iPhone 4S, the average search results on an iPhone ranged from the offensive to the practical. Whatever way you look at it though, Siri did indeed change how we interact with our phones and their search capabilities.
Microsoft followed suit and implemented their own voice recognition functionality in the Windows Phone platform. And while not perfect, the Redmond behemoth has been working hard to improve the service allowing for better reduction of background noise and dialect recognition.
In a new video post, Stefan Weitz and Michael Tjalve – both from the Bing Speech Team – have demoed the latest improvements to the voice search capabilities in Windows Phone 8.
Working with Microsoft Research, the Bing Search Team have come up with a new methodology; Deep Neural Networks, or DNN for short, to provide greater accuracy and speed when using voice search functionality. It’s a new version to the existing ‘Acoustic Model’ already in place in speech recognition. According to Tjalve:
It’s essentially a computational framework or how we process speech within the brain.
All this means that the DNNs are able to learn more quickly and allow Bing voice capabilities to get even close to the way we, as people, recognise speech. This, and improvements to recognising speech patterns, along with being able to block out background noise and ambient sounds have helped improve word error rates by 15 percent.
If you watch the video, at the 1:13 mark, you’ll see an example of how the latency has been reduced in providing quicker results to Bing voice searches. So that’s twice as fast, and more accurate than ever before!
It’s an exciting time, and improvements like this are helping drive the way we interact with our phones and the search services that we use on a daily basis.
Source: Bing Blogs
21 Comments - Add comment