Microsoft has slowly but surely been increasing its efforts to compete with Apple and Google in the voice recognition game. First came Windows Phone 7's speech commands, then Kinect's own voice integration, and most recently Windows Phone 8's speech API for developers to tinker with. But what's coming next? We may now have the answer thanks to a leaked demonstration from MSFTKitchen, which was filmed at Microsoft's recent TechFest event.

Touting "deep neural network technology" applied to the same backend every Windows Phone accesses today, Microsoft Research has enabled some vast improvements to speech recognition. Not only are basic voice queries more than half a second faster, the recognition is also more accurate in the presence of background noise. As you'll see in the video below when the presenter tried to search the phrase "Starbucks in Washington" with background noise, the current software thinks he said 'sky in washing machine' whereas the new software picked it up flawlessly.

However, it appears the big new benefit of this new technology is the ability to stream voice searches. In today's implementation of voice recognition, you speak to the phone and then it takes a few seconds before the search results load. With this new version of the speech recognition though, results appear as you speak which is frankly kind of incredible.



Microsoft is definitely on the right track with its current speech capabilities with Windows Phone 8, opening it up to developers to write custom commands for their own apps. However, this technology could push that feature far beyond what we experience today and may significantly increase the utility of voice search as well. Rumor has it the Blue updates for both Windows and Windows Phone will incorporate improvements to the Bing search experience, and we have a sneaking suspicion this tech may well play a part in that.

Via: PocketNow | Source: MSFTKitchen