Micrsoft improves Windows Phone voice recognition using Deep Neural Networks data
Microsoft has rolled out improvements to Windows Phone voice recognition, using the work of Geoffrey Hinton, the founder of Google-acquired DNNresearch.
According to Microsoft, the company has been silently rolling out this functionality over the past few weeks. Voice recognition has gained an increase of 15% accuracy, and now registers at twice the speed. In a video demo, Microsoft showed improved results for voice texting and search, and better accuracy in noisy environments.
To achieve these improvements in functionality, Microsoft applied data from Hinton’s work with Deep Neural Networks (DNN), which has helped to build better models of how the human brain processes information.
This proved useful to Google, who acquired DNNresearch, and used this data to power the Google+ photo search functionality. DNN data provided, among other things, a better data model for object and facial recognition. Similarly, it is now aiding Microsoft to build a better acoustic model for voice recognition.
In order to implement this research into Windows Phone products, the Bing development team has been working in close tandem with Microsoft Research, a series of labs which perform a variety of product research for Microsoft.
The new, improved voice recognition is currently only available in the United States.