Microsoft apparently had a confidential event for its employees earlier this month, wherein it demonstrated a bunch of future technologies that it has been working on. And what we have before us is more than just a sneak preview of a prototype Bing app for Windows Phone by way of a live demo. Microsoft demonstrated a bunch of improvements on the speech recognition feature. On MSFT Kitchen, these improvements are listed as – word error rate reduction, reduction in latency for speech requests and better handling of speech requests in noisy situations.
The live demo had Microsoft test the voice recognition feature side-by-side on the existing system and on the new system. The first test had the demonstrator speak a few words into the system, and the new system returned with results faster than the existing system. To be precise, while the existing system took 1.01 seconds to return with results, the new system did it in .56 seconds.
In the next step, the demonstrator turned on the streaming mode on the new system (to the right). With the streaming mode enabled, the demonstrator managed to have his spoken words appear on the screen, even as he was still talking. While this was happening on the new system, the existing system continued to show text once the demonstrator paused.
Now since the words appeared on the screen instantaneously on the new system, the search results were thrown up that quicker (.43 seconds), whereas the same activity took 1.83 seconds on the existing system.
He then moved on to test the accuracy of this new system and chose to do it in a situation where accuracy suffers the most – in noisy places. To introduce some noise into the whole setup, the demonstrator turned on some background noise and said that if all goes well, the existing system would face troubles in understanding what the speaker is saying, while the new system would come out unscathed. As expected, while the new system accurately typed out:: "Starbucks in Washington", the existing system, in all inaccuracy typed out "Sky in washing machine".
At the end of the demo, Eric Rudder, Chief Technical Strategy Officer, said that in addition to a 12-15 percent increase in word error rate accuracy in current tests, another 10-15 percent is “in the bank,” as per his conversation with Microsoft Research speech scientists.
While the demonstrator took note that certain enhancements are already being rolled out to Windows Phone customers in select areas, what’s unclear is the timeline they’re following for everything else shown.
Just this morning, we got more evidence that Windows Blue could be a better and sharper touch-based operating system, as an Internal video has leaked on YouTube. Besides touch improvement, a host of other new tweaks, including better speech recognition, could be part of the new Windows OS codenamed Blue.
Published Date: Mar 22, 2013 13:53 PM | Updated Date: Mar 22, 2013 13:53 PM