Microsoft makes history: human-like speech recognition achieved

Microsoft's Speech & Dialog research group

Microsoft has outdone itself. Just one month ago, the company's Speech & Dialog research and development team beat out IBM's record results, posting a 6.3 percent WER (word error rate) in the Switchboard speech recognition test. Those results were pretty close to human performance but still short. That previous best is no longer valid as that same team has made history in speech recognition.

In the same test, Microsoft's speech recognition system managed a 5.9% WER. When the team compared those results with human transcriptionists, they were identical, making the speech recognition software as accurate as human beings when it comes to writing/reciting words spoken to them.

Geoffrey Zweig, who manages the Speech & Dialog research group, was happy to finally make this breakthrough. He said the historic results were “the culmination of over twenty years of effort". Harry Shum, executive vice president of the Microsoft Artificial Intelligence and Research group, couldn't believe the results came as quick as they did.

"Even five years ago, I wouldn’t have thought we could have achieved this. I just wouldn’t have thought it would be possible".

-Harry Shum

Indeed the achievement is one for the record books. Though there's a moment for celebration, the team is not done yet. Much has been accomplished since DARPA started the work to make speech recognition possible back in the 1970s, but there are still errors in speech recognition by a computer that could be capable of perfection (or close to it).

There's also the greater error rate in continuous conversation that will need some work. In a paper published by Cornell, Microsoft's same speech recognition software achieved an 11.9% WER in the CallHome testing portion, where friends and family members carry on open-ended conversations as the system tries to interpret and report what it's hearing. Even though that 11.9% WER is also a new best, Microsoft's software still has plenty of opportunity for growth in that testing environment.

There's plenty of immediate benefits to this technology for Microsoft. Shum says that the new technological breakthrough “will make Cortana more powerful, making a truly intelligent assistant possible". Perhaps we'll see the benefits in one of the next updates to Cortana across the Windows 10 platform.

Source: Microsoft

Report a problem with article
Next Article

Windows 10 build 14951 for PC and Mobile now available in the Fast ring - here's what's new

Previous Article

Apple sends out invitations for October 27 event

26 Comments - Add comment