Skype can be a universal translator: Microsoft Research

Microsoft Research, which delves deep into fields like computing and information technology, is in sharper focus after the software behemoth's acquisition of Nokia's phone business in September last year, and its attempts to gain a firmer foothold in the rapidly growing mobile devices segment, dominated by Google and Apple. There's also the niche area of wearable technology which is on the brink of going mainstream.

Peter Lee, who recently took over as the head of Microsoft Research, shared his thoughts on emerging technologies on a recent to Bangalore.

Was Microsoft slow in moving to the mobile?

For the past 20 years, MSR has been doing research relevant to the mobile. We were not late in porting cutting-edge technologies to consumer devices. MSR tells the product line about the way the world is going. But the world isn't always ready to take the product. Microsoft was one of the first companies to sell tablets with features like speech and handwriting recognition. Arguably we were too early.

 

 What's the progress in speech-recognition technology?

For 11 or 12 years, the quality of speech systems didn't improve from any company, and many of them gave up. But we kept the research going. In 2009, we had a major breakthrough that saw 30% improvement in speech recognition quality. People are using speech a lot in Xbox One. It's magical when I go into my living room, say "Xbox On", everything turns on, and immediately it says "Hello Peter, Welcome Back", because it recognizes my face and my voice. Research is not about speed, but also about the longevity. It's a lot about patience as well.

Can you elaborate on that breakthrough?

This breakthrough used a concept in artificial intelligence called Deep Learning, which in turn is based on neural network. The system learns by associating certain representation. The learning can be so good that if you show the picture of a dog, the system will not only recognize it as dog, but will also tell you the breed. It's like a child learning to recognize something.

What are its applications?

There are many. For example, when you give a search for 'patent leather shoes for teenager' it understands you want to shop. So, it understands not just the meaning, but the intention as well. We provide translations for over 45 languages.

At the height of the end-of-the-world buzz in 2012, there was a rush of tourists to the Mayan state in Mexico. The government there worked with Microsoft to have the Mayan languages translated for the tourists.

A few years ago, chief research officer Rick Rashid demonstrated speech recognition breakthrough via machine translation that converted his spoken English into English text, then to Chinese text and finally to Chinese in his own voice - all live while he was addressing a group of students. We are now looking at Skype and wondering if it could be a universal translator.

What's the status of wearable technology?

We are doing a lot of research in wearable technology. The fact is most of the sensors in the devices are of poor quality. There is a lot of noise that the sensors don't filter out. We think that Deep Learning can get past the noise. One of our focus areas is the GPS sensor. There is an algorithm in GPS that is extremely power hungry. By leveraging big cloud infrastructure, we will be able to deliver an energy-efficient GPS. We are working with wildlife authorities in Africa, and testing them: for example, to track the movement of zebras.