Siri speech recognition capabilities will be greatly enhanced

Foreign media recently published an article saying that the neural network algorithm has become the mainstream technology, can significantly improve the accuracy of speech recognition, technology giants to make products and services more intelligent, such as Google's Android voice recognition and Microsoft SkypeTranslate. Only Apple Siri has not embraced the technology, but the company has acquired Nuance voice technology license, but also recruit a lot of talent in the field, indicating that the neural network boost Siri version is coming.

The following is the main content of the article:

Neural Networks

It all started with a small academic gathering at Whistler, British Columbia, Canada.

The topic at the time was speech recognition, along with untested new machine intelligence called deep learning - that helped computers recognize spoken language more accurately. Sponsored by Microsoft and held before Christmas 2009, two of the researchers invited a world-renowned depth learning expert, Geoff Hinton from the University of Toronto, to make a speech.

Hinton's philosophy is that machine learning patterns can largely resemble human brain functioning. He wanted to create a "neural network" to gradually form the understanding of spoken language, such networks are also more and more appear. Neural networks were prevalent in the 1980s, but they did not reach their full potential until 2009.

According to Peter Lee, director of research at Microsoft, voice researchers at Whistler are friendly to the concept, but "not very interested." Those researchers had already developed their own algorithms at that time. However, Deep Learning was worth a look at the Microsoft team, so the company got some engineers to work with Hinton's research team and run some experiments with real data.

Peter Lee recalls that the test results were "astonishing" - a 25% improvement in speech recognition accuracy. You know, it is a 5% improvement is called a transformable area. "We announced the results of those experiments, and the world changed immediately," he said.

Now, almost five years later, neural network algorithms have become mainstream technologies that make computers smarter in new and exciting ways. Google uses them to enhance Android voice recognition. IBM also uses them. Most notably, Microsoft applied neural networks to Star Trek-style Skype Translate translations, enabling the service to translate your words into another language instantly.

Hinton pointed out that "people did not believe this kind of technology from the very beginning, but now our technology has already become the mainstream."

Siri will be upgraded

However, a well-known company has not turned to neural network algorithm: Apple, its Siri software will usher in an upgrade. Although Apple has always been tight-lipped about its internal operations, the company appears to have licensed voice recognition technology from Nuance, perhaps the most famous voice recognition technology provider. The AI ​​research community believes that Siri is about to change.

They pointed out that it is clear that Apple has set up its own voice recognition team, the Siri version of neural network boost impending.

Peter Lee knows this very well. Apple last year recruited Alex Acero, one of his senior managers. Arcelor now serves as senior director of Apple's Siri department, where he studied voice technology for nearly 20 years at Microsoft. At Microsoft, his subordinates include Li Deng and Dong Yu, two researchers who invited Sinton to attend the academic gathering.

Apple also dug up some voice researchers from Nuance, including Siri manager Gunnar Evermann. Arnab Ghosha, a researcher at the University of Edinburgh, is also among the voice researchers led by Apple.

"Apple not only recruited manager-level talent but also recruited team leaders and research-class titans," said Abdel-rahman Mohamed, a postdoctoral fellow at the University of Toronto who was pursued by Apple. They are building a very strong voice recognition research team. "

Ron Brachman, Yahoo's research director who helped launch Siri's predecessor, notes that Apple's iPhone digital assistant goes far beyond relying on voice recognition technology. Microsoft's Peter Lee believes that Apple takes six months to catch up with Microsoft and Google, and start using neural networks will give Siri a great improvement.

"All the big players, except Apple Siri, are fully technology-based," he said. "I think Siri's move to neural networks is just a matter of time."

http://www.wired.com/2014/06/siri_ai/

Translator: Le Bang