Microsoft's AI Now Hears You as Well as Humans

Senior Writer
Updated

Microsoft says it has achieved a breakthrough in speech recognition: It's developed a system that can understand normal speech just as well as people who transcribe conversations for a living.

Microsoft's system understands acoustics and details like a speaker's pitch and how fast or slow they speak, according to a company research paper. And its word error rate, how often it transcribes a word incorrectly, has dropped to 5.9 percent, on par with the best of us (humans).

Credit: MicrosoftCredit: Microsoft
Microsoft's system can still be tripped up with an "um," "uh" or other stammering sound that come up in normal conversations. But as it stands now, this AI advance could already help make Microsoft's Cortana voice assistant more effective, and "truly intelligent," says Harry Shum, who heads up Microsoft's Artificial Intelligence and Research group.

MORE: Here Are the 10 Best Smartphones Available

Beyond improving its digital assistant, Microsoft's speech recognition system could be put to better use for fun, on the Xbox, and for work, with a better voice-to-text transcription being one of many possible uses.

Microsoft's record-low word error rate is a result of more than 20 years of research into speech recognition, according to Geoffrey Zweig, manager of Microsoft's Speech & Dialog research group.

But the company says it still has work to do to make sure its speech recognition system works in real life, like when you're walking down the street past car horns honking and construction noise, or when you're at a party with many voices in the background.

Of course, it takes time to have implement new technologies into the products consumers buy, so Microsoft's announcement really serves as a public reminder that the company is working on speech recognition at the same time its major competitors are.

Earlier this month, Google unveiled its much-anticipated Pixel phones equipped with the company's natural language processing system in its Google Assistant feature.

And Apple is expected to come out with a smarter version of its Siri voice assistant, now that it's hired Carnegie Mellon University professor Russ Salakhutdinov as its director of AI, making it a stronger competitor against Amazon's Alexa in the home.