Researchers at Microsoft claim to have created a new speech recognition technology that transcribes conversational speech as well as a human does (via The Verge).
The system's word error rate is reportedly 5.9 percent, which is about equal to professional transcribers asked to work on the same recordings, according to Microsoft.
Microsoft researchers from the Speech & Dialog research group (Image: Allison Linn)
"We've reached human parity," said chief speech scientist Xuedong Huang in a statement, calling the milestone "an historic achievement".
To reach the milestone, the team used Microsoft’s Computational Network Toolkit, a homegrown system for deep learning that the research team has made available on GitHub via an open source license. The system uses neural network technology that groups similar words together, which allows the models to generalize efficiently from word to word.
The neural networks draw on large amounts of data called training sets to teach the transcribing computers to recognize syntactical patterns in the sounds. Microsoft plans to use the technology in Cortana, its personal voice assistant in Windows and Xbox One, as well as in speech-to-text transcription software.
But the technology still has a long way to go before it can claim to master meaning (semantics) and contextual awareness - key characteristics of everyday language use that need to be grasped for Siri-like personal assistants to process requests and act upon them in a helpful way.
"We are moving away from a world where people must understand computers to a world in which computers must understand us," said Harry Shum, who heads the Microsoft AI Research group. However it will be a long time before computers can understand the real meaning of what's being said, he cautioned. "True artificial intelligence is still on the distant horizon."
The system's word error rate is reportedly 5.9 percent, which is about equal to professional transcribers asked to work on the same recordings, according to Microsoft.
"We've reached human parity," said chief speech scientist Xuedong Huang in a statement, calling the milestone "an historic achievement".
To reach the milestone, the team used Microsoft’s Computational Network Toolkit, a homegrown system for deep learning that the research team has made available on GitHub via an open source license. The system uses neural network technology that groups similar words together, which allows the models to generalize efficiently from word to word.
The neural networks draw on large amounts of data called training sets to teach the transcribing computers to recognize syntactical patterns in the sounds. Microsoft plans to use the technology in Cortana, its personal voice assistant in Windows and Xbox One, as well as in speech-to-text transcription software.
But the technology still has a long way to go before it can claim to master meaning (semantics) and contextual awareness - key characteristics of everyday language use that need to be grasped for Siri-like personal assistants to process requests and act upon them in a helpful way.
"We are moving away from a world where people must understand computers to a world in which computers must understand us," said Harry Shum, who heads the Microsoft AI Research group. However it will be a long time before computers can understand the real meaning of what's being said, he cautioned. "True artificial intelligence is still on the distant horizon."
51 comments
Apple's VP of Health Dr. Sumbul Desai recently sat down for an interview with MobiHealthNews where she discussed Apple's health products, the company's relationship with the FDA, the...
Apple maintains a photography tutorial section on its website for those interested in stepping up their iPhone photography, and it's regularly updated with new videos. Apple also uploads its...
It's no secret that Facebook is harvesting incredible amounts of data on all of its users (and some that don't even use the service), but what may come as a surprise is just how detailed and...
Apple plans to close both of its retail stores within the Eastern District of Texas in a few months from now in an effort to protect itself from patent trolls, according to five sources familiar with...
Apple chipmaker TSMC will miniaturize its fabrication process to 5 nanometers in a bid to secure orders for Apple's processors in its 2020 iPhones, according to industry sources cited in a...
Adobe has released an update for Adobe Premiere Pro via the Creative Cloud app to resolve an audio issue that left some users with blown-out MacBook Pro speakers.
Earlier this month, we...
Apple CEO Tim Cook is set to deliver Stanford's 2019 commencement speech on Sunday, June 16, the university announced today.
Stanford chose Cook because he has been a "prominent voice on...
You may or may not know this, but there's a feature in Safari that's designed to let you select a portion of text and send it to a friend over Messages using the Share feature in Safari.
...
