In October last year, Microsoft revealed voice recognition technology which could transcribe the words of a phone call just as well as a trained human transcriber. Professionals hit a 5.9% word error rate, matching the results of Microsoft’s system.
In a blog post published on Sunday, Microsoft Research’s Chief Speech Scientist, Xuedong Huang has announced the company has now surpassed that record.
Microsoft’s speech recognition system has “reached a new industry milestone” and notched a 5.1% error rate, beating the levels of professional human transcribers.
The result is a major AI milestone, bringing Microsoft one step closer to developing artificial intelligence which will understand the meaning behind words, instead of simply recognising them.
Huang states, “Moving from recognising to understanding speech is the next major frontier for speech technology”.
The achievement arrived much sooner than Microsoft expected, as in 2015 Huang informed media that developing a system which could surpass a human in transcription was “four to five years away”.
The result only took two years.
Microsoft states there will be challenges ahead, especially as it focuses on improving speech recognition technology to adapt to noisy environments or highly accented speech.