Researchers at Microsoft have announced that their new speech recognition technology can transcribe the content of a conversation with “the same or fewer errors” than actual human professionals trained in transcription, even when the human transcript is double-checked by a second human for accuracy.
At 5.9% word error rate, Microsoft claims that the system is “about equal” to professional human transcriptionists, which uses the company’s homegrown deep learning Computational Network Toolkit, to achieve the feat. Microsoft will reportedly use the technology in its intelligent personal assistant Cortana across platforms.
