Microsoft has announced its speech recognition technology now performs as well as the average human.
In a paper published by the firm’s Artificial Intelligence and Research division, engineers said the word error rate was now 5.9 per cent. That’s right where humans, with proper organic ears and brains are too.
The results were gleaned by asking humans to transcribe the same conversation. Last month, Microsoft got its score down to just 6.3%, but the new mark is the lowest error rate ever recorded.
Related: Windows 10 Anniversary Update review
“We’ve reached human parity,” said Xuedong Huang, the company’s chief speech scientist in a blog post. “This is a historic achievement.”
Harry Shum, EVP of the division added: “
In real world terms, Microsoft says this will improve Cortana and speech input on the Xbox platform. It’ll also boost accessibility tools and speech-to-text transcriptions services Microsoft says.
“This will make Cortana more powerful, making a truly intelligent assistant possible,” Shum added.
Microsoft says the next step is to go from recognition to understanding.
Video: iPad Pro vs Surface Pro 4
Editorial independence
Editorial independence means being able to give an unbiased verdict about a product or company, with the avoidance of conflicts of interest. To ensure this is possible, every member of the editorial staff follows a clear code of conduct.
Professional conduct
We also expect our journalists to follow clear ethical standards in their work. Our staff members must strive for honesty and accuracy in everything they do. We follow the IPSO Editors’ code of practice to underpin these standards.
Editorial independence
Editorial independence means being able to give an unbiased verdict about a product or company, with the avoidance of conflicts of interest. To ensure this is possible, every member of the editorial staff follows a clear code of conduct.
Professional conduct
We also expect our journalists to follow clear ethical standards in their work. Our staff members must strive for honesty and accuracy in everything they do. We follow the IPSO Editors’ code of practice to underpin these standards.
