New technology flagged by Microsoft lets you converse in any language instantaneously in your own voice.
Using deep neural networks, in which a machine’s learning is patterned after human brain behaviour, the technology takes accuracy and speed of translation far beyond what was thought possible, Microsoft says in a YouTube demonstration.
It takes data about the user’s voice to recreate it in another language with an accuracy that takes the previous best-case error rate of one word in four to one word in seven or eight.
Although research into computer speech recognition has been underway for the past 60 years, until recently even the best systems had word error rates of between 20-25%.
This new technology has an error reduction rate of 30%.
Rich Rashid, head of Microsoft Research, says this is the biggest breakthrough in the accuracy of speech recognition in more than 30 years, suggesting it could lead to business opportunities.