Microsoft claims to have made significant progress in getting its automated speech-recognition systems—as used in services such as Cortana—to do their thing as accurately as people can.
The company’s researchers have produced a new paper describing the advances they have made since last year, when they said their systems had achieved “human parity” with a 5.9% error rate in transcribing audio conversations.
Since that happened, researchers at rival IBM achieved an error rate of 5.5%—and while they were at it, they said their research showed real human parity would involve an error rate as low as 5.1%.
Now, Microsoft’s researchers say they’ve reached that new milestone, meaning they’re claiming for the second time to have developed speech-recognition systems with a human-level accuracy rate.
Get Data Sheet, Coins2Day’s technology newsletter.
“Reaching human parity with an accuracy on par with humans has been a research goal for the last 25 years,” the researchers said in a blog post.
The researchers said the improvement to Microsoft’s systems came from tweaks to its neural net-based acoustic and language models. Neural nets, or networks, are programs that take their inspiration from the workings of organic brains. These artificial brains provide a key building block for current work on “artificial intelligence,” and are used in speech and image recognition, among many other applications.
[Tempo-video id=”5541625085001″ account_id=”2111767321001″ autoplay=true]
“Moreover, we strengthened the recognizer’s language model by using the entire history of a dialog session to predict what is likely to come next, effectively allowing the model to adapt to the topic and local context of a conversation,” Microsoft’s researchers added.
