Microsoft develops human-like speech recognition system
Photo shows the team responsible for the breakthrough, although their work has just begun because the next step is implementing the new system into Microsoft products. Image Microsoft
After more than 20 years of research, a team at Microsoft has achieved a noteworthy breakthrough in speech recognition. The system that they developed is said to be as good as human hearing.
To be exact, the system’s “word error rate” is tagged at 5.9%, which is on the same level as a professional human transcriber. It is definitely more than good enough for conversation, reports TechCrunch.
Speech recognition has been pursued by researchers for decades. While quality had steadily but slowly improved, the latest advances have been achieved through the use of neural networks and machine learning.
The research paper states, “Our progress is a result of the careful engineering and optimization of convolutional and recurrent neural networks. These acoustic models have the ability to model a large amount of acoustic context.”
The team made full use of Microsoft’s open-source Computational Network Toolkit to achieve its goals.
There has been no word yet about the new system’s implementation on Microsoft’s product but their AI assistant Cortana can easily be assumed as a prime candidate to receive this technology update to better serve users and one-up Google’s own virtual assistant. Alfred Bayle
