New voice recognition software from Microsoft should understand words just as well as a human being. That claim has been made by researchers of the United States technology giant in a recently published scientific paper.
The computer system that has been developed by the researchers is able to dictate a conversation between two individuals. During experiments the computer made the same or less mistakes compared to its human counterparts.
The voice recognition software harnesses the power of a so called neural network, which allows it to make connections between certain words, greatly reducing the number of mistakes. Multiple computers needed to be connected in order to generate sufficient computational power.
Microsoft calls the system ‘historic’, but emphasizes that computers are still unable to write texts without making mistakes. When certain phrases and words are spoken out really fast they would still be written down incorrectly.
In the future the United States tech giant wants to use the voice recognition software for its virtual assistant Cortana, which already comes pre-installed in Windows 10. Aside that, accessibility functions for impaired users could be improved using the new voice recognition finds.