Historic Achievement

Untitled Document

Human Parity in Voice Recognition


 

 

Voice recognition applications based on artificial intelligence processing of human speech have been around for quite some time and are becoming part of our everyday life. It is becoming more and more difficult to talk to a human when accessing different services and the quality has been improving, especially in the last couple of years.
Based on the recent announcements from Microsoft, it looks like humanity has reached another milestone in the integration of intelligent machines in our lives – the human parity.

As reported by Microsoft’s chief speech scientist Xuedong Huang, a word error rate (WER) of 6.3 percent is achieved in benchmark evaluation against the industry standard NIST 2000 Switchboard speech recognition task. It is the lowest in the industry and a tremendous improvement in comparison with the WER of 43 % from twenty years ago.
These results are made possible by the use of deep neural networks, inspired by the biological processes of the brain that utilize a new kind of cross-layer network connection. Another contributing factor is the sophisticated optimizations that enable deep learning algorithms to run an order of magnitude faster than before, as a result of a breakthrough for parallel training on graphics processing units (GPUs). 

Protecting your electrical assets? today and tomorrow