Microsoft has successfully developed a speech recognition system that can accurately recognize the words of a conversation like humans do. The Microsoft Artificial Intelligence and Research team reported about the speech recognition system that makes fewer or sane errors than professional transcriptionists. Last month the researchers reported the word error rate of 6.3 percent whereas in the current month it comes down to 5.9 percent. The 5.9-word error rate is the lowest ever recorded error rate against the “switchboard” standard speech recognition task. In the Microsoft blog post-Xuedong Huang, the company’s chief speech scientist said that we have reached human parity and this is a historic achievement. The milestone implies that for the first time in human history, a computer becomes able to recognize the words in the conversation as well as a person would. The research in speech recognition was begun in the early 1970s with DARPA and the research milestone came after the research of decades. The milestone is going to have broad implications for the business products and consumers that can be augmented by the speech recognition.
It includes the consumer entertainment devices like Xbox, personal digital assistance such as Cortana and accessibility tools including instant speech-to-text transcription. This will make the Microsoft personal assistant Cortana more powerful than ever, making it a truly intelligent assistant possible. The Microsoft team is now using Computational Network Toolkit (CNTK) to reach the human parity milestone. The research team has made this home-grown system for deep learning available on GitHub through an open source license. The Microsoft’s Computer Network Toolkit runs on the specialized chip called as Graphic Processing Unit that has the ability to process the deep learning algorithms. It has vastly improved the speed and through which team was able to do research and reached the human parity. Furthermore, Researchers are now working more on various methods to make sure that the speech recognition system works well in the real-life settings. This includes the places with a lot of background noise like a party or a busy highway. Researchers in the long term are going to focus on the methods to teach the computers to transcribe the acoustic signals that come out of the people’s mouth and to understand the words that they are saying.