Microsoft, who has been working on a new speech recognition system, announced a breakthrough as their new product is just as exact as a human.
The team of scientists to achieve the possibly historic feat is part of Microsoft’s Artificial Intelligence and Research division. According to the statement, they have developed a speech recognition system that can recognize and transcribe words just as well as a human.
The speech recognition systems area first started being explored back in the 1970’s with the DARPA, a U.S. agency whose task was to accomplish technological breakthroughs.
Their mission later garnered the help and collaboration of many other research organizations and also of most of the industry’s big companies.
The new speech system developed by Microsoft may come to mark a milestone in the technology’s timeline and is sure to garner a widespread usage.
The technology could place its mark on quite a wide variety of domains as most business and consumer products could greatly benefit from better, more exact speech recognition features.
Microsoft’s new speech recognition software system registered the best results when compared to other similar systems, and also a human level of accuracy.
Nonetheless, this doesn’t mean that it is perfect. Just as us humans, the computer had a 5.9 percent word error rate (WER). The number dropped as compared with the team’s 6.3 percent result reported last month and came to fit into the human parameters of understanding.
The same test was applied to a number of people that were asked to transcribe the same, exact Switchboard standard conversation.
The system’s 5.9 percent WER rate was just about equal to most people’s results, and almost the same with that of the professional transcriptionists’ which participated.
As such, the company’s chief speech scientist, Xuedong Huang, declared that the computer has now acquired human parity as their new test results were the lowest to ever be recorded, and also the closest to the human level of understanding.
The company has now reached the goal they set for themselves no more than a few months ago, and which was helped by the use of the latest technology in the neural network area.
As the respective network technology was used in all the system’s configurations, it allowed for a better word to word generalization, based on neural language models.
The new speech recognition system could come to be used not only in AI digital assistants such as Microsoft’s Cortana, which will certainly become smarter and more powerful, but also in other consumer entertainment areas.
Devices such as Xbox and game consoles, accessibility tools and instant speech-to-text apps and transcription could be the next beneficiaries of the system, and the speech recognition’s usability could also span well beyond this said areas.
Image Source: Wikimedia