"Computer speech recognition has reached the same level as human ears," Microsoft announced

ByDion gillard

The cognitive ability of a computer that is rapidly evolving has reached a new milestone. It was announced that a speech recognition system developed by a research team that conducts research on artificial intelligence (AI) at Microsoft has reached a level that can recognize human speech with precision not different from actual human beings and generate it in letters It was.

Microsoft says speech recognition technology reaches "human parity" - CBS News
http://www.cbsnews.com/news/microsoft-speech-recognition-technology-understands-conversation-as-well-as-people-do/


According to the announcement, the speech recognition system developed by the research group of the department "Microsoft AI and Research Group" aiming at the development of products and services utilizing AI at Microsoft,Accuracy of error recognition rate 5.9%It is possible to recognize with. It seems that the precision which is equal to or slightly higher than that of the person who works "text raising" which transcribes the contents which I said in the letter.

Xuedong Huang, who leads research at Microsoft, said about this performance, "In our development we were able to reach the same level as many people, which is a historical feat." Also, Harry Shum, executive in the division, said, "It was impossible to think that we could realize this kind of performance five years ago," revealed the rapid technological progress occurred It is.

Although Microsoft's speech recognition technology can be said to have reached a very high recognition rate, there are still cases where false recognition still occurs. For example, there are cases in which the word "Have" is incorrectly recognized as "is". However, it seems that this is heard at the same level even when people hear it, and it seems that mistakes occurred, it can be said that it is attributed to the original way of speaking rather than "misrecognition".

In order to achieve this performance, of course, the latest computer technology Deep learning is used. Improvement in accuracy has been achieved by using Microsoft's Computational Network Toolkit (CNTK) that improved processing speed using a dedicated chip and learning by deep learning.

Although it is a computer speech recognition technology that has reached the same level of precision as a human being, the next task is to improve the recognition rate in the same situation as human daily life. In addition to the good environment where the voice can be heard clearly, another new performance is required to correctly recognize the voice even in the situation where the surrounding noise exists, so the development of speech recognition in the real world is necessary in the future The situation that will come. In addition, the research team is looking at the development of technology that can hear even "until who is talking" by recognizing the difference in sound.

ByJD Hancock

When these technologies are realized, it is finally the time when robots live life like human beings arrive, and there is a concern that robots will be beyond human society as in the movie "Terminator" According to the research team, it seems likely that it will have a long way to reach that situation. According to Geoffrey Zweig of the research team, because the technology realized this time is "voice recognition" and the technique of "understanding content" is another thing, the research team's team said, "The next frontier to be aimed at is" It is to go from "recognition" to "understanding", "I am talking about the future path.