Although speech recognition is considered to be one of the most widely used human-computer interaction methods, its experience does not seem to be ideal. The recognition speed is slow and the accuracy is low. This is the impression of many users on speech recognition. However, the researchers' tests showed that speech recognition can organize text messages faster and more accurately than manual typing on mobile phones.
Computer scientists from Stanford University, Washington University, and Baidu recently conducted a test: Experiments with Chinese and English typing contests using the iPhone as a hardware device. One party was the "Deep Speech 2" software released by Baidu Silicon Valley Artificial Intelligence Lab last year. For the representative, the other party is 32 typing experts aged 19 to 32 years old. The recording speed and accuracy are recorded by the back-end test software.
James Randy, a professor of computer science at Stanford University, said, "These people have been typing, so we've made speech recognition software match the typing that these people are good at."
The results show that the speed and accuracy of speech recognition are better than those of traditional keyboard input, whether English or Chinese.
English: Speech recognition is three times faster than typing, and the error rate is 20.4% lower. Mandarin Chinese: Voice is 2.8 times faster than typing, and the error rate is 63.4% lower.
According to Shan Shanshan, a doctoral student in the Department of Computer Science at Stanford University who participated in the study, “We knew that the effectiveness of speech recognition was good, and the expected recognition speed was faster than keyboard typing, but the results were surprising. This speed was almost 3 times faster. ."
Of course, Baidu's Deep Speech 2 is not just an example. It is also applicable to other speech recognition software. Randy said, "In the past two to three years, thanks to big data and deep (artificial intelligence) learning, speech recognition has indeed improved a lot, and faster and more accurate results can be achieved."
So these researchers believe that other new versions of high-accuracy speech recognition software can achieve the above results.
Recommended reading:
Where are the pain points for speech recognition, and how do you do it from interaction to precise identification?
Decline and Self-Redemption of Nuance, the World's Largest Speech Recognition Company