Speech recognition technology has become the highlight of the technology circle. Whether it is Baidu's secret or search for a robot that has just been developed to replace the interpreter, the speech recognition technology is really stronger. In this article, Jamie Condliff introduces new research and proves that artificial intelligence can not only recognize people's speech content through speech, but even if they can't hear the sound, artificial intelligence can read lip language smoothly, even more effectively than artificial.
As we all know, lip reading is very difficult, depending largely on the context of the language and its understanding, and these are only visually presented. But researchers are showing us that machine learning can identify lines of silent video more effectively than professional lip readers.
In one project, a team from the Department of Computer Science at Oxford University developed a new artificial intelligence system called LipNet. According to the Quartz news website, this system is based on GRID datasets, and GRID is read by people. A clear facial video compilation of seconds sentences. Each sentence follows a string of the same pattern.
The team used this data set to train the neural network, similar to the nature of performing speech recognition. In this process, although the neural network can recognize export-type changes over time, it can learn to relate this information to the interpretation of the content. But the artificial intelligence system does not analyze the footage continuously and intermittently, but considers the overall content so that it can understand the context from the analyzed sentences. This is very important because people's mouths are often much less than people's voices.
At the time of testing, this artificial intelligence system was able to accurately identify 93.4% of the words, and many artificial lip reading volunteers did the same test, but the accuracy rate was only 52.3%.
According to New ScienTIst, another team from the Oxford University's Department of Engineering Science who has been working with Google's DeepMind, an artificial intelligence system, has completed a more difficult task. They are not using a neat data set like GRID, but a series of 100,000 small videos taken from BBC TV. These small videos come in many languages ​​and have different lighting effects and movement of the speaker's head position.
The University of Oxford and DeepMind's team used a similar approach to successfully develop an artificial intelligence system with a recognition rate of 46.8%. This far exceeds the accuracy of manual recognition, and the accuracy of manual recognition to be error-free is only 12.4%. Of course, there are many reasons why accuracy is so low, including the transition from light and direction to deeper language complexity.
In terms of differences, these two experiments show that artificial intelligence systems far outweigh artificial recognition in lip reading, and it is not difficult to imagine that the application potential of such software is enormous. In the future, Skype can make up for a lot of deficiencies, such as when the caller is in a noisy environment, or that those who have hearing impairment can pick up the phone to "listen" what others are saying.
Strong and durable
Aluminum alloy plug, TPE cover
Hidden design
One second stretch, free storageUltrasonic welding pressure
Close fitting, firm resistance to fall
Strong and durable
Priority TPE
It's hard to break even if you stretch multiple times
Stretching or shrinking only one end can easily to damage the components and cause jamming
No winding
Stretch when used, shrink when not used
Five lengths
Each pause is a length, suitable for multiple occasions
Notice
Both cables are stretched at the same time
Do not stretch unilaterally
Data Cable Wiring,Original Data Cable,Computer Transfer Cable,Line Data Charging Cable
Guangzhou HangDeng Tech Co. Ltd , https://www.hangdengtech.com