Active noise control technology is used by noise-canceling headphones to minimize or completely block out outside noise. However, despite the many advancements in technology, people still don't have much control over which sounds their headphones block out and which they let pass.
Now, deep learning algorithms (算法) have been developed by a group of academics at the University of Washington that enable users to select which noises to filter (过滤) through their headphones in real-time. The system has been named “semantic hearing” by its creators.
The AI-powered headphones remove all background noise by streaming recorded audio (音频) to a smartphone that is linked to the devices. Through this process, the headphone users can choose to strengthen or cancel out 20 types of sounds, using voice commands or a smartphone app. The headphones will then only let through the sounds that have been chosen by the wearer.
“The challenge is that the sounds headphone wearers hear need to syn (同步) with their visual senses. This means the neural (神经的) algorithms must process sounds in under a hundredth of a second.” said senior author Shyam Gollakota, a UW professor.
Due to this time constraint, the semantic hearing system chooses a process that relies on noises communicated on a device like a linked smartphone. Furthermore, in order for humans to continue to effectively experience sounds in their environment, the system needs to maintain these delays because sounds coming from different directions enter people's ears at different times.
Trials were undertaken by the researchers in a variety of settings. The semantic hearing system was able to isolate target sounds, while at the same time removing background noise. In terms of the system's audio output for the desired sounds. 22 participants gave it an average rating higher than they assigned to the original noise recordings.
There were, however, a few disadvantages: the AI-powered system occasionally had trouble recognizing sounds that were too similar. The researchers said that the system could produce better results if its machine learning models were trained on more real world data.
【小题1】What can deep learning algorithms do?| A．Improve users listening ability. |
| B．Help users remove unwanted noise. |
| C．Stop people from entering noisy areas. |
| D．Create communication between users. |
| A．Select headphone users. |
| B．Process data without noticeable delay. |
| C．Follow the way people speak. |
| D．Improve the quality of smartphones. |
| A．It has improvement in sound quality. |
| B．It helps them recognize participants' voices. |
| C．It has more disadvantages than advantages. |
| D．It's suitable to strengthen background noise. |
| A．The semantic hearing system still has some drawbacks |
| B．The semantic hearing system can recognize human speech |
| C．AI noise-canceling headphones let you choose what you hear |
| D．AI noise-canceling headphones now have a commercial version |
