The so-called "Cocktail Party Problem" means that the current speech recognition technology can already recognize a person's words with a high degree of accuracy, but when the number of people speaking is two or more people, the speech recognition rate will be greatly reduced. The computer cannot distinguish the target sound from the sound that other people emit at the same time. This is a well-known problem in the field of computer speech recognition.
Today, scientists from Duke University in the United States have skillfully solved this problem with a simple 3D printing device. Scientists have published a paper in the Proceedings of the National Academy of Sciences to introduce this wonderful 3D printing solution.
They first created a large, thick 3D printed plastic tray with 36 openings on one side of the plastic tray and a number of honeycomb channels leading to a microphone in the center of the plastic tray. It is essentially a single sensor listening system that combines acoustic metamaterials and compression sensing technology.
“Unlike previous studies that often relied on signal and speech processing techniques to solve the “cocktail party†problem, the approach we proposed was a unique hardware-based approach that utilized well-designed acoustic metamaterials.†Researchers write "We firmly believe that this approach will not only solve the cocktail reception problems that researchers in all fields have been hoping for over the past decades, but that the system design method that combines the design of the physical layer with the computational perception will be Traditional acoustic sensing and imaging methods have an impact."
The study was conducted under the leadership of Steven Cummer and Yangbo Xie.
So what is the working principle of it? According to the researchers, the 36 channels leading to the microphone each have a unique 3D printed shape that differentiates their functional characteristics, making the sounds transmit to the center in a subtle way that can be distinguished from each other and by a single sensor. receive. According to Yangbo Xie, we humans can't tell the difference, but the algorithm based on the sensor almost always tells us which of these sounds come from.
As for the correctness of the solution. The researchers stated in their paper: "This device with a compact array of resonant metamaterials has been proven to distinguish between overlapping audio from three separate sources with a correct rate of 96.67%." This simple but effective method has been hailed as an excellent solution. But the only problem at the moment is its size - it's about the equivalent of a very thick pizza, but obviously it has plenty of room to optimize the design and use it for further applications. Scientists at Duke University say it can be used for acoustic imaging and sensing applications such as hearing aids, so we may see a smaller version of the technology in the near future.
George Nelson Wall Clock,Home Eames Bird,Multicolored Coat Hanger,George Nelson Wall Ball Clock
Ningbo Realever Enterprise Limited , https://www.realeverfurnishing.com