Wired In: Deming Chen

Wired In: Deming Chen

Each week, staff writer Paul Wood chats with a high-tech difference-maker. This week, meet DEMING CHEN, a University of Illinois professor and co-founder of Inspirit IoT, which aims to improve security — starting on this campus.

How did you become an entrepreneur in addition to a high-tech researcher?

Actually, I was involved with two startups previously when I was a Ph.D. student at UCLA. Both companies were founded by my Ph.D. adviser. Through those experiences, I realized that the fastest way to transfer the most-recent and promising research results from your own lab to the industry is doing it yourself because you are the most familiar person with the technology, and you believed it enough to have started that research at the first place. So later on, when I became a professor myself, I am convinced that the goal of my research is not just producing research papers and Ph.D. students but also connecting research results to practices and making research innovation useful in the real world.

Who else is on the team?

The startup is actually incubated in the Advanced Digital Sciences Center, a research center operated by Illinois faculty members but located at Singapore.

You set out to solve a security problem at the Urbana-Champaign campus all 6,370 acres of it.

Yes. Once I realized that the campus covers such a large area, I thought that we had to have a cheap device, so the campus can afford to install them in many places. But it needs to be a smart device, so it can help campus to achieve the goal of being a peaceful and safe community for learning.

What is Inspirit IoT's low-cost solution for an audio-based security system that can alert a central command center when trouble breaks out?

We have designed and manufactured a new printed circuit board, which consists of four microphones, associated amplifiers and analog-to-digital converters. This board interfaces to the Lattice XO3 FPGA board through a pin-header. The XO3 FPGA receives the digitalized audio, normalizes the amplitude, and extracts four digital-signal-processing features. These features consist of short-term energy, subtracting windowed median, zero-crossing rate and auditory roughness. The FPGA buffers a window of sound samples and computes the features across the window. The features are then fed into a small classifier module on the FPGA, which detects and classifies the event type, while rejecting non-suspicious events. Suspicious audio events are combined with sound localization data, which provides an event direction using the microphone array. Such information can then be sent to a central command center to trigger a security alarm.

How big is the system and how much does it cost to build?

The system is about 10 centimeters long and 2.5 centimeters tall. It would cost about $25 to build. For mass production, the price can be further reduced. It can operate by itself if connected to a battery pack or a solar panel, or plugged into another embedded system, such as a camera system.

Has it been tested in real life?

We are developing a new version of the board. Once that board is done, we plan to try it out in real life under different use scenarios, such as campus security surveillance and smart home care.

How will 'smart' sensors only report suspicious activity — gunfire or screams — to protect privacy and preserve computational power?

Our device is designed to detect suspicious activities, such as gunshot or screaming, in real time as opposed to recording audio and streaming it somewhere else. Thus, our device is not spying on everything people say, and can protect privacy. Meanwhile, since our system is only looking for suspicious sounds, it does not need a large deep neural network to perform detailed speech recognition, which means we just need a small chip, saving computational power significantly.

Can it be combined with a network of security cameras?

Yes. Smart sound can work together with security cameras to make the whole system smart. For example, once the sound device detects a suspicious sound, the security camera can stream video of the event to a human operator. The human operator can watch the video in real time to make the eventual decision to activate an alarm or to notify authorities; such a human in the loop strategy can significantly reduce false positives. Meanwhile, since our device can localize the sound, it can also guide the camera to mechanically turn to the direction of the suspicious sound, zoom in, and trace the event in real time.

How do you do it so inexpensively?

Through a set of smart designs for both the physical board and the algorithm run on the board. The physical board is designed using the minimal number of physical components possible while embedding a cheap FPGA chip as the computing component. Meanwhile, our sound recognition algorithms are custom designed targeting the limited computing resource on the board. Such a software-hardware co-design strategy brings the overall cost down while still maintaining a high quality of results.

The project won first place at the IEEE/ACM Design Automation Conference 2017 International Hardware Design Contest this summer. Will that stimulate the growth of the start-up?

Certainly. We are using this technology to target several interesting applications. The first is smart campus surveillance. Currently, all the cameras installed at UI campus are just recording devices. They are used to review what happened after a criminal event already took place. If a smart sound board, such as ours, can be plugged into the camera system, an alert signal can be sent to a central station or to a policeman nearby when the sound board detects gunshots, people crying for help, continuous screaming, etc., and the central station and the policeman can react quickly. This will make the system proactive instead of reactive. Another example is smart home aid. Societies are aging, especially in some Asian countries. As a result, many elderly people are living alone. If we can install smart sound devices at homes that can send an alert when an elderly person falls down or cries for help, it can provide smart home care, and also give people a sense of safety and security.

What's your best advice for someone who's starting up?

Be open-minded and look for the opportunities surrounding you. Even many doors are closed, if you don't give up and keep searching, there will be a new door open to you.

Did you ever make any mistakes in your early years?

I took a winding path when I was young. I was a chemistry major when I did my undergraduate study in China. After I came to America, I "wandered in the wilderness" for several years, and then decided to switch to computer science for another undergraduate degree. After that, I joined a company as a programmer. Eventually, I decided to go back to school for a Ph.D. But when I graduated with a Ph.D. in 2005, I was already 37 years old. My young colleagues entering the university as assistant professors were mostly 27 years old. So I am 10 years late. But I know God gives people second chances, and that is why I am here.


Favorite app: WeChat

On Facebook, I follow: Mainly my friends and some departmental and university news.

Book or Kindle? What are you reading right now? Strangely, I am not a fan of books. If I have time, I prefer watching a movie on Netflix. It would be even better if my wife can watch it with me.

Do you have any wearable electronics? The Fitbit. I mainly pay attention to my sleep quality and the steps I take each day.

Do you have an entrepreneur hero? Jack Ma of Alibaba. Ma took four years to pass the Chinese college-entrance exam. That must have been very hard for him. It means that he entered college when he was 22, definitely a late start. He later became a lecturer of English and international trade in a common Chinese university. He applied for Harvard 10 times and got rejected. However, he never gave up, and he is bold and wise to have delved into the rising power of Internet in China in the mid-'90s. Now, in 2017, Fortune ranked Ma second on its World's 50 Greatest Leaders list. It is amazing.