Congratulations to ECE undergraduate students Stephen Xia, Tianyi Yao, and Zichao Wang, for placing first in the 2015 IEEE Region 5 Student Paper Awards, held in New Orleans this past weekend. The group were finalists after placing first in the South Area Competition held last month.
The paper, titled, “Selective Transparent Headphones,” explores a system that forwards certain sound signals to a headphone wearer, while suppressing the rest. Specifically, the system would separate human speech from background noise, forwarding only the human speech to the headphone wearer.
“When you are wearing a pair of good noise isolating headphones and are listening to music, you might want to hear some of the sounds around you, such as people talking to you. We think it would be great if headphones could selectively choose important sounds around you, and propagate these sounds to the music listener. The music listener could listen to music, and at the same time hear important sounds around him, for example speech or a car horn,” Wang explained.
The group implemented a blind-source separation algorithm to separate statistically independent sound sources, and an artificial neural network to computer the amount of human speech in each source. The system is the first step in the design of a pair of selective transparent headphones that can function in a real-world setting.
“Zichao had been interested in doing something related to acoustical signal processing, while Tianyi had always been interested in machine learning. As a result, these were two of the major elements that influenced the subject of our project and how we attempted to complete it,” Xia said of the project’s origins.
“This is definitely one of the most thrilling projects I have ever worked on. I have been interested in machine learning since freshman year and this project gives me the unique opportunity to incorporate machine learning algorithms into a real-time signal processing system,” Yao said.
Yao was responsible for the construction and training of the artificial neural network, and Xia and Wang were responsible for implementing the blind-source separation algorithm.
“The team is really great because we all wanted to do something related to music, and all of us bring different expertise necessary to this project. Stephen and I both know about music - Stephen plays the piano and I play the violin. Tianyi and Stephen are our experts in machine learning and Blind Source Separation, respectively,” Wang explained.
Dr. Richard Baraniuk, Victor E. Cameron Professor of Electrical and Computer Engineering, and Dr. Mohammad Golbabaee, Postdoctoral Research Associate in the DSP group, advised the paper.
“I was thrilled when I found out that we were nominated as finalists [of the IEEE Region 5 Competition]. I felt that the hard work we have put into this project has been recognized and we are highly motivated to further build on this project,” Yao said.
“There is a feeling of accomplishment whenever your peers commend you for any work you have produced. This time is no exception,” Xia added.