Minje Kim, an assistant professor of intelligent systems engineering at the Luddy School of Informatics, Computing, and Engineering, is part of a group that has been honored with a 2020 Best Paper Award by the Institute of Electrical and Electronics Engineers (IEEE) Signal Processing Society.
The IEEE Signal Processing Society Best Paper Award is one of the most prestigious paper awards in the signal processing field. Kim published the paper, “Joint Optimization of Masks and Deep Recurrent Neural Networks for Monaural Source Separation,” in 2015 with colleagues from the University of Illinois. The IEEE Signal Processing Society confers honors for past papers that have proven to have made an impact on the field.
The research focused on the use of deep recurrent neural networks to help separate individual audio signals from a source that features several mixed signals. The process could be used for monaural source separation with applications in speech recognition, singing voice separation, and speech denoising tasks.
“I'm deeply humbled to receiving this prestigious award,” Kim said. “Since beginning to work on deep learning techniques, the field has changed a lot. It's a rapidly changing area, leading the advances of AI. What we did back then might be something primitive compared to what's happening now. However, I also came to believe that if researchers are excited about solving a particular problem, and if they succeed in building a team that respects one another, research can result in something widely well received.”
The study changed the direction of Kim’s research at the time.
“Since it was my first deep learning project that led to multiple publications, I realized the power of deep learning and became much more serious about it,” Kim said. “Since I was always into the idea of making an AI model more compact, I was able to think of another idea about compressing deep neural networks. I also believe that one of the reasons for the paper’s success was the open-sourced nature of the project. I started to care more about disseminating my project results in the form of an open-sourced effort whenever possible.”
Kim is currently the director of the Signals and Artificial Intelligence Group in Engineering at the Luddy School, and he continues to focus on developing machine learning models for audio signal processing applications, such as speech enhancement, source separation, MPEG audio coding, music information retrieval, etc., stressing their computational efficiency in resource-constrained environments or in applications involving multichannel observations.
The paper will be awarded the honor during the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2021) in Toronto in June.
“Few aspects of research are more rewarding than seeing your work make a long-lasting impact in your field,” said Kay Connelly, the associate dean for research at the Luddy School. “Minje’s research continues to serve as a foundation for advancements, and this recognition only helps reinforce the leadership role played by Luddy faculty in areas across computing.”