News & Events

Efficient Neural Audio Processing Models

Minje Kim, Indiana University

Wegmans Hall 1400

Abstract: While advancements in deep learning technology have led to a near-human performance in many AI related tasks, the typical success criterion has usually been the accuracy of the model. Network compression is an emerging research area, where the goal is to streamline the neural network inference process, so that the trained models can be deployed to small devices with limited computing resources. In this talk, we will explore various network compression options, such as low-bit quantization, hashing, perceptual loss functions, and architectural considerations. We examine the performance of those efficient models in the context of audio signal processing, especially in terms of speech enhancement quality and speech coding efficiency.

Bio: Minje Kim is an Assistant Professor and the Director of graduate Studies in the Department of Intelligent Systems Engineering at Indiana University. He is core faculty of the Data Science and Cognitive Science programs, and adjunct faculty of Department of Statistics. He earned his PhD in the Department of Computer Science at the University of Illinois at Urbana-Champaign. Before joining UIUC, he worked as a researcher in ETRI, a national lab in Korea, from 2006 to 2011. His research focuses on developing machine learning algorithms applied to audio processing, stressing their computational efficiency in the resource-constrained environments or in the applications involving large unorganized datasets. He received Richard T. Cheng Endowed Fellowship from UIUC in 2011. Google and Starkey grants also honored his ICASSP papers as the outstanding student papers in 2013 and 2014, respectively. He is an IEEE Senior Member and also a member of the IEEE Audio and Acoustic Signal Processing Technical Committee.