Darius Petermann, an Intelligent Systems Engineering Ph.D. student at the Luddy School of Informatics, Computing, and Engineering, recently received a Best Student Paper Award at the prestigious IEEE International Conference on Acoustics, Speech and Signal Processing 2023, the world’s premier technical conference focused on signal processing and its applications.
The annual ICASSP conference was held in Rhodes Island, Greece with more than 7,000 attendees. A total of 6,127 papers were submitted. Typically, half of the papers are student-authored. Five received best-student-paper recognition.
The paper, “Hyperbolic audio source separation,” introduced a novel framework for audio source separation using hyperbolic embeddings.
Audio source separation consists in extracting the various constitutive sources found in a given audio mixture, such as individual instruments or a voice in a musical recording.
“In this work,” Petermann said, “we exploit the ability of hyperbolic manifold to compactly represent hierarchical structure in order to learn hierarchical relationships between sources and time-frequency features.”
Hyperbolic embeddings, a special kind of machine learning, can improve audio source separation better than other methods.
Petermann’s work was done during his summer internship at the Massachusetts-based Mitsubishi Electric Research Laboratories. MERL is renowned for its work in artificial intelligence, computer vision and robotics.
“It was his second internship at MERL,” said Minje Kim, ISE associate professor and Petermann’s Ph.D advisor, “which signifies his academic excellence approved by the MERL researchers.”
Kim said the internship project was separate from Petermann’s Ph.D. research, and indicates his ability to do impressive research independently.
“Internships are an integral part of the ISE Ph.D. program,” Kim said. “I hope this success inspires students to go out and solve real-world engineering problems.”
ICASSP covers a wide range of topics in signal processing. Those include acoustics, speech processing, audio signal processing, image processing, machine learning for signal processing, signal processing for communications, signal processing for radar and sonar, signal processing for medical imaging, and signal processing for security and defense.
ICASSP attracts thousands of professionals from academia, industry, and government. It is a premier forum for sharing new research results, learning about the latest advances in signal processing, and networking with other professionals in the field.