Partitioned convolution is an important technique for low-latency filtering using long finite impulse response (FIR) filters. It is typically used to simulate, in real-time, the acoustic response of a space (convolution reverb).
Partitioned convolution operates by breaking up the impulse response of a long FIR filter into many smaller filters.
This talk presents an algorithm for optimizing the partition scheme for parallel computing architectures.

Bio:
Eric Battenberg is a grad student studying Electrical Engineering and Computer Science at UC Berkeley. He did undergraduate work at UC Santa Barbara, focusing on signal processing and high frequency circuit design. His research interests include music information retrieval, audio signal processing, applications of machine learning, and architecting parallel music software.