This article discusses a new approach to universal sound separation (USS), a method capable of separating various sources of sound. The authors focus on the concept of sampling frequency independence, which has been largely overlooked in USS research. They propose a network, SuDoRM-RF, that uses sampling-frequency-independent (SFI) convolutional layers. These layers can handle various sampling frequencies by generating convolutional kernels in line with an input sampling frequency. This advancement could significantly improve the performance of audio signal processing tasks.
Publication date: 25 Sep 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2309.12581
