Decorrelating Feature Spaces for Learning General-Purpose Audio Representations

We introduce DECAR, a self-supervised pre-training approach for learning general-purpose audio representations. Our system is based on clustering: it utilizes an offline clustering step to provide target labels that act as pseudo-labels solving prediction task. develop top of recent advances in computer vision and design lightweight, easy-to-use scheme. pre-train DECAR embeddings balanced subset the large-scale Audioset dataset transfer those representations 9 downstream classification tasks, including speech, music, animal sounds, acoustic scenes. Furthermore, we conduct ablation studies identifying key choices also make all our code pre-trained models publicly available.