Multi-Scale Multi-Band DenseNets for Audio Source Separation

Explore a 24-minute presentation on Multi-scale Multi-band DenseNets for Audio Source Separation, delivered by the Fellowship.ai team. Delve into the novel network architecture that extends the densely connected convolutional network (DenseNet) to tackle the complex challenge of separating audio sources. Learn about the incorporation of up-sampling layers, block skip connections, and band-dedicated dense blocks to enhance performance in audio source separation tasks. Discover how this approach leverages long contextual information to outperform state-of-the-art results on the SiSEC 2016 competition, while requiring fewer parameters and less training time. Gain insights into the paper's methodology, problem definition, terminology, and comparisons with previous methods. Understand the intricacies of the proposed architecture, including composite layers and the multiband multiscale internet concept.