Language models have been successfully used to model natural signals, such as images, speech, and music. A key component of these models is a high quality neural compression model that can compress high-dimensional natural signals into lower dimensional discrete tokens. To that end, we introduce a high-fidelity universal neural audio compression algorithm that achieves ~90x compression of 44.1 KHz audio into tokens at just 8kbps bandwidth. We achieve this by combining advances in high-fidelity audio generation with better vector quantization techniques from the image domain, along with improved adversarial and reconstruction losses. We compress all domains (speech, environment, music, etc.) with a single universal model, making it widely applicable to generative modeling of all audio. We compare with competing audio compression algorithms, and find our method outperforms them significantly. We provide thorough ablations for every design choice, as well as open-source code and trained model weights. We hope our work can lay the foundation for the next generation of high-fidelity audio modeling.

Graph Convolutional Neural Networks (GNN) based recommender systems are state-of-the-art since they can capture the high order collaborative signals between users and items. However, they suffer from the feature leakage problem since label information determined by edges can be leaked into node embeddings through the GNN aggregation procedure guided by the same set of edges, leading to poor generalization. We propose the accurate removal algorithm to generate the final embedding. For each edge, the embeddings of the two end nodes are evaluated on a graph with that edge removed. We devise an algebraic trick to efficiently compute this procedure without explicitly constructing separate graphs for the LightGCN model. Experiments on four datasets demonstrate that our algorithm can perform better on datasets with sparse interactions, while the training time is significantly reduced.