Sparse multi-label linear embedding nonnegative tensor factorization for automatic music tagging

In this paper, a robust framework for automatic music tagging is proposed. First, each music recording is represented by its auditory temporal modulations. Then, a multilinear subspace learning algorithm based on sparse label coding is proposed to effectively harness the multi-label information for dimensionality reduction. The proposed algorithm is referred to as Sparse Multi-label Linear Embedding Nonnegative Tensor Factorization. Finally, a recently proposed sparse representation-based method for multi-label data is employed to propagate the multiple labels of the training auditory temporal modulations to annotate the auditory temporal modulations extracted from a test music recording with the sparse ℓ₁ reconstruction coefficients. The proposed framework outperforms both humans and state-of-the-art computer audition systems in the music tagging task, when applied to the CAL500 dataset.