The paper proposes an improved model of Independent Low-Rank Matrix Analysis (ILRMA) for dealing with Blind Source Separation (BSS) of audio and speech signals. The new model takes into account the dependency between spectral coefficients from different frequency bands, using Sinkhorn divergence to optimize source model parameters. To reduce the algorithm complexity, the Kronecker product is applied to decompose the modeling matrix. The use of cross-band information improves the BSS performance, but also significantly increases the number of parameters to be estimated and computational complexity. An efficient algorithm is developed to implement the Sinkhorn divergence based BSS algorithm, reducing complexity by an order of magnitude.
Publication date: 3 Jan 2024
Project Page: https://arxiv.org/abs/2401.01762v1
Paper: https://arxiv.org/pdf/2401.01762
