Sparse Gaussian Process Audio Source Separation Using Spectrum Priors in the Time-Domain
This addresses the problem of scaling time-domain audio source separation for longer audio frames, though it is incremental as it builds on prior GP methods.
The paper tackled the computational complexity of Gaussian process audio source separation by introducing an efficient method using variational sparse GPs and spectral mixture kernels, which outperformed existing techniques like LD-PSDTF, KL-NMF, and IS-NMF.
Gaussian process (GP) audio source separation is a time-domain approach that circumvents the inherent phase approximation issue of spectrogram based methods. Furthermore, through its kernel, GPs elegantly incorporate prior knowledge about the sources into the separation model. Despite these compelling advantages, the computational complexity of GP inference scales cubically with the number of audio samples. As a result, source separation GP models have been restricted to the analysis of short audio frames. We introduce an efficient application of GPs to time-domain audio source separation, without compromising performance. For this purpose, we used GP regression, together with spectral mixture kernels, and variational sparse GPs. We compared our method with LD-PSDTF (positive semi-definite tensor factorization), KL-NMF (Kullback-Leibler non-negative matrix factorization), and IS-NMF (Itakura-Saito NMF). Results show that the proposed method outperforms these techniques.