Using the CSM and VSM techniques to speed up the ICA algorithm without a loss of quality

Abstract

Abstract:In blind source separation problems that are implemented based on the independent component analysis (ICA) algorithm, the separation speed and quality are related inversely. In this paper, the proposed algorithms eliminate this tradeoff by generating a faster separation while maintaining the quality. In the proposed algorithms, in each frequency bin and in all of the learning steps, the separation quality of the separating matrix is compared with another one that we define as a situated matrix, and the best matrix is considered as an initial separating matrix in the next learning step.In this paper, we propose 2 algorithms based on the constant situated matrix (CSM) and the variable situated matrix (VSM). Using the simulation results, on average, the proposed CSM and VSM algorithms are about 3 and 6 times faster than the ICA algorithm, respectively, while the quality of the separated signals remains almost unchanged or becomes slightly better.

High-speed separation of anechoic mixtures of speech signals using fusion of iterative and closed-forms separation method

“…Here we apply the FICS to the VICA [30] to introduce the FVICA (FICS+VSM+ICA) algorithm and investigate its separation speed and quality. Signal to distortion ratio (SDR), signal to interference ratio (SIR), and perceptual estimation of speech quality (PESQ) are calculated for measuring the separation quality (see Appendix 2, on the journal's website).…”

“…Although SICA is at least 3 times faster than ICA, still this result can be further improved. In [30], we presented the variable situated matrix (VSM) technique to speed up ICA without loss of quality, leading to introduction of the VICA algorithm. In the VICA algorithm, in all learning steps of each frequency bin the separation quality of the separating matrix is compared with our defined situated matrix and the best one is selected as an initial separating matrix in the next learning step.…”