Evaluation of SIMO Separation Methods for Blind Decomposition of Binaural Mixed Signals
Tomoya Takatani, Satoshi Ukai, Tsuyoki Nishikawa, Hiroshi Saruwatari, Kiyohiro Shikano · Institutional Repositories DataBase (IRDB) · 2005
High-fidelity blind source separation (BSS) using Single-Input Multiple-Output (SIMO)-model-based Independent Component Analysis (SIMO-ICA) is now being studied by the authors. This paper describes a comparison of two types of SIMO-ICAs with different constrains and the conventional methods, and gives explicit discussion on the sensitivity of the parameters settings in the methods. In order to discuss the difference, the source-separation experiments using the mixed binaural sounds are carried out under the same real acoustic conditions. The experiment results reveal that SIMO-ICA-IG outperforms SIMO-ICA-LS and the conventional methods, and the parameter setting in SIMO-ICAIG does not depend on the source signals’ properties compared with that of SIMO-ICA-LS.