Independent component analysis for simultaneous active noise canceling and blind signal separation

A new algorithm is presented to perform active noise canceling and blind signal separation simultaneously. In many real-world problems speech signal is contaminated by noises, some of which are completely unknown while the other may be estimated by microphones located near the noise sources. Electric 'line' signals of audio equipments may also be used to estimate the noises. The active noise canceling removes the estimated noises with reverberation, while blind signal separation extracts speech signal from unknown noisy mixtures. Both algorithms are based on independent component analysis (ICA), which assumes statistical-independence among acoustic sources. The ICA-based active noise canceling utilizes higher-order statistics, and outperforms the standard least-mean-square (LMS) algorithm with quadratic statistics in real-world applications.