A new preprocessor for speech recognition system using blind signal separation

Abstract

In previous years, numerous studies to pick up a target signal under a noisy environment are proposed. Especially, a blind signal separation method attracts the interests of many researchers because of its flexibility of system configuration. However, there are several problems to apply actual systems; required computational load and insufficient, stability to convergence. In this paper, a new simple blind signal separation method is proposed for 2-inputs and 2-outputs system. The results of computer simulation show the coherence between original signal and estimated one is more than 0.99 even if the level difference between signal and noise is 40 dB.