A method for noise reduction in speech signal based on singular value decomposition and genetic algorithm

This paper presents a new approach to speech signal enhancement. The proposed technique is based on singular value decomposition (SVD) which is in time domain and separates the signal and noise subspaces. Genetic algorithm is used for optimizing the SVD method. It is a random technique based on natural evolution. In this approach, genetic algorithm estimates the initial parameters needed for SVD in order to enhance a speech signal. The basic algorithm is developed using white noise, but we have shown that it can easily be extended to the case of colored noise using some prewhitening techniques. The experimental results indicate that the presented algorithm improves SNR considerably.