A Speech Enhancement System Based On LMS Algorithm

In speech signal detection process, signal is always very weak, and it's easily be interrupted and overwhelmed by noise, so it is difficult to extract pure speech signal from contaminated signal. One of the key technologies to solve this problem is speech enhancement. In order to eliminate the influence of noise and improve the quality of collected speech signal, People designed a variety of speech signal processing systems based on various speech enhancement methods.This paper describes a variety of speech enhancement methods, such as spectral-subtraction method, wavelet transform method, Kalman filter method and so on, and compares their advantages and disadvantages based on computer simulation results. Adaptive noise cancellation method is one of the most widely used methods for speech enhancement at present. It mainly consists of two parts:adaptive filter with adjustable parameters and adaptive algorithm.This paper mainly introduces the principle and application of adaptive filtering algorithm. The step factor of traditional adaptive least mean square algorithm (LMS) is fixed, which lead to the contradiction of convergence rate and steady-state error. In order to fix this defect, this paper introduces a kind of SNR-based variable step size adaptive noise cancellation (LMS) algorithm. Through the comparison with VS-LMS algorithm, we can find that this improved algorithm has faster convergence rate, smaller Steady-state error, lower computational complexity and better Noise immunity.This paper takes mine accident observation system as its research background, builds a speech enhancement system based on the improved variable step size LMS algorithm using GUI tools of MATLAB. In order to verify the validity of this system, it adds two kinds of noise to the pure speech signal:impact noise and white Gaussian noise. From the computer simulation results we can find that this speech enhancement system has good noise reduction effect.