Abstract
A common problem for modern communications and audio recordings is background noise caused by automobiles, cars, trains, planes and alike. A common remedy is to use speech signal enhancement methods to improve the perceptual quality and intelligibility or to restore the contaminated speech signal back to its original form. It is well known that denoising is a compromise between the removal of the largest possible amount of noise and the preservation of signal integrity. To address this issue, various speech signal enhancement techniques were proposed; this paper proposed a new hybrid technique for speech signal enhancement which combines spectral subtraction and Kalman filtering methods. Firstly, the spectral subtraction pre-processed the noisy signal in order to initially reduce the noise level, the quality of the speech signal is then improved by Kalman filtering method. The effectiveness of the proposed method was evaluated by using white and coloured noise (car and street) from the Noizeus database. Our Results demonstrate that the hybrid approach consistently outperforms individual spectral subtraction and Kalman filtering methods. For white Gaussian noise, an average percentage SNR improvement of 10.7% over spectral subtraction and 2.7% over the Kalman filter was achieved. Higher performance gains are recorded over spectral subtraction and Kalman filtering methods particularly on real-world noises, such as car and street noise.

Key words: Speech signal enhancement, Spectral Subtraction, Kalman Filtering, Denoising, Signal-to-Noise Ratio