Stationary Bionic Wavelet Transform and Teager Energy Operator for Speech Enhancement

In this paper, we have proposed a new speech enhancement technique based on the time adaptation of stationary bionic wavelet thresholds. The time dependence is introduced by approximating the Teager Energy (TE) of the stationary bionic wavelet coefficients. The Stationary Bionic Wavelet Transform (SBWT) was used in this work, in order to have a perfect reconstruction of the speech signal. Experimental results based on Mean Square Error  (MSE) between the original speech signal and the reconstructed speech signal, show that the SBWT is better than the Bionic Wavelet Transform (BWT). This fact can improve the speech enhancement system based on thresholding in the bionic wavelet domain. We have also used the pause detection to introduce more improvement of the noise level estimation in the stationary bionic wavelet subbands. We have also used the time-space adapted thresholds in order to improve the procedure of stationary bionic wavelet thresholding and avoiding severs degradations of the enhanced speech signal. The proposed technique was evaluated by comparing it to others techniques such as the technique based on Wiener filtering and the techniques based on thresholding in wavelet and bionic wavelet domains and the technique based on undecimated wavelet packet-perceptual filterbanks and MMSE-STSA estimation. The two thresholding techniques are based on time adaptation of wavelet thresholds too. The obtained results from the SNR, SSNR, ISd and PESQ computation show that our proposed technique outperforms the others techniques.