An Improved Speech Enhancement Method based on Teager Energy Operator and Perceptual Wavelet Packet Decomposition

Abstract

According to the distribution characteristic of noise and clean speech signal in the frequency domain, a new speech enhancement method based on teager energy operator (TEO) and perceptual wavelet packet decomposition (PWPD) is proposed. Firstly, a modified Mask construction method is made to protect the acoustic cues at the low frequencies. Then a level-dependent parameter is introduced to further adjust the thresholds in light of the noise distribution feature. At last the sub-bands which have very little influence are set directly 0 to improve the signal-to-noise ratio (SNR) and reduce the computation load. Simulation results show that, under different kinds of noise environments, this new method not only enhances the signal-to-noise ratio (SNR) and perceptual evaluation of speech quality (PESQ), but also reduces the computation load, which is very advantageous for real-time realizing.