The authors propose a new speech enhancement approach based on the application of the wavelet packet transform with an optimal decomposition. The approach uses the principal component analysis (PCA) and an improved version of the robust PCA by imposing the nonnegative factorization on the low-rank matrix. In order to detect noisy time–frequency zones, a subspace decomposition based post-processing technique was implemented. A number of simulations were then used to evaluate performance under various types of noise. Standard objective measures, as well as subjective evaluations, show that this approach outperforms the comparable speech enhancement methods for noise-corrupted speech at low levels of signal-to-noise ratios. The technique creates the least distortion in the enhanced speech while suppressing less noise than the on-line semi-supervised Nonnegative Matrix Factorization.
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.