Improved Empirical Mode Decomposition Using Optimal Recursive Averaging Noise Estimation for Speech Enhancement

In this paper, a new approach for robust speech enhancement based on improved ensemble empirical mode decomposition (EMD) using optimized log-spectral amplitude noise estimation is presented. In this approach, a noisy signal is decomposed adaptively into a sum of oscillating components that belong to intrinsic mode functions (IMFs); then, each component is enhanced separately to provide less-corrupted IMFs that are used by the Hurst exponent method to construct an estimate of a clean signal. This new framework takes advantage of adaptive noise estimation performed by improved minima-controlled recursive averaging for noise estimation and optimally modified log-spectral amplitude to enhance the noisy EMD components. Through experimental evidence, the objective evaluation of quality and intelligibility demonstrates that the proposed method performs significantly better than the baseline techniques, including the most recently developed EMD-based speech enhancement methods.