Empirical Mode Decomposition for Speech Synthesis

Abstract: A new method of Synthesis by Analysis for EMD multi-component signals of fast changing instantaneous attributes is introduced. It makes use of two recent developments for signal decomposition to obtain near mono-component signals whose instantaneous attributes can be used for synthesis. Furthermore, by extension and combination of both decomposition methods, the overall quality of the decomposition is shown to improve considerably summarize the use of Empirical mode decomposition (EMD) for denoising a speech signal. EMD, introduced by Huang et al in gives time-frequency representation of non-linear and non-stationary signals. It decomposes a signal into a sum of finite zero-mean, oscillating components called as intrinsic mode functions based on the local time characteristics of the signal. Main essence of the method is it�s adaptive and data driven nature.

Keywords: Empirical mode decomposition, IMF, Hilbert transform, Denoising technique.