Speech based gender identification using empirical mode decomposition (EMD)

Abstract

Traditionally for feature extraction, decomposition techniques such as Fourier decomposition are used to capture signals. But those methods have some margins like – it only works for linear and stationary data. On the other hand, in real world, we found data that are in non-linear and non-stationary. EMD or Empirical Mode Decomposition technique is a new approach introduced by Huang et al (1998) that can take any complicated signal and decomposed it to IMF. It extracts the amplitude and frequency information of a signal at a particular time. It is robust for non-linear and non-stationary signal processing. In this paper, I am using EMD as a new approach for gender identification based on speech signal. Gender identification based on the voice of a speaker consists of detecting if a speech signal is given by a male or female. Detecting the gender of a speaker has several applications.