A Speaker Verification System Based on EMD

Most of the speech utterance feature extraction methods are based on the assumptions: utterance signal is short-term stable and independent between each other adjacent frames. This approach ignores the dynamic characteristics of speech signal. For the time-varying characteristics of the speech utterance, we propose a new feature extraction method based on empirical mode decomposition EMD. We can extract the LPCC feature parameters from different stages of IMFs which are decomposed by EMD process. A speaker verification system based on EMD is proposed and it is shown that it is better performance than the one with traditional LPCC features based on short-term process.