A novel preprocessing method is proposed. It has a reduced complexity and therefore is aimed to be used in low power, VLSI implemented, speech recognizers. Our algorithm extracts a feature vector made from up to 3 feature ...