Windowing functions for the average magnitude difference function pitch extractor

This paper describes the fundamental properties of the Average Magnitude Difference Function, (AMDF) when used for pitch extraction on human speech, such as the ability to work in the absence of the fundamentals, and responses to pitch and formant resonances. Requirements for the algorithm´s correct functioning are discussed, including reasonable bounds on window size and requirements on a nonlinear post processor. Finally, three experiments are described: 1) comparison of rectangular window versus exponential window AMDF, 2) comparison of the accuracy of the indicated pitch versus window length, and 3) effects of inverse filtering on the AMDF.