A pitch detection algorithm with hypothesis and test strategy by means of fast surface AMDF

This paper describes a pitch extractor algorithm(V/UV decision and fundamental frequency estimation). The basis of the method is a modified AMDF(Average Magnitude Difference Function) algorithm. The main modifications are: Segmentation of the speech signal, fast computations by area comparisons. This last algorithm is shown to be 5 or 6 times faster than another AMDF algorithm. A hypothesis and test strategy is used. Hypotheses are generated by the modified AMDF. Then the hypotheses are verified by simple filterings of the signal and dynamic programming scheme.