PERIOD ESTIMATION USING MINIMUM ENTROPY DECONVOLUTION (MED)

In this paper, the Minimum Entropy Deconvolution (MED) method, develop ed for seismic signals, is used to address the problem of period estim ation in periodic and quasi-periodic sequences. First, a very intuitiv e explanation of the MED method is given, using a geometric approach. Then, periodic sequences, which can be viewed as the result of a convo lutional process similar to that of seismic signals, are postulated as candidates to be processed by the method. Tests performed on a set of signals - including natural and artificial, periodic and quasiperiodi c - produce filter outputs showing the desired comb-type shape. These output markers are successfully used in individual period determinatio n. The MED algorithm operating conditions that lead to useful results are described. In order to introduce the application of the MED method to the problem of pitch determination, voiced speech sequences are in cluded among the examples shown. A complete test set of the Spanish vo wels is used to evaluate the performance of the method as a pitch esti mator. Results obtained and their qualitative validation reveal its ro bustness, even for individual period measurement.