Automatic and accurate pitch marking of speech signal using an expert system based on logical combinations of different algorithms outputs

An expert system comprising a new pitch marking algorithm based on the estimation of the ideal excitation signal, using energy equalization of harmonics of the fundamental frequency present in speech, and three other competent tools is devised and explained in this paper. This expert system uses simple logical combinations of these tools outputs. The behaviour of a human expert is taken into account in developing the post-processing that is necessary to complete each tool and to further improve the results of their combinations. It is noted that, in most cases, combining the results of the new tool and the Childers method, itself based on what goes on behind hand marking by a human expert, is satisfactory. However, accurate and complete pitch marking is best achieved with all four outputs at the expense of some higher processing time.