Robust Pitch Detection Algorithms for Estimation of Fundamental Frequency of Prolonged Vowels Phonations at Pathological Voices

This paper presents a design for two novel methods of speech fundamental frequency (f0) detection in pathological voices. The methods are tested on the database of vowelsustained phonation. One method is based on the detection of maxima and the other on signal filtration with band pass. Compared to other well known f0 detection methods, our algorithms are designed with respect to speech pathology detection. Moreover, the designed methods support detection of the other voice parameters, such as jitter, shimmer and harmonic-to-noise ratio (HNR). The comparison of our results with results generated by the help of the Praat algorithm reaches a high value of correspondence: for the maximum method, the accuracy is 88,4 % and for the pass band method it is 83,9 %. The detection leads to the creation of a self-automated method, which is robust in detecting f0.