A Procedure for Formant Domain Extraction

Abstract

This paper describes a procedure for extracting the formant domains, in which there are the formants of a vowel, from a measured vowel spectrum by means of a digital computer, KDC-I, as a step preceding an accurate extraction of the formants. The formant domains are expected to be obtained as the frequency regions, where a spectral value is larger than a certain threshold value. A locally weighted mean of spectral values was adopted as the threshold value. The spectral data used for the test of this procedure were obtained from the connected vowel sounds of the words spoken by two male speakers. The formant domains may be said to be extracted fairly well for the tested utterances, although they are small in number.