Software for generation and analysis of vowel polygons

This paper presents developed software generating and analyzing vowel polygons. The speaker independent software was developed in MATLAB environment, uses both Mahalanobis distance between vowels based on mel-cepstral coefficients and also feed-forward neural network for vowel recognition in fluent speech in Czech language. Format frequencies were estimated using LPC spectra. In relevant formant planes, all possible polygons are created by formant positions extracted from found vowels. In the comparison with other formant processing programs, described software uses formant polygon as an integral unit not each individual formant separately for the analysis. Developed program can be useful in speaker identification, recognition of actual emotional state of speaker, investigation into some illnesses and other speech processing applications.