A Fractal Approach to the Features of Speech Consonants

Abstract

Fractal dimension (D) quantifies the roughness of a temporal signal and estimates its degree of freedom, allowing a good approach for its fluctuations. This present study of consonants follows the D-assessment of vowels which was presented at the Copenhagen Collegium Meeting. Using a 16 kHz time sampling the D-values of consonants were studied in the consonant-vowel context of the French language. Each consonant was pronounced four times by six males and six females. For D-measurement of long consonants the same method was used, i.e. the dyadic box-counting method and its 10 points of D-measurement (10pD) as that used for vowels. In the aim to approach infinitely small time scales, and to appreciate at least the tendency of the 10-point set, i.e. the D value to which tends this set, the slope of the three last points (3pD) was also calculated. For the plosion part of plosive consonants, a semi-continuous box-counting method devoted to the D-measurement of a short, single-dimension temporal signal was designed. This study consistently demonstrates that (i) there is a significant difference between males and females, as far as voiced and non-plosive consonants are concerned; (ii) plosive consonants are not fractal; (iii) among long consonants, D-value of fricatives are significantly different (p<0.01), as far as 3pD measurement is considered; and (iv) in the case of nasal consonants [m] and [n], this categorization is efficient for both 3pD and 10pD measurements (p<0.05). There results will be commented on and discussed with the aim of clinical use.

• Consonant Fractal Signal Processing Speech