There have been various studies in the past reporting the duration of fricatives, and of the VF and FV transitions, and investigating the influence of word- and sentence-position, stress and vowel context on fricative durations (Klatt 1971; Klatt 1974; Klatt 1975; LaRiviere et al. 1975; Umeda 1977; Manrique and Massone 1981; Behrens and Blumstein 1988; Docherty 1992; Shadle and Mair1996; Crystal and House 1997). The difference in duration between unvoiced and voiced fricatives, and the frequent devoicing of voiced fricatives, have also been reported in many studies (Docherty 1992; Shadle and Mair 1996; Pirello, et al. 1997; Smith 1997; Jongman, et al. 2000). In this poster, we present a novel methodology of temporal analysis and results including the durations of the fricatives, and of the VF and FV transitions and a study of devoicing, together with discussion of some possible causes of this phenomenon. The correlation between devoicing and duration was also investigated. The data presented in this paper could be used to improve the naturalness of synthetic speech.
There is some perceptual evidence (Cole and Cooper 1975; Scully 1979; Hogan and Rozsypal 1980) that the vowel to fricative duration ratio is used as a primary cue for voicing, so we've also extracted and analysed new acoustic data from the corpora described by Jesus (2001), namely the ratio of fricative duration to the preceding and following vowel duration. Results seem to confirm our initial predictions, i.e., that the medians of the ratios of unvoiced fricatives is greater than those of voiced fricatives, and that devoicing lowers the vowel to fricative duration ratios. This new acoustic evidence suggests that Portuguese ``maybe changing in the direction of using'' fricative to vowel duration ratio ``as the primary cue for the voicing contrast'' (Klatt 1976, p. 1219). Segmental durations of fricatives, vowels, and VF and FV transitions also provide information about the relative timing of the oral closing/opening and glottal opening/closing gestures (Klatt 1976, p. 1214), and could be used to model the durational behaviour of Portuguese in a speech synthesis by rule system (Klatt 1976; Umeda 1976).
Jesus2002ps.zip (PostScript file, compressed with Winzip 8.0)
Jesus2002pdf.zip (PDF file, compressed with Winzip 8.0)