Prosodic hierarchy and spectral realization of vowels in French

The aim of this study is to relate spectral realization of vowels and prosodic hierarchy in continuous speech. The IRISA speech alignment system is used and formant values of oral vowels are automatically measured in a total of 500,000 segments from around 30 hours of journalistic broadcast speech in French. The link between the duration of vowels and their spectral realization (through their formant values) has been validated for a long time since Lindblom (1963), i.e. the longer the vowels the more hyperarticulated they are. Similarly, the relation between prosodic constituents of different levels (the prosodic hierarchy) and the duration of phonemes close to these boundaries has been demonstrated (for French, Fougeron 2001; Tabain 2003ab). In this study we show that the level of prosodic constituent influences the spectral realization of vowels. Although significant differences can't be established for all levels and phonemes, we observe a general prosodic hierarchy (from syllable to word, then accentual phrase and finally intonational phrase) based on spectral measurement results, showing that the higher a vowel is in the prosodic structure of French, the more hyperarticulated it is.