The production of fricatives is not yet fully understood because the mechanism is particularly complex. Studies of Portuguese fricatives have been very limited, so in this thesis a novel methodology of corpus design, and temporal and spectral analysis techniques were developed to enhance our description of the acoustic properties, and to increase our understanding of the production of fricatives. The data presented in this thesis could be used to improve the naturalness of synthetic speech.
Corpora were devised that included the fricatives /f, v, s, z, ʃ, ʒ/ in the following contexts: sustained, repeated nonsense words of the form /pV1CV2/, Portuguese words containing fricatives in frame sentences, and the same set of words in sentences. Four subjects (two male, two female) were recorded saying the corpora, using a microphone in the acoustic far-field and a laryngograph. Temporal analysis of the fricatives revealed a large number of devoiced examples. Analysis of variance showed that devoicing was significantly more likely for word-final fricatives and posterior place of articulation.
In addition to the fricatives listed above, we also noticed other fricatives occurring as allophones of /ʀ, ɾ/ in 100 words out of 365. Durations of the fricative segments were comparable to /ʀ, ɾ/ and thus shorter on average than fricatives /f, v, s, z, ʃ, ʒ/. Some of the speech segments were continuous “noisy signals” very similar to those of fricatives. The spectral peak frequencies of the fricatives occurring in place of /ʀ/ were compared to the other fricatives, which indicated a place of articulation further back than /ʃ, ʒ/, and compared to velar and uvular fricative results previously reported for other languages. These comparisons indicated that the uvular fricatives [χ, ʁ] and the voiceless tapped alveolar [ɾ̥] were given the phonological role of /ʀ/ and /ɾ/ respectively, though these fricatives have not previously been reported as phones of standard European Portuguese.
The fricative spectra were parameterised in terms of our knowledge of the underlying aeroacoustics. The parameters spectral slope, frequency of maximum amplitude, and dynamic amplitude were developed to characterise fricative spectra. The parameters behaved as predicted for changes in effort level, voicing, and location within the fricative. Some combinations were also useful for separating the fricatives by place or by sibilance.
A preliminary cross-language study of Portuguese and English fricatives produced by two bilingual siblings is also presented. Although results for Portuguese and English fricatives seem to be very similar this maybe due to the use by bilinguals of different production strategies from monolinguals which attenuate cross-language acoustical contrasts. The English corpus developed for the bilingual subjects could be used to study monolingual English speakers.
Jesus2001PhD.zip (PostScript file, compressed with Winzip 8.0)
Jesus2001.pdf (Acrobat 6.0 file)