Speaker Plínio A. Barbosa

Plínio A. Barbosa
Associate Professor, University of Campinas, Brazil

Plínio A. Barbosa obtained his PhD degree in 1994 from the Institut National Polytechnique de Grenoble, Grenoble, France. He is currently Associate Professor of the Dept. of Linguistics of the Instituto de Estudos da Linguagem at the State University of Campinas (Brazil), and responsible for the Speech Prosody Studies Group, a team of researchers and students devoted to the analysis and modeling of speech prosody. He has published more than 140 papers in pier-reviewed journals and congress proceedings and is the author of three books: Incursões em torno do ritmo da fala (2006), Manual de Fonética Acústica Experimental with Sandra Madureira (2015) and Prosódia para o Ensino Superior (2019). His interests are directed to Experimental Phonetics and Prosody. He is the editor of the Journal of Speech Sciences, a member of the editorial board of Phonetica, the International Journal of Speech Technology, and of six Brazilian journals. He is current member the IPA, ISCA, Abralin and IEEE Associated and a board member of the SproSIG.
- Tuesday 03 December 2019
  13:45 - 14:45 KEYNOTE: Stylistic and cross-linguistic differences in the prosodic organization of breathing, stressing, and pausing
  
  Speech breathing is very different from tidal breathing both with respect to its overall amplitude and the duration of inspiration and expiration phases. In terms of these two phases, speech breathing is strongly asymmetrical, with the expiration phase easily exceeding 2 to 4 seconds. This extended expiration time window is the foundation on which pauses, syllables and stress groups are built. In the first part of this lecture, experimental data from read and spontaneous speech in different speaking styles is presented to show how duration patterns of pauses, syllables and stress groups change cross-stylistically and how breathing patterns are adjusted by speakers to meet these stylistic changes and to support the realization of prominences and phrase boundaries in each style. For example, recent contrastive analyses of breathing patterns in read and narrated speech in Brazilian Portuguese revealed style-specific changes in inspiration amplitude and duration as well as in breath group duration and, moreover, gender-specific differences within these stylistic changes. Reading is also characterized by a higher frequency of inspiration, whereas narration shows a higher amplitude of inspiration. In the second part of this lecture, it is demonstrated that the relatively consistent and style-specific pause and prominence patterns allowed developing an algorithm for the automatic detection of terminal boundaries (of utterances in speech-act terms) in read and narrated speech. The algorithm was successfully tested for Brazilian Portuguese (BP), European Portuguese (EP), French and German. The audio file alone is sufficient for the algorithm to work, i.e. it does not require any previous transcription or linguistic-analysis efforts. The algorithm operates in two stages: first, it detects vowel onsets (VO) and then, it normalizes VO-to-VO interval durations in order to obtain smoothed z-scores; z-score peaks that exceed the threshold value of 2.5 are considered terminal boundaries. Proportion of hits for reading (circa 70 %) are better than for narration (circa 60 %) for all languages, with performance levels generally being higher for EP and French.