Tree-based modeling of intonation

Applying data mining techniques to corpus based prosodic modeling

“…The type of intonation units generally considered in studies of intonation include: sentences, intonation groups, stress groups or syllables. Dierent proposals can be found in the literature when one or several of these units are to be selected as the basic reference unit on which the matching between the text of the message and the F0 contours should be laid: accentual phrase (Sakurai et al, 2003), accent groups (Santen and Möebius, 2000), words (Veronis et al, 1998), syllables (Taylor, 2000;Lee and Oh, 2001;d'Alessandro and Mertens, 1995) and Interperceptual Centre Group (GIPC) (Holm, 2003). Moreover, in superposition models the use of more than one unit is proposed (Fujisaki and Hirose, 1984;Sakai, 2005).…”

“…Some of them are position, number of syllables, number of words, conjugation (Sakurai et al, 2003), part of speech (Sakurai et al, 2003;Veronis et al, 1998;Holm, 2003), number of syllables from previous stress or accented syllable (Taylor, 2000)and structure of the sentence (Lee and Oh, 2001;Holm, 2003). Furthermore, each of these prosodic features can have a dierent cardinality depending on the proposal.…”

“…As for the function-form correspondence, several proposals can be found in the state of the art on how to obtain the right relation between the acoustic parameters (representing the F0 contours) and the prosodic features: stochastic models (Veronis et al, 1998), neural networks (Holm, 2003;Sakurai et al, 2003), linear regression (Sproat and Olive, 1995) and decision and regression trees (Lee and Oh, 2001;Taylor, 2000;Eide et al, 2003). In all these methodologies, two main limitations arise: lack of robustness to cope with data scarcity training conditions and limited capabilities to provide experimentally contrastable prosodic information.…”