Speech rhythm measuring and modelling: pointing outmulti-layer and multi-parameter assessments

In this paper we summarise early contributions to rhythm measurement and assessment carried out in our laboratory in Turin (following criteria defined by Ramus et alii, 1999, and Grabe & Low, 2002). We briefly illustrate the need to reserve more attention to rhythm perception in order to include the apparently relevant role of f0 in speech rhythm evaluation. Preliminary testing showed that duration could be a good correlate, giving a physical estimate of rhythmic (maybe derived) properties (see Romano, 2009), but we believe that most direct correlates should take into account (distance, extension and shape of) peaks and general melodic profiles. Linguists’ intuitions about speech rhythm are mainly based on perceived or expected properties (often perhaps heavily influenced by the knowledge of phonological features; see e.g. Canepari, 2006, and Ghazali et al., 2002, for examples on how impressionistic evaluations could match experimental/instrumental research). Even though common people perhaps associate perceived rhythmic properties to stress occurrence conditions or relate it to an impressionistic evaluation of faster vs. slower (or rapid changes in) tempo, this paper aims to show that general intuitions could help to understand the basis of a well assessed dichotomy between two basic rhythm types. In this view, this contribution also aims at accounting for recent papers which address their interests in extending the assessment dimensions to other variables (as Lee & McAngus Todd, 2004, with intensity and rhythmograms) or suggest different ways to calculate metrics (as it is proposed by Mok & Dellwo, 2008, with DeltaS, or by Gibbon & Gut, 2001, with the rhythm ratio, both accounting for syllable durations). Furthermore, on the wake of Bertinetto & Bertini (2008 and forthcoming), we propose to integrate rhythm metrics into a rhythm model, possibly a multi-layer one, where durational properties could be associated with strong-weak measures or other stress/syllable properties at different levels (Gibbon & Gut, 2001, Barbosa, 2006) in order to merge high level (linguistic) information with measures of more than one parameter (namely pitch, duration and perhaps intensity).