Data Driven Intonation Models

To solve these problems, we can of course refine the models, but this can result in endless tinkering in which the models lose their original elegance and simpUcity. The basic problem is that, when we design a model, we wish it to be elegant and easy to interpret, but we always nm the risk that some intonational phenomena that we haven t taken into accoimt may appear and be outside the model s capabilities. We then face the choice of adding to the model, which often appears as a hack , or designing a completely new model. 

In the following sections, we give an overview of the techniques used, but, since many of these follow the synthesis techniqnes described in Chapters 15 and 16, we wait until then to give a Mly formal account. 

A common criticism of the Fujisaki model [285], [436] is that there are many contours which the model can not generate. These include gradual low rising contours which occur when a low accent occurs early in a phrase, and the effect of phrase final lowering where the contour falls past the normal baseline to a final baseline in declarative utterances. Suggestions that fix the problems 

---

Buhmann, J., Vereecken, H., Fackrell, J., Martens, J. P., and Coile, B. V. Data driven intonation modelling of 6 languages. In Proceedings of the International Conference on Spoken Language Processing 2000 (2000). 

While many of the AM models and deterministic acoustic models provide useful and adequate representations for intonation, the trend is clearly towards the data driven techniques described in Section 9.6. These have several advantages besides bypassing the thorny theoretical issues regarding the true nature of intonation, they have the ability to automatically analyse databases, and in doing are also inherently robust against any noise that can occur in the data, whether it be from errors in finding FO values or from other sources. 

In recent years, the notion of explicit models has been challenged by a number of data-driven techniques which learn intonation effects from data. 

---