The Fujisaki model and Superimpositional Models

Fujisaki s intonation model [166] takes a quite different approach to the models previously discussed in that it aims for an accurate description of the FO contour which simulates the human production mechanism. Fujisaki s model was developed from the filter method first proposed by Ohman [335]. 

In the model, intonation is composed of two types of components, the phrase and the accent. The input to the model is in the form of impulses, used to produce phrase shapes, and step functions which produce accent shapes. 

This mechanism consists of two second order critically damped FIR filters (these are introduced fully in Section 10.4). One filter is used for the phrase component, the other for the accent component. The FO contour can be represented by equations 9.1, 9.2 and 9.3. 

Although the mathematics may look a little complicated, the model is in fact very simple. Each phrase is initiated with an impulse, which when passed through the filter, makes the FO contour rise to a local maximum value and then slowly decay. Successive phrases are added to the tails of the previous ones, thus creating the type of pattern seen in figure 9.10. The time constant, a, governs how quickly the phrase reaches its maximum value, and how quickly it falls off after this. 

Accents are initiated by using step functions. When these step functions are passed through the filter they produce the responses shown in figure 9.11. The accent time constant, P, is usually 

Accents are initiated by using step functions. When these step functions are passed through the filter they produce the responses shown in Figme 9.11. The accent time constant, (i, is usually much higher than a, which gives the filter a quicker response time. This means the shape produced from the accent component reaches its maximiun value and falls off much more quickly than the phrase component. Phrases are initiated 

---