Human and Machine Perception:
Emergence, Attention and Creativity

Pavia, September 14 - 17, 1998
HMP98 Home Page

AT THE EDGE OF CAOS

Roberto D'Autilia
Dipartimento di Fisica, Università di Roma "La Sapienza"
Piazzale Aldo Moro 2, I-00185 Roma, Italy
e-mail: dautilia@romagtc.roma1.infn.it

Dissipative systems feeded by an external source of energy can model complex biological behaviors. In particular we consider the self-oscillating systems driven by external controllers: dissipative systems with a zero frequency source of energy, whose oscillations are determined by the nature of the system so that the motion does not change for small variations of the system itself. Together with this feature of stability, these systems show also different kinds of motion, from the very regular to the caos.

These models can be used to understand sources of complex information, and in particular the speech. Different speech features, like, for example, the vowels sound, the prosodic patterns or the sibilant sounds, can be achieved driving the motion of the system in the suitable regime. The behavior, similar to the arising of different regimes in fluidodynamics, can be compared with the changing of the vocal cords strength and the buccal cavity configuration when we articulate the speech. In this framework it is easy to understand why we meet many difficulties if we study these phenomena by means of linear models (see for example the spectral representation in the Helmholtz theory).

The vowel sounds analysis shows how the brain controls the system driving it among different regimes, from the very simple and predictable "singing" to the caotic one corresponding to the sibilant sounds. It has also to be noticed that machines based on these ideas have been developed since many centuries to communicate emotions: one of the most important example is given by the bowed string.

The developed theory gives interesting results if we use it for the study of the prosody. The prosodic patterns are important to give intonation, communicate emotions etc., but how to integrate prosodic information into artificial speech systems is still an unsolved problem. We propose a possible solution in terms of self-oscillating systems.

URL: http://vision.unipv.it/hmp98
e-mail: hmp@ipvvis.unipv.it