Preface
This volume contains invited and contributed papers presented at the 9th International
Summer School “Neural Nets E.R. Caianiello” on Nonlinear Speech Processing: Algo-
rithms and Analysis, held in Vietri sul Mare, Salerno, Italy, during September 13–18,
2004.
The aim of this book is to provide primarily high-level tutorial coverage of the
fields related to nonlinear methods for speech processing and analysis, including new
approaches aimed at improving speech applications.
Fourteen surveys are offered by specialists in the field. Consequently, the volume
may be used as a reference book on nonlinear methods for speech processing and anal-
ysis. Also included are fifteen papers that present original contributions in the field and
complete the tutorials.
The volume is divided into five sections: Dealing with Nonlinearities in Speech Sig-
nal, Acoustic-to-Articulatory Modeling of Speech Phenomena, Data Driven and Speech
Processing Algorithms, Algorithms and Models Based on Speech Perception Mecha-
nisms, and Task-Oriented Speech Applications.
Dealing with Nonlinearities in Speech Signals is an introductory section where non-
linear aspects of the speech signal are introduced from three different points of view.
The section includes three papers. The first paper, authored by Anna Esposito and Maria
Marinaro, is an attempt to introduce the concept of nonlinearity revising several nonlin-
ear phenomena observed in the acoustics, the production and the perception of speech.
Also discussed is the engineering endeavor to model these phenomena.
The second paper, by Marcos Faundez-Zanuy, gives an overview of nonlinear pre-
dictive models, with special emphasis on neural nets, and discusses several well-known
nonlinear strategies, such as multistart random weights initialization, regularization,
early stop with validation, committees of neural nets, and neural net architectures.
The third paper, by Simon Haykin, faces the problem of processing nonlinear, non-
Gaussian, and nonstationary signals describing the mathematical implications derived
by these assumptions. The topic has important practical implications of its own, not
only in speech but also in the field of signal processing.
Acoustic-to-Articulatory Modeling of Speech Phenomena deals with problems re-
lated to the acoustic-phonetic theory in which basic speech sounds are characterized
according to both their articulatory features and the associated acoustic measurements.
Fundamental and innovative ideas in speech production are covered. This section con-
tains three papers. The first paper, auth ored by Eric Keller, discusses voice quality
within a large predictive and methodological framework. Voice quality phenomena
are reviewed at two levels: (1) at the level of independent variables, topic-, affective-,
attitude-, emotion-, gender-, articulation-, language-, sociolect-, gender-, age- and
speaker-related predictors; and (2) at the level of dependent variables, where the em-
pirical identification of voice quality parameters in the speech signal are summarized.
Specifically, Fant’s original and revised source-filter models are reviewed.
