Speech Prosody 2002

Aix-en-Provence, France
April 11-13, 2002

A Method for Automatic Extraction of Fujisaki-Model Parameters

Pierluigi Salvo Rossi (1,3) , Francesco Palmieri (2), Francesco Cutugno (3)

(1) Facoltà di Ingegneria, Università degli Studi di Napoli "Federico II", Italy
(2) Dipartimento di Ingegneria dell’Informazione, Seconda Università di Napoli, Italy
(3) C.I.R.A.S.S., Università degli Studi di Napoli "Federico II", Italy

The utility of a model describing pitch profiles in speech signals is of fundamental importance in many application areas and especially in natural-sounding text-to-speech system. Fujisaki-model [1] has shown considerable accuracy on many languages, despite its simplicity. The inverse problem, i.e. the extraction of the input parameters which generated an observed pitch contour, that could be of great interest in the field of automatic extraction of prosodic parameters from a given speech signal, is a much harder task. This paper suggests a method for input parameters estimation based on two steps: an initial guessing algorithm based on relative extremes, and a refinement procedure based on a gradient optimization algorithm. Preliminary results of analysis/synthesis of pitch contours show excellent performance of the proposed method.

Full Paper

Bibliographic reference.  , Pierluigi Salvo Rossi (1,3) / Palmieri, Francesco / Cutugno, Francesco (2002): "A method for automatic extraction of fujisaki-model parameters", In SP-2002, 615-618.