Prosodic evaluation of speech

Abstract

This work presents English prosodic evaluation advances: specifically intonation and accent as spoken by hispanics. A widely used algorithm Hirst (Momel) [D. J. Hirst and R. Espesser, Travaux de l’Institut de Phontique d’Aix 15, 71–85 (1993)] to determine the Fundamental Frequency contour (F0), based on quadratic regression and splines is reproduced, and an alternative method, with comparable results, based on signal processing is proposed. We are presently experimenting with Hidden Markov Modes for accent evaluation. Intonation is the melody of speech and helps in transmitting emotions and intentions. The feature most commonly associated with intonation is the Fundamental Frequency of the speech signal. The accent refers to the emphasis given to a particular syllable within a word. The features associated with accent are: duration, pitch, and intensity of vocalic nuclei. The Hirst (Momel) algorithm is divided into five steps: preprocessing of Fundamental Frequency, estimation, partition and reduction of target candidates, and interpolation of the curve. The alternative method consists of: Preprocessing of Fundamental Frequency; Detection of voiceless regions; Fill-up of voiceless regions; Digital filtering, downsampling and upsampling. (To be presented in Spanish.)