ISCA Archive ICSLP 1992
ISCA Archive ICSLP 1992

An interactive system for automated pronunciation improvement

Jean-Paul Lefevre, Mervyn A. Jack, Claudio Maggio, Mario Refice, Fabio Gabrieli, Michelina Saving, Luigi Santangelo

In this presentation, we will introduce a project whose final objective is the development of a workstation designed to improve the pronunciation of a given language by non-native speakers. The SPELL (Interactive System for Spoken European Language Training) workstation is intended to be a self-instructional device aimed at intermediate ability foreign language learners. Three languages are involved including English, French and Italian. This study is based primarily on the analysis of the speech characteristics of non-native speakers and on the development of tools to be used in the automated assessment and improvement of non-native language pronunciation. The output of the first phase of the project will be an initial demonstrator system able to process the speech of non-native speakers in order to identify and correct pronunciation errors. This will involve the analysis of the phonetic characteristics of the speaker's mother tongue, the identification and measurement of his/her pronunciation errors, as well as the use of aids helping him/her to understand his/her errors and showing how he/she can correctly pronounce words and short phrases using both audio and visual feedbacks. In a second phase, refinements will be introduced in order to achieve a marketable product. The technical objectives of the project are to develop methods for analyzing the characteristics of speech produced by non-native speakers (both at macro and micro levels), to develop metrics for identifying differences between a non-native speaker's pronunciation and a model offered by the system, and to provide user friendly feedback which will help to improve pronunciation. The main technical innovation behind SPELL is the application of well founded phonetic and phonological principles for teaching selected aspects of the pronunciation of the language under consideration. This paper begins with a general description of the framework of the SPELL system. This general discussion will set out some basic assumptions about the system and the phonetic issues which it addresses. Then, some technical issues about the user interface software, which takes advantage of an object oriented approach in a powerful windowing environment, are also discussed.