ESSV Konferenz Elektronische Sprachsignalverarbeitung

Title: A Pronunciation Tutoring System for Basque - First Development Steps

Authors: Igor Odriozola, Oliver Jokisch, Inma Hernáez, Rüdiger Hoffmann

Abstract:

In this paper, we introduce the first steps of the integration of Basque language in the AzAR system, the pronunciation tutoring software developed for German and Slavonic languages. At the time of designing the curriculum for Basque, we noticed that developing prosodic (suprasegmental) aspects is as necessary as teaching pronunciation (segmental) aspects. Therefore, the design of the system includes both parts. On one hand, the initial steps of the development of a CAPT (computer-assisted pronunciation teaching) system are introduced. It relies on a standard automatic speech recognition (ASR) system based on hidden Markov models (HMMs), which uses GOP scores (Goodness of Pronunciation) as confidence scores. The process of calculation the decision thresholds is explained for languages like Basque that do not have a specifically designed database for verification. Some preliminary results and conclusions are explained. On the other hand, a new module for AzAR is introduced, which aims to automatically assess the prosody of students. The module consists of computing the RMSE between the f0 curves of the student’s speech and the reference voice’s speech, after aligning and subtracting the mean. First results show that the global distance scores thus obtained are smaller among Basque speakers than comparing speakers of different nationalities with the reference voice. We conclude that, although more experiments and results are needed, it can be useful to receive an automatic feedback from the system. Finally, some conclusions and reflections about future works are introduced.


Year: 2012
In session: Sprachtechnologie und Anwendungen
Pages: 323 to 330