Abstract

Motivated by a desire to assess speaking and reading skills and perform unsupervised tutoring of non-native speakers in a foreign language, robust evaluation of speech variability and pronunciation quality must incorporate perceptually meaningful information from many domains of speech analysis–spectral and prosodic, segmental and suprasegmental, and so on. In this paper we present three techniques for pronunciation evaluation on multiple time scales, as well as details of two example language-learning applications currently being implemented with these methods.

Date

2006

Authors

Joseph Tepperman, Jorge Silva, Abhinav Sethy, Shrikanth Narayanan

Journal

Proc. International Conference on Intelligent Systems And Computing: Theory And Applications ISYC06, Ayia Napa, Cyprus