Building a pronunciation dictionary for the Kabyle language

Abstract

This paper presents the methods used and results obtained for the creation of a Festival-compatible pronunciation dictionary of above 10k words for the kabyle language. Kabyle is a berber dialect spoken in Northern Algeria. This dictionary will be useful in the design of text-to-speech and automatic speech recognition systems for the kabyle language. It was built using a bootstrapping method in which we incrementally build rules to predict word pronunciations while correcting wrong predictions. We thus obtain a large pronunciation dictionary as well as a set of rules to predict pronunciations for unknown words. The rules are embedded in Classification and Regression Trees and achieve 91,62% of correct prediction rate for entire words and 97,85% for phonemes.