Developing intonation corpora for isiXhosa and isiZulu

Abstract

In order to bring the tools of statistical pattern recognition to bear on intonation modelling, we need tailor-made corpora in the languages of interest. We describe how two such corpora were developed (for isiZulu and isiXhosa, respectively). We also show how those corpora can be used without further interpretation to gain insight into matters such as overall pitch contours and gender differences, and discuss the additional steps that will be required to create truly generative models from these corpora.