Developing a corpus to verify the performance of a tone labelling algorithm

Abstract

The authors report on a study that involved the development of a corpus used to verify the performance of two tone labelling algorithms, with one algorithm being an improvement on the other. These algorithms were developed for speech synthesis purposes with the aim of improving the perceived naturalness as well as the intelligibility of the speech produced by the synthesizer. The corpus used to test the algorithms consisted of 45 Sesotho sentences specifically chosen to represent the contexts that are necessary for the application of the algorithms. Statistical methods were used to prove the reliability of their transcripts.