Visualisation tools to analyse phonetic confusions for speech perception tests

Abstract

To enhance the intelligibility of synthetic voices, especially for the hard-of-hearing, these voices need to be evaluated in a phonetically systematic way. This enables identification of problematic sounds and confusions that come with them, thus allowing suitable adjustments to the voice. Firstly, modified forms of confusion matrices are introduced to examine and compare phonetic confusions. This is followed by the consonant confusion cloud plots, developed to visualise confusion data by focusing on individual consonants. These tools enable us to identify the distribution and relationship of confusion with the target phone in terms of manner, place and voicing in one glance.