Statistical Description of Phonetic Balance for Greek Utterances

To assess the accuracy of a multi-modal voice and face biometrics solution developed within the H2020 SpeechXRays project, we investigate by computational means some statistical indicators of the Greek language that capture the properties of the language beyond the standard distributions of phonemes and words and word clusters. We extend the usual phonetic balance through the inclusion of the distribution of two-letter consonant clusters and a function that measures the vowel-consonant and consonant-vowel transitions and show how they can be used to construct test utterances that are statistically representative to the Greek language.