Testing voice-based biometrics authentication platforms for Romanian utterances through infrequent consonant clusters

Investigating the occurrence rate of two-letters consonant clusters in the Romanian lexicon, we rank the clusters by their frequency and determine those with minimal occurrence rates. Our table of infrequent consonant clusters can be used to construct Romanian utterances for voice-based biometrics authentication platforms. The main message of our results is that one should go beyond the somewhat standard distribution of phones when describing the phonetical balance of the test phrases/utterances used in calibration of these platforms. To this end, we propose the distribution of clusters as the closest complementary statistical indicator. Our results are relevant for a wide range of voice-based cyber-security solutions.