A large-scale forensic discrimination experiment is described which investigates how well same-speaker speech samples can be discriminated from different-speaker speech samples using acoustic parameters (F-pattern and duration) from Australian English vowels. A multivariate likelihood ratio is used, under both optimum and realistic conditions, as a discriminant function on the five tense and six lax vowels phonemes of 171 male speakers. In 171 target trials and 54,140 non-target trials, comparing samples with just one token per vowel each gave EERs of between 17% and 40%, which dropped to 10% (optimum) and 14% (realistic) when fused. It is also demonstrated that kernel density modeling outperforms normal, and that performance degrades under realistic conditions.