A likelihood ratio scoring technique for speaker verification is described. The likelihood score for the speaker whose identity is claimed is compared with the scores of a "cohort" of other speakers assigned to that speaker. The likelihood ratio is used as a "normalized" verification score. This normalization technique can be viewed as providing a dynamic threshold which compensates for some kinds of trial-to-trial variations. In particular, it is shown that the use of cohort normalized scores compensates for the degradation obtained by comparing verification utterances recorded using an electret microphone with models constructed from training utterances recorded with a carbon button microphone. Cross-microphone verification equal-error rate drops from 22% using unnormalized scores to 4.8% using cohort normalized scores.