In this paper we describe a secondary processing algorithm designed to improve word spotting performance by reducing the sensitivity of a primary HMM word spotting system to false alarms while maintaining high recognition accuracy. The concept behind the algorithm is to rescore the putative events hypothesized by the primary word spotter by generating a "secondary" score, which is designed to discriminate between true keyword occurrences and false alarms, and then combining it with the original HMM score. The secondary processor makes use of variable duration speech segments produced by a deterministic acoustic segmentation algorithm. The secondary processing algorithm is evaluated on a keyword spotting task using the Road Rally Database. Performance is shown to improve significantly over that of the baseline word spotting system with the greatest improvements taking place at low false alarm rates.