The performance of automatic speech summarisation has been improved in previous experiments by using linguistic model adaptation. One of the problems encountered was the high computational cost and low efficiency of the development phase. In this paper we compare our original development approach of evaluating summaries produced by an exhaustive search over all parameters with a much faster development method using an expectation maximization algorithm that minimizes perplexity in order to find the optimal combination of linguistic models for the speech summarisation task. Perplexity proves to be sufficiently correlated to the objective evaluation metrics used in the summarisation literature that it can be used in this fashion. For a much reduced computational cost (approximately 500 times faster), final relative improvements are very similar to those previously obtained, ranging from 1.5% to 21.3% on all investigated metrics for summaries made from automatic speech recogniser transcriptions.