This presentation will give a very brief summary of keyword spotting (KWS) and then will re-present the iterative Viterbi algorithm (IVA) for this task along with some further improvements in accuracy. Traditional keyword spotting algorithms use a garbage models to fill in the space before and after the keyword. The main problem with this is that the garbage models are very difficult to estimate. The IVA does not require the use of garbage models, instead it adapts a garbage parameter to find a segment in the utterance which maximises the average observed posterior. (This is the accumulated posterior divided by the length of the segment). Results obtained with this methods were using 117 independently chosen words using the BREF test set. The results suggest that the IVA algorithm can find more keywords than the output from an automatic speech recognition system, but with a few more false alarms.