An initial attempt to improve spoken term detection by learning optimal weights for different indexing features.