Over the last decade, various detection mechanisms for spoofed speech have been proposed. Thus far the development focus has been on detection accuracy, largely ignoring secondary goals such as computational complexity or storage effort. In this work, we use empirical mode decomposition to compute intrinsic mode functions which are then demodulated to obtain features consisting of short-time statistics of instantaneous amplitude and instantaneous frequency. These features are then used with a simple k-nearest neighbours classifier. We further show that voiced segments from short speech signals can be used in the feature extraction resulting in a spoofing detection competitive with top-performing systems while having up to 103x less computation.