An audio processing system makes use of a number of levels of compression or data reduction, thereby providing reduced storage requirements while maintaining a high accuracy of keyword detection in the original audio input.