Recognizing keyboard clicks and segmenting waveform audio data

I am trying to find an appropriate method for keyboard click sounds (hits) recognition in waveform audio data and segmenting those clicks into separate "windows" for use with machine learning algorithms as features (after FFT and calculating frequency distributions for each "click" window). Looking at recorded keyboard clicking samples using Audacity intuitively it seems that it might be feasible to simply scan over the range of samples and try to find maximum sum of the closest n samples and that might probably constitute a window with the peak in the beginning (max value).. well i'd probably try to take into account the nature of key click sound and perhaps look for a segment that's close to the typical distribution (pattern of typical key click segment). Of course this sounds like a very dummy approach so my question is are there some useful tested techniques/algorithms for this task of audio segmentation by peaks? I really hope i'm missing out..

I must mention that i only need to identify the peak key touch phase as i'm probably taking into account only some defined short period of time to be used as a window.