So my work involves looking at a bunch of waveforms in the context of classifying events. I often am looking for new ways to represent my waveforms, and in my searching, I came across audio embeddings (briefly mentioned at the end of this video ~ 28:49). The method is to take one cycle snapshots of your waves and treat them like words, with the surrounding snapshots acting as the context. But in order to do this, don't we need to discretize the signals? Or can we do this with a continuous set of values on either side of the snapshot?
I would be very interested in using this sort of representation in my work, but I'm not sure how to pretreat my waveforms prior to implementing the skip-gram model.