One of the biggest differences was transients, and it makes complete sense. If I compressed the transients “channel” separate from the channel with no transients (input-transients), perhaps I could get a cleaner sound. This is all speculation and I apologize if this sounds stupid.
So what would be the best way to detect the transients using a multiband approach? I choose to use multiband because transients don’t necessarily occur on the entire frequency spectrum. My current method analyzes transients (uses a slow envelope and fast envelope and finds the difference plus a little extra) on a given band and generates a “score” for how much it thinks the given sample is part of a transient. I can then add all of the scores from the bands up and average them, giving a normalized score of 0 -> 1 that can be used as a mixing variable (input * score = separated transients).
This algorithm yields ok-ish results. I tested it on a drum loop with transients of varying volume, found some things of interest:
– Loud transients were detected pretty much perfectly
– Quiet transients had trouble being detected
Any suggestions on improving this?
Read more here: Source link