Regarding the sampling rate: I switched down to 16kHz from 48kHz, because I found that with my approach there wasn't much of a difference in the key-matching p

Regarding the sampling rate: I switched down to 16kHz from 48kHz, because I found that with my approach there wasn't much of a difference in the key-matching performance. At the same time, I gained computational performance. At some point, I was looking at the spectrum of the key sounds and I think that almost the entire signal is in the lower few kHz. It might be just my keyboard - not sure. I still have the tool, so I can easily check if there is some high-frequency portion that I am missing.