better to have a short "clean" noise sample than a long one with inconsistent noise. I'm learning to use SoX, so it turns out that my workflow makes more sense to do noise reduction before normalization, because otherwise I would have to know the gain ahead of time and apply it to the noise sample too for the noise reduction to have any significant effect.Ībout a second should be fine. Is it the sort of system that looks for a commonality in the noise sample, such that it would be better to have a longer sample of "silence" even if it might have some momentary low-level background sounds, or is it better to have even a really short sample of just "pure" noise, if that's even possible? How long is long enough? 0.1 sec? 1.0 sec? 10 sec? if you have a bad noise sample, you get bad NR results with signal getting cut off or noise let through in a less than ideal fashion.Īny tips on this? I don't understand how it works enough to know what a good noise sample is. the noise gate uses this to determine which bands to open to let signal through. What's usually most critical is finding a good signal-free sample to profile the noise.
0 kommentar(er)
