de-emphasis (↔ preemphasis)

I feel like why not when we have preemphasis. A use case would be when people use librosa as a preprocessor of a system (speech analysis/enhancement for example) where they’d need to recover the preemphasis-ed input signals. Any thoughts?

[keunwoochoi]

👍 thanks for the update and the answers!

[bmcfee]

Yeah, this seems doable.

The preemph filter uses b=[1, -coef], a = [1], which gives zeros, poles, and gain z=[coef], p=[], k=1. The inverse filter would exchange poles and zeros, putting a pole at coef and producing coefficients b'=a, a'=b. The inverse filter then has a pole at 0 < coef < 1 (for nontrivial filters) which is stable.

I tested this out and it seems to work well enough for default settings, at least after the filter warms up for a bit. The hiccup here is getting the initial conditions right.