Impulse perturbation normalization typo

[tabasy]

As this is my first issue, I MUST thank anyone contributing to this great repository, so I do :).

Without any context, this normalization line seems wrong (first min should be np.mean).
impulse_norm = (impulse.samples - min(impulse.samples)) / (max(impulse.samples) - min(impulse.samples))

This totally changes the RIR signal and adds an overlapping echo to the output. Seems like nobody is using this augmentation :) or they didn't manually check the outputs.

An example RIR file from BUT reverb dataset is attached for anyone like myself who seeks a RIR example to check the augmentation.
rir_1sec.zip

[anteju]

@tabasy, thank you for reporting this issue.
I agree, subtraction of min appears to be incorrect.
We'll take a look at this and open a PR with a fix.

[anteju]

This issue has been fixed in PR #6505