Hello!
Anyone have an idea how to achieve this?
Either an inverse filter or spectral subtraction.
Regards
De-reverb based on Impulse Response
-
- KVRist
- 484 posts since 8 May, 2007
If you say that deconvolution is not the answer, then there are only two possibilities that I can think of:pluginnow wrote: Mon Dec 04, 2023 6:08 pm Yes, but we can only invert the minimum phase of the impulse response.
I believe that after it, the only option is to spectral subtract.
1) We are misunderstanding what you are trying to accomplish.
OR
2) You are misunderstanding something about how deconvolution and/or "de-reverb" works.
To review in case of #2:
Convolution can be accomplished by multiplication of complex values in the frequency domain.
Deconvolution can be accomplished by division of complex values in the frequency domain.
“De-reverb” does not involve “spectral subtraction,” instead complex spectral division.
If this doesn’t help, then exactly what are you trying to accomplish?
-
- KVRist
- 484 posts since 8 May, 2007
For that case I asked "What exactly you are trying to accomplish?" The subject line you wrote could be many different things.
Although it seems to me to be an unusual situation, I am assuming that you have a result of a convolution reverb and the IR that was used to create this result, and that you want to recover the original dry signal.
If that's correct, and if it were me, I would use a program I wrote years ago called "unreverb" that uses deconvolution and recover the dry signal. Normally I use it to recover the IR from the wet signal using the original dry signal, but it works either way.
I'm curious as to the this idea that de-reverb involves spectral subtraction comes from. Perhaps your definition of "spectral subtraction" is not what I think it is.
-
- KVRist
- 484 posts since 8 May, 2007
After thinking about it for a while, I now believe that my previous post is incomplete without a demo video.
The video demonstrates a quick-and-dirty deconvolution using rectangular windows. It shows an unreverbed, dry, electric-guitar DI signal, a room IR from MConvolutionEZ, the reverbed version of the dry signal using the room IR, then the de-reverbed or unreverbed signal, then a graph showing a typical region of the difference between the dry and the recovered signal.
The difference between the dry and the recovered signal is about -99 dB RMS, showing that deconvolution is indeed the inverse operation of convolution and that division of complex values in the frequency domain can indeed accomplish deconvolution pretty well if the IR used for reverb is available.
QED.
The video demonstrates a quick-and-dirty deconvolution using rectangular windows. It shows an unreverbed, dry, electric-guitar DI signal, a room IR from MConvolutionEZ, the reverbed version of the dry signal using the room IR, then the de-reverbed or unreverbed signal, then a graph showing a typical region of the difference between the dry and the recovered signal.
The difference between the dry and the recovered signal is about -99 dB RMS, showing that deconvolution is indeed the inverse operation of convolution and that division of complex values in the frequency domain can indeed accomplish deconvolution pretty well if the IR used for reverb is available.
QED.