Product - Deconvolver - Q&A

In theory, it is possible to remove a reverb from an acoustical recording given the room's impulse response at the recording position is known.  In practice, this may not work well, because a real sound source (if it's not a speaker cabinet) is hardly ever in a static position while deconvolution requires that the performer stays "still like a rock" relative to the microphone and room's borders.  These small movements of the performer usually result in various "swishing" artifacts, making output of the deconvolution process unusable.

On the other hand, the farther away the performer from the mic and reflective surfaces is, the lesser the chance the aforementioned problem will manifest itself.

Sine sweeps are one of the best test signals you can use to capture impulses.  You can generate sweeps inside the Deconvolver itself.  You can also use any custom test signals stored in WAV format, generated by any other software.

You should probably insert some amount of leading and trailing silence when running the sweep through FX, so that the recorded file stays several seconds longer than the sweep itself.

Yes, this is normal: this means your recorded signal had a lot of leading silence.  You are safe to remove this leading silence.

Various "hardware" impulses technically work like reverbs, but they "color" the sound instead of adding reverb tail - hence, they should be used at 100% wet level.  Note that compressors cannot be modeled with usual WAV impulse files - only their coloration is applied, not the compression itself.

Unfortunately, the minimum-phase transform is quite an unstable technique, so the sonic problems you are hearing are to be expected.  In general, there is no much sense in using the MP transform since it mostly is an experimental technique.  In practice, when you record, for example, a speaker cabinet, it already has a minimum-phase impulse response, so there is no need to additionally force it to be minimum-phase.

Deconvolver's test tone generator wasn't designed that way - the sweep runs from minimal to maximal frequency to cover the specified period of time.  So, only the first and last frequencies cross zero strictly.

Moreover, there is no mathematical deconvolution requirement for the sweep to be "phase-locked" the way you have described.

This problem is known: the fact you get a trace of the original sine sweep in the deconvolved impulse file tells that the device (in its particular mode of operation) you are recording has some sort of non-linear behavior - it may have harmonic distortion or chorus/phaser like behavior happening.  This situation cannot be resolved by any means, because Deconvolver can only produce impulse responses of linear time-invariant (LTI) devices (e.g. non-ventilated rooms at moderate temperature).

It is not necessary to normalize the files before you pass them to Deconvolver.

This situation is similar to the one where you get sine sweeps in the resulting file.  The usual reason for such behavior is some non-linearity or discontinuity in the recorded test tone file being deconvolved, which is either introduced by the medium you are recording or due to an abrupt start or finish of the recorded test tone file.  Make sure you leave some leading and trailing silence.

The recorded file should be longer than the test tone (test signal) file, because otherwise there will be an empty result as deconvolver "subtracts" (for a lack of a better term) the test tone from the recorded file.  If the recorded file is shorter than the test tone, it means there is not enough information in the recorded file available.  The minimal recorded file's length is equal to test tone's length plus tail's length.  For example, if reverb is 5 seconds long, the recorded file should be longer than the test tone by at least 5 seconds.

Please be warned: if you are recording an acoustical reverb and are stopping the recording process prematurely, your effort may go to waste.  Stop the recording only when the recording level meters reliably settle on an ambient sound level, after the test tone finishes playing.  For example, when capturing a large hall, the reverb tail may be longer than 8 seconds.  Not recording reverb's tail will produce a wrong deconvolution result.

Usually, 12 seconds should be enough.  Longer signals are better suited for capturing larger spaces in noisy conditions.

Deconvolver is mostly immune to the sample rate setting: it does not have the sample rate as a factor.  At higher sample rates there may be some sort of non-linear components or "noise" beyond 20kHz frequencies that make the MP transform problematic.  It depends on the recorded sound material.

Not only those, Deconvolver supports all standard sample rates.  Moreover, Deconvolver is sample rate-agnostic at its base.  However, the sample rates of the test tone and recorded files should match.

The click at the end of the test-tone is not a problem.  You may also generate a tone with fade-outs if you find it suits you better.