How to remove reverb from a recorded voice?
Is there a difference between a monophonic audio signal and a stereophonic reverb related to this removal process?
One old technique for this is to copy the track or tracks, invert the phase on each and apply a high ratio compressor to the inverted track. When you mix the treated tracks with the originals, the parts that were left by the compressor will cancel out. See De-Verb for Free: Removing Reverb using Free Plugins for a fairly good explanation of this technique. Fair warning though, it can require a fair amount of tweaking time to get the sound right.
A stereophonic signal gives you a much better starting base since there is a simple correlation between the initial wave fronts and more information about the room transfer function.
There are ways of removing the less phase-consistent elements you get in the reverb. However, the better you do this kind of reverb-removal, the more artifacts you get in the process (those tend to be squawking, blubbery sounds known as "musical noise"). This is usually an acceptable trade-off in speech processing applications. For artistic purposes, this kind of musical noise (the accidental leftovers from removing the decorrelated parts of noise) tends to be quite distracting.
So one has to keep the adaptive filters at rather moderate effectiveness.
A monophonic signal gives the filtering a lot less to work with as the spatial correlation is no longer available as a criterion and only temporal correlation measures can be used for trying to guess signal from noise.
Natural reverbation will be hardest to remove, followed by analog reverbs (such as coil reverbs) followed by digital reverbs. For simple digital reverbs with known basic structure, dereverbation might work well, but only if the digital reverb has been applied after recording and there is no further playback/recording through room acoustics involved.
A quick-and-dirty method might be to use a noise gate.
Reverb is often much quiter tha the voice itself so you could apply a noise gate to the track and set the threshold to be as low as possible such that it captures only the voice.
If you're not familiar with a noise gate, it's a device (software or physical) which silences output if the signal is less than a set amount (the threshold).
So is someone says something and there is trailing reverb afterwards, you may be able to set to set the threshold such that their voice is loud enough to peep through, but the reverb isn't so it gets silenced.
It won't get rid of the reverb present when people are speaking, only the gaps between. That is : if someone says "Potato", the initial P might have a reverb over the rest of the word. This will remain - it's only when the word finishes that the noise gate would kick in and remove the reverb.
It depends a lot on the recording. Eg whispers might well get cancelled as well.
It can have the effect of making people sound like they have a cold, because some consonant sounds get muted too like 'S' and "th"- makes it sound like one's nose is blocked. Some noisegates also have a setting relating to the amount of time a sound must be below threshold in order to be silenced. As speech is fairly quick, it may be possible to experiment & get the right settings to do this successfully.
If the reverb is loud compared to mthe voice then this probably won't work too well as cuting the reverb will most likely have an undesirable effect on the speech.