Echo removal function?

[BillFrezza]

Is there an echo removal function somewhere in Audacity? It should be possible to build a digital filter that does that, maybe with a tunable time constant and damping parameter you could play around with to get it match the room conditions just right. I'd love to remove a slight but annoying echo from an audio interview I recorded in a boxy room and my hope was I could find such a function in the Effect menu, but no such luck.

[kozikowski]

No, but if you wanted to program one for us you would be the biggest hero on the planet. Echoes are one sound problem that can kill a production. See #1.

The Four Horsemen of Audio Recording (reliable, time-tested ways to kill your show)
-- 1. Echoes and room reverberation (Don't record the show in your mum's kitchen.)
-- 2. Overload and Clipping (Sound that's recorded too loud is permanently trashed.)
-- 3. Compression Damage (Never do production in MP3.)
-- 4. Background Sound (Don't leave the TV on in the next room.)


Koz

[BillFrezza]

Ha, in my younger days when I could both write code and do Fourier Transforms in my head I would have taken on such a challenge. But my engineering days are three decades in the rear view mirror and I can't even do simple arithmetic anymore without a calculator.

Seriously, though, one should be able to model a rectangular room with bare reflective walls and then deconvolve the echo. The trick in using such a function would be discovering, through trial and error, the handful of key parameters required to damp down the echo without introducing additional artifacts. I've got to believe it's possible, especially with non-causal digital filtering techniques. Surely, there's a signal processing engineer out there willing to attempt it?

[kozikowski]

There are expensive software packages with convolver tools and modules, but every time I ask a competent user how they handle room echoes, they "hear their mom calling" and have to go.

And don't call me Shirley.

Koz

[kozikowski]

I thought for a long time that you could do OK by creating an artificial echo, add it to the show and see if the overall volume went down. Then keep changing the size, polarity and delay of the signal until no further improvement is possible. Then make a new one and go through it all again. That has the advantage that you can "stop helping" any time you wanted with any number of echoes limited only by the time you had to wait.

Of course, that may be how Convolver works. I don't know.

Koz

[Gale Andrews]

If you can't spend money

• Deverb using noise removal:
  http://forum.audacityteam.org/viewtopic ... 64#p251864
  However Sensitivity is different now in Noise Reduction in 2.1.0 - I've had more success in 2.1.0 using a small section of "reverb only" as the Noise Profile.
• Deverb using centre isolation in rjh stereo toolkit:
  http://forum.audacityteam.org/viewtopic ... 89#p216689

Gale

[kozikowski]

If you can't spend money

And if you can?
Koz

[kozikowski]

I've had more success in 2.1.0 using a small section of "reverb only" as the Noise Profile.

Then something strange may be happening behind the curtain. Echoes or room reflections are the performer's own voice coming to the microphone multiple times by way of bouncing from the walls and ceiling. There are no more or fewer frequencies or tones.

Noise Removal worked by changing the sensitivity to produce an artificial difference between higher volume sound (voice) and lower volume sound (echo). It was never anything to write the papers about because the sensitivity slider wouldn't go far enough to be really effective.

So apparently, Noise Reduction needs the hand-clap at the beginning of the lecture to be useful. You need an impulse sound so you can identify the trailing echoes clearly. If you try to do it from speech, it will never be more than hit or miss.

The movie clapboard has magnets at the closure to prevent stick bounce. Both the film and the sound people are looking for one, single, clean, clear sound impulse to use for sync.

So all you need to do is pop up just as the performer starts to speak, yell "SOUND MARK" close the board and leave before they arrest you.

Piece of cake.

Koz

[Gale Andrews]

If you can't spend money

And if you can?
Koz

In theory, deconvolution if you have a good sample of the reverb, with as little of the original sound as possible:
http://forum.cakewalk.com/tm.aspx?m=173996

SPL De-verb (Windows VST and Mac VST/AU):
http://spl.info/index.php?id=463&L=1

Zynaptiq Unveil:
http://www.zynaptiq.com/unveil/

Plus if you can't spend, but you can compile on Linux:
http://www.hydrogenaud.io/forums/index. ... ntry214926.


Gale

[steve]

Plus if you can't spend, but you can compile on Linux:
http://www.hydrogenaud.io/forums/index. ... ntry214926.

Unfortunately that one is becoming very difficult to build due to the age of the libraries used.