Reducing Recorded Echo

[David Ennis]

This question comes up a lot. And of course, the answer is that echo must be prevented, not treated.

However, there actually is a method for reducing recorded echo that was described in this forum by Bob Grant. I've tried it extensively and found that how well it works for me depends how much higher the amplitude of the main signal is than that of the echo. A small amount of echo can be almost eliminated. But I was unable to improve a large amount of echo at all without seriously degrading the main signal.

Also, low level intelligence in the signal tends to get lost-- part of the baby may get thrown out with the bath water. For example, consonant sounds at the beginning or end of some words from some speech may suffer. The more dynamics the main signal has, the harder it is to find the right settings. For instance, the nuances of dramatic dialog will probably not fare as well as an address by a public speaker.

In the interest of having a self-contained thread on this topic, I'll quote Bob here rather than link you to his post.

" ... Basically you make a copy of the track and invert it. Now you have 100% cancellation. Now apply compression to one of those tracks. Trick here is to get the knee of the compressor so that the compressor is just compressing the wanted part of the signal and not touching the lower volume echo / reverb. So now the echo will still cancell out but the compression will alter the wanted part of the track so it doesn't get cancelled. Adjust attack and decay times of the compressor to taste as well as the level of one of the tracks. Adding Eq after the two tracks are mixed as well will help.

You'll never get it 100% as good as it would have been minus the echo but you can make a very signifcant improvement."

I start (in Vegas) with 10 to 1 compression, 250 milliseconds of attack and 250 milliseconds of release and adjust the threshold ("knee") until the sound is optimal. Then I adjust the release downward until distortion outweighs the echo reduction benefit. I find that the process is much less sensitive to release time than to attack time. There's nothing magic about 10 to 1 compression--a wide range of ratios works. There is probably some correlation of optimal ratio to the type of signal and type and amount of echo, but I haven't tried to study that.