View Full Version : OT: audio echo removal

16-06-2004, 02:41 PM
I have recorded some audio and it was recorded in a big hall witha crap mic setup, I can't recorded it again.

the audio i have is ok but has a big echo kind of thing, you can tell its a big empty hall.

is there any digital way i can remove some or all of the echo?

free software would be nice :-)

16-06-2004, 02:52 PM
I doubt thats going to be easy.

Adding echo is easy, as its just a delayed version of the audio.

Removal isn't as easy, as the frequency spectrum is the same as that of the audio, and software would have a job trying to recognise what was an echo and what was primary audio.

Some brick wall filtering could restrict the range of frequencies, but it may also degrade the quality of the primary audio.

I am sure they could do it in CSI (or FBI, CIA etc).

16-06-2004, 03:07 PM
I had a good laugh at csi the other night,they were able to zoom in on a photograph taken by a security camera of a reflection caught in someguys sunglasses,they then produged a perfectly detailed image of writing on a patch of some other guys jacket.

Excellent stuff.

Not to mention the 3d mock-ups they produce and manipulte on their comps within seconds.

16-06-2004, 03:40 PM
could i not subtract and echo version for what i have??

thinking in a math kind of way.......

i want just the X bit and i have

x+0.5x+0.25x+0.125x so that is a echo of x getting smaller each time it echos.

if i make a new track that is

if i had a echo function that takes x and adds the 0.5x and so on, i could get this


i then subtract the second one from the first and get just X

hope this makes sence....

16-06-2004, 03:56 PM
Well, that is ONE of the corrections, but 't' pops up in a few spots, and delta t for each little x. Oh darn, but good luck anyway.

Graham L
16-06-2004, 03:57 PM
Well, there's the varying time delays ... dispersion ... to be considered. If it was easy ... :D

It's always easier to add "noise" than to remove it. One of the major problems in air accident inquiries is extracting what was actually said on the flight deck from the cockpit recorder.

Most of the work on echo removal seems to be in telephony, where the bandwidth is limited, and there are obvious commercial applications.

Have a look at an academic (http://www.el.bqto.unexpo.edu.ve/~jaguero/docs/98_0222.pdf), and a commercial (http://www.nctclearspeech.com/cs-aec.pdf) approach.

16-06-2004, 05:09 PM
Even if it were easy to do it via your "0.5x+0.25x..." method, you would lose some of the original source.

Unless what you recorded is comprised of a single short tone, then any echo will be overlapping the original source, and you'd risk taking out some of that at the same time.