How do I get rid of the feedback in a WAV file with Audacity?

October 30, 2006 2:48 PM

I have a WAV file of a lecture I'm trying to edit into a Podcast using Audacity. There's some feedback and ringing during some of the speaker's talk. How do I get rid of it (or at least minimize it)?

Admittedly, I'm a novice with Audacity and sound editing; this is podcast #8. I couldn't control the feedback because I wasn't running sound; I (or rather, a student assistant) plugged the recorder straight into the sound board. (It's a Marantz, so it's not exactly a cheap digital audio recorder.)

It's not as if the feedback/ringing dominates the talk, but there are 2-3 minute passages where the ringing is more pronounced and annoying. And it is recorded as a WAV file rather than as an MP3.

Any ideas as to how to get the sound out or at least minimize the impact of the feedback?

posted by dw to technology (13 comments total)

Besides using an EQ to notch down whatever frequency the ringing is at, there's not a ton you can do. Room tone, once it's on tape, is almost impossible to remove. If you do have any success minimizing the offending frequency, try and use it only where necessary - I haven't used Audacity since it was horrible, but hopefully it gives you the option to either bypass the EQ most of the time (when there's no ringing) or allows you to apply the EQ directly to the unruly areas of audio.

If you'd like, I can give a few suggestions towards having an idiot-proof setup in the future, as to avoid this problem, but for now, I'm going to guess that you're stuck with what you've got.
posted by god hates math at 3:08 PM on October 30, 2006

I haven't used Audacity, but in a Audition/Cool Edit Pro, you could use one of the following approaches:
- use noise reduction, sampling the section with the feedback. This could be useful if you've got a nice chunk of the whine by itself
- use a notch filter. Through trial and error, or by using a spectrographic display, find the frequency of the whine, and filter it out
- use the EQ method as mentioned above. It's not perfect, and it won't be as narrow as a notch filter, but it's quick and easy.

If you'd like, compress and send the WAV file over to me, and I'll try getting rid of it. GMail address in profile.
posted by herrdoktor at 3:21 PM on October 30, 2006

1: If it's recorded in stereo you could try separating the left and right channels into 2 mono files and see if the feedback is less dominant on one of the sides, and if so keep it as a mono recording.

2: Or / and try putting a graphic eq across it, or a parametric where you can set the q (or octave) to a narrow band, and keep sweeping around the frequencies, cutting and boosting until think you've isolated the frequency of the feedback, and set it to cut it enough so it tones it down without affecting the frequency of the speech you want to keep clear, if possible. Just play it by ear.

3: I personally would load the file into wavelab (or any other editor that supports direct x plugins), load the sonic foundry noise reduction plugin (I think cool edit pro used to have this functionality built in, maybe see if adobe audition still has it), find a section where you can hear *just* the feedback (whatever background noise there is won't matter too much, so long as there's none of the speech you actually want or anything else of a similar frequency playing), loop it and take a footprint of the feedback in sfnr, and then reduce it gradually so it's not affecting the speech but eliminating the feedback, maybe doing it over a couple of passes rather that a significant reduction in one pass.

There's a couple of things you could try off the top of my head, anyway.
posted by chrissyboy at 3:31 PM on October 30, 2006

A narrow EQ cut should probably do the trick decently. I don't know if you're going to be able to do much with Audacity, though. If you're really stuck, send it to me and I can try to do it in ProTools.
posted by ludwig_van at 3:36 PM on October 30, 2006

On Preview: Get him to do it :)

Or get a copy of Adobe Audition and use the built in noise reduction plugin.
posted by chrissyboy at 3:36 PM on October 30, 2006

I second the suggestion to try Audacity's noise removal effect. It can be an incredible tool, and though it doesn't exactly result in broadcast-quality audio, it's often a lot nicer to listen to.

Just select a short, 'pure' section of the feedback (without other noises), select the Noise Removal effect, click "Get Noise profile", then select the audio you want to fix, then select Noise Removal again and click "Remove Noise".

If that doesn't produce great results, fiddle the noise profile thingy...
posted by stokast at 3:41 PM on October 30, 2006

it doesn't exactly result in broadcast-quality audio, it's often a lot nicer to listen to

If you're hearing really noticable artefacts of the noise reduction processing (that fft flangy / doubling chirpy type sound) then that's where you want to be doing a few passes reducing a small amount at a time imo.
posted by chrissyboy at 3:50 PM on October 30, 2006

Turns out there's a Nyquist notch filter plugin (way down the page) that works mostly perfectly. Well, once I figured out the Hertz level by trial and error. And the sound is now a little scratchy. But the ring is only an issue for about 8 minutes of the 70 total.

Noise removal was a pain; I couldn't get a clean profile, and the flanging was pretty hideous. I use it a lot, but guess not this time.

And I'll see if I can get some funds to get Audition or ProTools. We're a poor college in a cash-strapped university, so the budget people loves them some open source.

Thanks for all the help.
posted by dw at 4:23 PM on October 30, 2006

Let me make a quick suggestion, re: purchasing.

Audition. As someone who has a Protools LE rig at home, and uses Audition every day at work, Audition is likely what you want. Or possibly Bias' Peak, if you happen to use a Mac. For the kind of single-file editing (with minimal post processing) that you'll be doing, Audition is perfect, and PT is overkill. PT can do almost anything you want it to, but you don't care about 98% of it. Add that to having to buy into Digi's hardware, and it's a deal-breaker, for me.

I'd never record an album in Audition, and I'd never work in radio with Protools.

And of course, it's all the better if you can get your hands on either/both of them ahead of time to help make the decision.
posted by god hates math at 4:47 PM on October 30, 2006

I'd never record an album in Audition, and I'd never work in radio with Protools.

That makes sense. I'm doing voice, not music with these podcasts, and the postprocessing is indeed minimal (truncating silence, minor editing, noise removal). The destination is computer speakers and iPod headphones, so I'm already thinking the sound will be tinny to begin with.

And hey, Audition is $150 with the educational discount.
posted by dw at 5:08 PM on October 30, 2006

I figured out the Hertz level

(pedant) Frequency. (/pedant)
posted by ludwig_van at 5:10 PM on October 30, 2006

Which is to say, you located the appropriate frequency. I don't think you'll ever hear anyone say "Hertz level."
posted by ludwig_van at 5:10 PM on October 30, 2006

Damnit Ludwig, I'm a web developer, not a sound engineer!

Really. I know as much about sound as I do about electricity. I've used it and know that I should shut my mouth when other people talk about it.
posted by dw at 5:26 PM on October 30, 2006