Topic: Recorded buzz - any way to remove?

[Spiggy Topes]

I'm new to all this, sorry, but have a whole series of problem recordings. They're voice recordings of interviews by phone, done using a voice recorder and a phone on speaker phone. Not until it came time to transcribe them was it apparent that many, but not all, included a background buzz that made the voices difficult or impossible to hear. I've dealt with that for future recordings by not using a wireless phone, but don't feel I can ask the interviewees to repeat unless there's absolutely no way that I can recover the buzzy recordings.

I don't have any experience with audio files, don't really want to spend money on software, so thought I'd try Audacity. Looking at the files at pretty much maximum zoom, the buzz is very obvious and very regular:



So I thought I'd try to find a section of silence (except for the buzz), copy to a separate track and invert it. Works beautifully. But the longest sample I have is seconds, and the whole piece is 20 - 30 minutes. I don't see an easy way in Audacity to cut and paste repeatedly without manually repositioning after each operation, and it's very, very easy to get just enough out of step that the two tracks get out of sync.

Is there any way that I can replicate the inverted section up a thousand-fold or so without repeated copying, repositioning and pasting? I know I can double up repeatedly, but it needs to be very, very precise and I keep thinking there must be an easier way. Tried the Compression and Hard Limit options, they just turn everything to mush. I don't need pristine, just intelligible.

Thanks, any help greatly appreciated!

[DigiGal]

You may want to look at a 15(?) day trial version of iZotope RX 3.  Not sure if the trial is crippled in any way though. If so there may be RX 2 copies available at a discount if you search since RX 3 is fairly new.


Another option is Sound Soap by the now defunct BIAS.

[Spiggy Topes]

Thanks, I may try that. I did think of a method to handle this, just after I'd finished posting, and it's not perfect but it works. I recorded 45 minutes of nothing but buzz, inverted that, matched the peaks from original to buzz track and played them that way. The buzz drifts enough that it's not 100%, but with millisecond tweaks to one track or the other I can keep them in sufficiently close sync that all that's left is a hint of the buzz and a far less intrusive high-pitched sound like frying eggs. I could try trimming that off based solely on frequency, I guess, but for purposes of transcription, matching up the inverted buzz seems to do the trick. 

[runonce]

Use the Audacity Noise Removal Tool....

Find a section of just the noise - use that as your sample, then select the whole track and apply the effect...

Should work well...

[DigiGal]

If you are all set great.

Just had this pop up as a suggested purchase in Amazon. Wasn't aware of it, another less expensive Izotope product ($30)...

http://www.amazon.com/eMedia-IZ11095-Izotope-Speech-Cleaner/dp/B002TXIB7U/ref=pd_rhf_ee_s_cp_1_121F?ie=UTF8&refRID=0FCDMPFK76XE3XAN3FVY

[hoppedup]

Use the Audacity Noise Removal Tool....

Find a section of just the noise - use that as your sample, then select the whole track and apply the effect...

Should work well...

x2

I just had a buzz in a PA system that I could not isolate through the spectrogram view. Or multiple attempts at notch filtering around what I thought it may be. So I selected a few seconds of it by itself between songs with the Noise Removal tool in Audacity and it did a pretty good job cleaning it up.

[Spiggy Topes]

I tried the noise removal tool in Audacity first, but I think it must have overlapped a lot of the frequencies in voice, as it either wiped everything or left a ... hard to describe, a kind of underwater gargle. Also tried the Compression, Hard Limit and Vocal Remover options, with no greater success. The inversion option seems to be the way to go. I haven't tweaked it much yet, but here's a sample result:

The original file with the inverted buzz in a second track:


and the result saved as a .wav file:


I can get a little closer by use of the Envelope tool, but there's still a degree of drift from one buzz to the other, so I still need to listen through and snip out a couple of milliseconds from one track or the other when the two files get out of sync, but the result is a file that was totally drowned out in buzz becomes understandable once more.

Thanks for the suggestions all!

Still no idea what causes it in the first place, of course...

[kirk97132]

Iztope is hands down the best noise removal tool.

[2manyrocks]

For some reason, some microphones pick up a buzz from telephones unless they are designed for telephone recording.  I don't know why, but it often happens.

Another possible method to try is to download the trial version of Reaper (isn't crippled), and use ReaFir to build a noise profile to subtract from the recording.  It's sort of awkward to find this feature in ReaFir, but there are three or four suboptions of which one is the noise subtraction profile.  Just capture the buzz on a quiet section and then the program extracts it from the recording.  Sometimes it works really well. 

If all that fails, then maybe someone would be willing to take a crack at it with Izotope for you. 

[ScoobieKW]

Best option is to use a purpose built interface.

The Gentner Hybrid Coupler
http://www.ebay.com/itm/like/200754591632?lpid=82

Or from JK Audio
http://www.jkaudio.com/radio-applications.htm

The Gentner unit was the industry standard for years, the JK Audio products are newer and more versatile.

[DigiGal]

Best option is to use a purpose built interface.

The Gentner Hybrid Coupler
http://www.ebay.com/itm/like/200754591632?lpid=82

Or from JK Audio
http://www.jkaudio.com/radio-applications.htm

The Gentner unit was the industry standard for years, the JK Audio products are newer and more versatile.

x2 

FWIW, for the same price I'd go for a new JK Audio Auto Hybrid over the Gentner Hybrid Coupler.  I've used both and prefer JK's Audio Hybrid.  http://www.fullcompass.com/product/243557.html

However, here's a BIN deal $50 on a used Gentner Microtel that would do the job for you, connect the Aux Out to your recorders input.  http://www.ebay.com/itm/Gentner-Microtel-Broadcast-Telephone-Handset-Line-Interface-No-AC-Adapter-/171178369826?pt=US_Other_Pro_Audio&hash=item27db06a322

[ScoobieKW]

Nice, I hadn't seen that JK unit, our shop owns these, http://www.fullcompass.com/product/260884.html along with the older Gentners. Definitely would go with the DigGal option over the Gentners. On the other hand, the only Gentners that end up in our repair shop have been from physical wear on the RJ-11 jack. Cheap fix and rare.

[rigpimp]

Iztope is hands down the best noise removal tool.

QFT.   There is a special place in heaven for the engineers that created the iZotope RX series.

[Spiggy Topes]

... and there's a special place in hell for thread hijackers! 

Okay, lots of recommendations for hardware and software, but it's too late for hardware and the Audacity software doesn't handle anything quite like this, and I don't expect it's what's usually meant by "noise", so I'm loth to buy software what won't fix the problem.

I have a workable solution, up to a point. It's still very uncomfortable to transcribe from, but my technique at least makes the conversation intelligible. There may be better and easier ways to do this, that don't involve quite so much tweaking though, and if anyone would like to take a crack at a better solution, here's a 30-second chunk to play with: http://members.shaw.ca/ccfc/Sound_Sample/ - in either .flac or .mp3 format, take your pick. It's a real oddity, the only one with a *double* buzz, but the same technique does work, just not very well.

So, my approach has been to record a piece with the same set-up but no voice, to get just the buzz. I've then inverted that to produce the anti-buzz and matched that millisecond to millisecond to the interview file.  You'll find ready-inverted buzz files in the same place. I've used Audacity's envelope tool to match peak heights from one track to the other, in order for one to cancel out the other. In this particular case, with the double buzz, I've added a second envelope-adjusted anti-buzz, and adjusted that to match the peaks of the second buzz. It works - just - , leaving a general background hiss like frying eggs. It needs constant tweaking, though, as every 15 seconds or so the alignment goes astray and has to be corrected. In a 40-minute conversation, that's a LOT of tweaking.

Anyway, if you'd like to give it a try, you're more than welcome. And if you have a better way to do this, I'm all ears!

[ScoobieKW]

As the guy accused of thread hijacking, I must defend myself.

When discussing a recording with flaws, it's perfectly reasonable to point out a way to avoid the issue in future recordings. While you may not ever need to record phone conversations, someone else searching this board may.

I'm glad you were able to make your recordings usable.

[Spiggy Topes]

Not an accusation, sorry if it came across that way. The thread started to drift very early, as threads do. Just thought I'd try to drag it kicking and screaming back in line.

[Marshall7]

my technique at least makes the conversation intelligible.

That's a lot more than can be said for this sentence of yours:

so I'm loth to buy software what won't fix the problem

To call out a guy for thread hijacking when he only meant to be of help is pretty fucking weak.

[Spiggy Topes]

Hey, all I said was I don't want to spend money on the problem if there's a fix that can be achieved without; I have nothing against anyone's postings here, I just came looking for a little advice, got it, didn't work, tried my own solution and I think it's as good as it's going to get. Take a look at the samples I put up, see if your suggested software would do the trick. I'm no expert in this kind of thing, but I don't think this is what you would normally categorize as "noise"; Audacity seems like a pretty solid tool and if it can't come close, then are you surprised that I'm not interested in spending money on it? For me, it's just become an exercise in tweaking a workable solution as far as I can with the tools I understand. If you think there's a better way to solve the problem, take the sample files and go for it, find a better way and explain what you did. *Then* call my response pretty fucking weak. Jeez.

[vanark]

weak.

[Spiggy Topes]

Not "pretty fucking weak", just weak, huh? Nothing better to offer? One tongue in cheek remark, and it turns a conversation into a shit storm. Who the fuck invited you anyway? Your first posting in this thread and it's one word. Weak. Witted, maybe.

[runonce]

Are you sure you're using the Audacity Noise Removal tool properly...?

You have to find, and select a small segment of just the noise...

Run the NR tool - and click "Get Noise Profile"

And then "Select All" - and re-run the Noise Removal Tool - this time omitting the "Get Noise Profile" step...

[Spiggy Topes]

Thanks for the return to sanity. Yes, as far as I can tell that's what I'm doing. Select a short speechless segment, sample that using "Get Noise Profile", then reopen the Noise Removal dialog with the whole file selected and click OK. I've tried a range of settings within the dialog, from both extremes to intermediate values, and no matter what, it doesn't leave the voice alone. I don't know how noise removal usually works, but I suspect it's by identifying frequencies in the sample and either eliminating or reducing those frequencies. So for a North American 60-cycle hum, or a British 50-cyle hum, it's a very specific target and easy to isolate. But here, the buzz is overwhelmingly load, and spans what looks to my untutored eye to be the same spectrum as the voice component. I've attached an image of the spectra so you can see what I mean.

If this is not how noise reduction works - I know it's a simplistic view, but I think it's close to the truth - then there may be tools to erase a repetitive spike such as we have here. If you have Audiocity yourself, and ten minutes to spare, please grab the samples I posted previously and give it a try. I've managed to salvage a number of the interviews with my own technique, but there are around 40 in total, all of 30 - 40 minutes, and some of them are hell to listen to, never mind correct.