Boy, I wonder if I downloaded the same clip. You clearly have a MAJOR noise problem. Sounds like a light dimmer to me. I've had this before, and there is no good solution, even with iZotope RX, and certainly with the Sound Forge NR plugin.

The problem is that the noise consists of sharp noise spikes. As anyone who is familiar with Fourier analysis will tell you, impulse spikes contain harmonics all the way to infinity. Thus, the noise reduction feature is powerless, because you end up with harmonics only 60 Hz apart, even up at 4,000 kHz. There is no way to remove frequencies that are this close together when you get up to those frequencies. The hum module in iZotope RX does nothing at all.

I started a thread about this three years ago after video I taped at a rehearsal dinner which got completely trashed either by the light dimmers or by a nearby TV transmitter at the Sears Tower. Here's a link to that post which describes how I fixed it. My method is very labor intensive, but it worked extremely well. If you are patient, the result can almost be perfect.

Here's a link to that thread over in the Sound Forge forum:

Continuous Impulse Noise (Buzz) Reduction

I just sent a feature request to the iZotope people describing my technique and asking them to incorporate it into a future release. We'll see what sort of response I get.

John, you caught me in mid-edit. I turned it up and heard the buzz too (I had the TV on). Could be dimmers, but it also sounds like my cheapo mp3 recorder. Since the spikes are in a linear rather than a harmonic progression, and very uniform, I tend to think it is the recorder. Even my H4, which is highly touted, has a "thing" at around 700 hZ. One thing we agree on -- this type of noise is almost impossible to extricate.

If you look at the link to my old post and follow the technique I describe there, it is possible to almost completely eliminate the noise. I just tried it with the sample posted here and achieved almost 100% reduction of the noise, and with virtually no noticeable degradation to the original audio. The problem is that the noise print has to be perfectly in phase (or actually perfectly out of phase) with the original signal. I can achieve this for about 2-3 seconds at a time, and then I have to copy and re-sync the inverted noise print. If you loop the noise print, as described in the link from my previous post, it doesn't take long to achieve a noise null.

In my feature request to iZotope, I suggested that they first figure out a way to idealize the noise print so it is a stable noise print. Then they need to find a way to lock it perfectly to the original signal, probably using an algorithm similar to a phase lock loop. There is a dynamite noise reduction tool waiting there for the industrious computer engineer who love this sort of challenge. My bet is that the iZotope people will bite on this one. It really fits their product nicely, and I think a lot of people have this problem (I'm pretty sure this is a light dimmer because they generate a huge impulse spike every time the triac is triggered into conduction, something which happens twice each second (once on the positive side of the sine wave, and once on the negative). The period of the noise in this sample is approximately 0.008 seconds which is 120 Hz.

Here's a link to two very short (1 second) clips where you can hear the before (which I added 20dB gain), and the after (where I took the +20dB clip, inverted it, and then added the result to the original.

Before/After

But wait a second, you say, that doesn't prove anything. If I take a clip, invert and add it to itself, I'll always get perfect silence.

Yes, but there is one thing I forgot to mention: I actually added the inverted noise from an earlier portion of the clip. Since the noise is almost perfectly repeatable, subtracting the noise print from any portion of the audio file produces almost zero where there is nothing but noise, but leaves an almost pure signal where there is audio.

OK, I just spent another minute on the project. Here is another short before/after clip, this time with the person speaking. With a little more effort, I could get this to come out even better:

Before/After with speech

These links are good for seven days.

John, I am not even worthy to READ your ideas and analysis! You is special people Meyer!!!

And yes, plug it to Izo - and a good view on their biz plan. Nice one!!

Often I think programmers need to be stimulated with what the actual "user" ( and Sony here too!) needs and what would be good for them.

I was using IzoRX the other day on that "other" sample we got from a chum here. And I was thinking, "I can HEAR what I want to keep. Why wont IzoRX just grab THAT and throw the other stuff away?" - Yes I realise it ain't that easy, but the clarity of wishing is!

Grazie

iZotope RX continues to amaze me. I just found an ancient live performance of Joan Baez singing "Diamonds and Rust." The scraping of her fingers on the wound bass strings of her guitar was really distracting. This is exactly the thing iZotope uses in their demonstration of the spectral repair, except that this had far more complex audio, with singing and audience mixed in. I was able to almost completely eliminate the scraping sound without doing any damage (that I could hear) to the recording. Absolutely amazing stuff.

I hear the same noise, John. It definitely sounds like induced hum, as said earlier, likely by a nearby dimmer. If the shooter used an external mike, it was most likely with an unbalanced cable.

I ran the bit through Levelator and then in Vegas added a low-pass filter with the rolloff at 1KHz and it sounded acceptable. Since the speaker's voice was very bassy, I didn't lose much of it even with the 1KHx rolloff.

But if it had been recorded closer to -12dBM then the same noise would have been practically unnoticed.

I can get pretty good results.
Use SF's Click and Crackle Removal FX.
Starting point:
Sensitivity = 18
Click Shape = 7
Max Click size = 2.0
Noise Level = High

A bit of low pass after that should clean it up a lot or Noise Reduction. The trick with NR is sometimes it pays to edit the noise envelope to avoid the wanted sounds being harmed too much.

The idea of adding the inverse of the noise goes back to the early days of vaccum tubes when the AC directly heated cathodes created a lot of hum. That was a pretty easy trick compared to cancelling external noises.
Problem with these kinds of noises is they drift in frequency , harmonics and amplitude. No doubt some smarts could take care of the frequency drift, the rest seems to be the stumbling block.

Bob.

Using the pop and crackle tool is probably the best automatic way to make some headway. Given that this is speech, you probably won't get the horrible "gargling" artifacts that you can get when you turn up the crackle settings in a music recording.

Here's the result of 23dB volume increase, followed by click/crackle removal in Sound Forge, followed by a light pass with the noise reduction module. I changed it to MP3 format to make it fast to download. Link good for seven days.

Quieter version, MP3 Format

**I was able to almost completely eliminate the scraping sound without doing any damage (that I could hear) to the recording.**

OMG John, that was her trademark. A Joan Baez recording without the string scratches is like a trip to Oz without the poppies. A call to millions of aged hippies -- unite!

BTW, I also tried my own version of the inverted feedback approach last night and got some improvement, too busy with another project now to carry it further.

The inverted feedback definitely produces better results, but it takes hours instead of minutes. The problem with the the pop/crackle is that it nails the one big impulse but misses the rest of the noise. As a result, when I did that pass with the noise reduction (after I did the pop/crackle pass), when I captured the noise print and looked at the spectral envelope, there were still those thousands of spikes, 60Hz apart, going all the way up to beyond 10,000 Hz. There is mathematically no way to eliminate those in the frequency spectrum because you need a filter with an almost infinite "Q" and this introduces all sorts of artifacts. So, there is still a residual buzz. By contrast, the inverted noiseprint works in the time domain (rather than the frequency domain) and therefore, to the extent you can line up the inverted version, and to the extent it is constant, it completely removes the noise, all the way up to the upper audio frequencies.

I'm hoping to engage the iZotope development team on this. They are a bunch of M.I.T. nerds and have been really responsive. They remind me a lot of what the software industry used to be like, back in the "go-go" software era of the 1980s.

Wow............ You all are amazing.

I'm not at my editing computer right now, but I'll give the primary options mentioned a whirl this evening on the full audio files and see what I get.

The actual taping session was about a year ago, and I can't recall exactly what the issue was at the time. I believe that there was a problem with a cable going into my V1U but required an adapter to go into the XLR, and I believe I had to replace that at my next taping event. It was one of those sessions where I was doing 30 things at once and wasn't picking up on details like this. Now I'm finally sitting down and editing all the footage, and I encountered the result of rushing too much in a scene prep.

If I've picked up one thing from all of the veterans on this forum in my learning over the past year, it's prepping lighting and audio. The last 3-4 interviews I did for this documentary, I think I got there about 3 hours ahead of time, set up the lighting and sound, and then just chilled out until the person was ready to tape.

Thanks so much everyone, and I'll let you know how it turns out and which method I used.

Edit: Okay, I have another favor to ask. I was just working with the footage/audio from this interview, and I realized that I ran two separate mics into the V1U XLR plugs. One was a lav mic, which was the audio in the original post above; this clearly had the problem with the wire or the adapter. The second mic ran to a Rode stereo Videomic that was on the camera. I didn't like the Rode audio, as it was about 6-7 feet away and a bit echoey. But given the problems you see with the lav mic audio, perhaps the Rode audio has more potential.

http://www.refproductions.com/Sidney/SidneyLeftChannel.mp3Lav Mic[/link]

http://www.refproductions.com/Sidney/SidneyRightChannel.mp3Rode Mic[/link]

If you all think the Rode mic -- despite the echo from the distance from the interviewee -- has more potential, do you have any recommendations to remove the slight echo and make it feel a bit warmer?

James

Here's the result of my "Brute Force" negative feedback technique, haven't used it in some time, but it had a huge effect on this audio.

It's so quiet in the gaps (< -50dB) while preserving the content that some will accuse me of using a noise gate, but I didn't . . .

I did lose a little in the quieter sibilants using this approach - it would be better if I backed off a bit on the feedback track and lived with a "little" more noise.

********

I'll post my workflow (the whole process takes about 5 min.) if you're interested in trying it this way.
Back to my other projects . . .

That sounds like you used a noise gate of some sort. It certainly is quiet during the pauses, but to my ears sounds somewhat strange when the voice suddenly appears from this dead silence, along with some sort of residual buzz.

The engineers at iZotope did reply and said they would think about the idea. In the meantime, they provided me with some settings for RX that I wouldn't have considered. It involves using their denoiser, but using the "C" algorithm, which takes a LOT of computing power. I'll try it, and if the results are good, will post the audio file.

OK, here's an MP3 version of the file restored using the iZotope RX settings I got from one of their support people. It sounds a little hollow to me, but other than that, pretty much all of the buzz is gone. I'm pretty impressed with the result.

I think that if I had more time, I could improve on this by doing 2-3 separate passes, using much smaller settings for each pass. This often results in a more "lively" sounding result, but with almost the same amount of noise reduction.

Here's the audio from the 1-pass iZotope RX using these settings (direct from iZotope):

Tonal and Broadband Controls: Linked
Threshold: 0
Reduction: 23.2
Musical Noise Suppr: 5
Residual Whitening: 5.0

iZotope RX restored version (MP3 file format)

(link good for seven days)

Nope, no noise gate at all, just brute negative feedback applied at full strength. That residual buzz when he speaks might be reduced with some adjustment of the feedback compression and EQ on the feedback track, but probably not a lot.

Like I said, it would sound more natural and take out less of the sibilants if applied it more conservatively, but I was interested in seeing how much of the noise I could take out without actually increasing it. Having done that, backing off to an "acceptable" noise floor of around -45dB makes it sound a lot more natural, although at that point the buzz begins to creep back in.

Hope you take it easy with the folks at izotope; in addition to Joan Baez without string noise, we could soon end up with Oscar Peterson with no humming, the Grateful Dead with the right chords, Stevie Nicks singing on pitch, and Ringo Starr playing on the beat. The possibilities are truly frightening . . .

Golly, those are all my favorites. I'll make sure not to mess with them.

Ever heard Stevie Nicks sing "At Last"? Does she sing on key? Maybe not, but who cares? Have a listen. Unbelievable:

At Last

Speaking of "at last," here is my final, last attempt at restoring this audio. I used almost all the ideas presented by others in previous posts: I started using the click/crackle tool in Sound Forge to get rid of the main impulse. This left the residual from each impulse. I saved the file from SF and imported into iZotope RX. I then sampled a noise segment and used the settings given by the iZotope engineers, but at only 12dB reduction instead of 23dB. When I was finished, since I had a little more time before my DVD burns were finished (I do this stuff when I have time to kill), I used the spectral repair function to remove the background bangs and crashes. While there is a little residual noise remaining, I think the result strikes a much better balance between realism and noise, and is pretty easy to listen to (speaking strictly about the audio quality; the actual material being discussed is anything by easy listening -- I'm still trying to get pre-Holocaust 16mm film from a high school friend whose grandfather shot it in 1930-31 Germany when he started to see the precursors to what eventually happened).

Oh, here's the link to the "final" file (good for 7 days):

Click/pop/iZotope denoise/iZotope spectral repair