Noise situation for a documentary

[Dennis Stevens]

Originally Posted by Dan Brockett

Hi Dennis:

Please tell me that you were a smart boy and recorded with a lav and a boom as everyone should everytime possible?

If you did, you should find that the lav will have a LOT less of the offending noise than the boom.

Dan

Thanks, and yes we did both lav and boom. The lav channel is a bit better, but not that much to my ears. But another point in our favor. Just yesterday I was thinking, 'oh yeah, should have used c-stands and blankets'.

We also recorded (as we always do) a good 40 seconds of room tone, for each interview since we re-arranged the background.

Had to tell people you can't talk AND you can't sniffle! OR snort! OR cough! OR fart! Don't breath for crissakes, you can do that later!

I'm going to put up a sample of the audio for those interested, didn't get to it yesterday.

[Dennis Stevens]

I have Adobe Audition 2.0, and was playing around with the noise reduction. It didn't clean up the sound to my satisfaction, so I'm thinking we will just have to re-interview people.

I've seen a lot of references to SoundSoap - Would you guys say more effective than Audition? I've seen SoundSoap 2.1 and SoundSoap Pro on the Bias website. Quite a price difference - is SoundSoap Pro much better at noise reduction, anyone know?

[Guy McLoughlin]

Originally Posted by Dennis Stevens

I have Adobe Audition 2.0, and was playing around with the noise reduction. It didn't clean up the sound to my satisfaction, so I'm thinking we will just have to re-interview people.

Hey Dennis,

I've had pretty good success with the Sony Noise Reduction plug-in 2.x ( now part of Sound Forge 9 ) the past couple of years. If you could post a sample of your audio, I'll see what I can cook up. If I'm happy with it, I'll send you the results to check out.

Originally Posted by Dennis Stevens

I've seen a lot of references to SoundSoap - Would you guys say more effective than Audition? I've seen SoundSoap 2.1 and SoundSoap Pro on the Bias website. Quite a price difference - is SoundSoap Pro much better at noise reduction, anyone know?

Yes, there's a big difference between the pro and non-pro versions, though I do most of my clean-up with the Sony NR tool because I find it easier to use ( in either Sound Forge or Vegas ) than the SSoap user interface.

[Dennis Stevens]

Originally Posted by Guy McLoughlin

Hey Dennis,

I've had pretty good success with the Sony Noise Reduction plug-in 2.x ( now part of Sound Forge 9 ) the past couple of years. If you could post a sample of your audio, I'll see what I can cook up. If I'm happy with it, I'll send you the results to check out.

Thanks, I will. I meant to do it yesterday, but didn't have time. In the next day, I definitely will.

[Dennis Stevens]

Here it is. It's about 60 seconds long, the first 30 seconds is just the room tone. The last 30 seconds is the person talking, of course.

http://www.catzillaproductions.com/Shared/Beth1.wav

It's uncompressed from Premiere Pro, but it's only 10 MB. One channel is the lav, the other is the shotgun. Well, that's how it was recorded, I'm not sure if that's preserved in the export.

[Guy McLoughlin]

Hi Dennis,

I took a quick stab at it using Sound Forge 9. Here is what I did to process your file:

- Split the channels, and decided that the right channel ( probably your lav ) was more useable.

- Ran DC Offset just in case.

- Minus 21 dB RMS normalization

- 3 and 4 passes with the Sony Noise Reduction plug-in. Made a noise print for each pass, then processed the audio at -6 dB in Mode 2.

- Tried a little Paragraphic EQ at -5 dB at 4,825 Hz wide 2.5 octave to try to smooth out the midrange a bit.

Here's a MediaFire link to check things out:

http://www.mediafire.com/?4kqckzgz12l

...Let me know what you think.

- Guy

[Dennis Stevens]

Wow. That is awesome. I thought Beth1R_4pass.wav was best. I used the NR tool in Audition on that.

Comparing Before and After is pretty incredible. Bravo, sir! Bravo!

I also know about mediafire now. I put my version (based on your version) here for you to listen -
http://www.mediafire.com/?1sqjtb0ezby

[Glenn Davidson]

Guy,
That Beth 1r 4 pass is great. Can you provide more details. I use Waves Restoration and it's common to get the artifacts like those heard in Dennis' version. A screen shot of your set-up would be helpful to correlate Sony's setting to Waves. Why RMS normalization? What is Sony's "Mode 2"? Why 6db under?

The key must be the multiple passes with gentle NR each pass. Process time would take a while especially on large file, but the results are fantastic. Good Job.

[Guy McLoughlin]

Originally Posted by Dennis Stevens

Wow. That is awesome. I thought Beth1R_4pass.wav was best.

Yeah, I thought so too, though I was doing this from a client's office with the client's crappy $30 Sennheiser DJ headphones, so I wasn't a hundred percent sure I was hearing the real thing. Beth's voice was sounding a little nasal, which might be her natural voice, which is why I tried using the Paragraphic EQ to hide this quality.

Originally Posted by Dennis Stevens

I used the NR tool in Audition on that.
Comparing Before and After is pretty incredible. Bravo, sir! Bravo!

Thanks. It's the process of trying to figure out how much NR to apply is the hardest aspect for me. If you don't do enough, the noise will still be front and centre. Go too far, and noticeable distortion starts to creep in.

Originally Posted by Dennis Stevens

I also know about mediafire now. I put my version (based on your version) here for you to listen -
http://www.mediafire.com/?1sqjtb0ezby

...Please don't take this the wrong way, it's possible that it's my equipment isn't reproducing things correctly ( I'm at home now using my Sony 7506's ), but I'm getting some clipping distortion when Beth speaks loudly, like when she says the phrase "Our organization". Visually the top of the sine wave also looks clipped too, so you might want to sort this out.

Also, it sounds like Audition is using a noise-gate which definitely helps to lower the noise floor, but I find it can draw attention to itself unless there's a music or ambience track to hide when the "noise floor drops out". ( i.e. Suddenly it's too quiet )

...Well, I hope this helps to save you from a reshoot. It certainly looks possible to fix your noise in post.

Cool stuff.


- Guy

[Guy McLoughlin]

Originally Posted by Glenn Davidson

Guy,
That Beth 1r 4 pass is great. Can you provide more details.

Sure thing. Though I don't know how much of this will translate unless you are using Sony Sound Forge with the NR 2.0 plug-in. ( now included with SF9 )

Originally Posted by Glenn Davidson

I use Waves Restoration and it's common to get the artifacts like those heard in Dennis' version.

What I like about Sony's SF is that I can apply each effect seperately, and use my own judgement on how happy I am with the results. I don't know Waves Restoration at all, but hopefully it's not too automated, so you can apply effects one at a time, and check your results as you go.

Originally Posted by Glenn Davidson

A screen shot of your set-up would be helpful to correlate Sony's setting to Waves.

You'll have to let me know how successful you were at translating them.

Originally Posted by Glenn Davidson

Why RMS normalization?

Root Mean Square Normalization, which is supposed to be a better representation of the overall "power" of the recorded sound file. I find that the standard Peak Normalization can sometimes really distort the volume of an audio file, unless you manually correct this afterwards. RMS Norm isn't perfect either, but I like it's effect more than Peak Norm.

Originally Posted by Glenn Davidson

What is Sony's "Mode 2"?

Sony's Noise Reduction 2.0 plug-in offers four modes of NR, which are essentially the strength of the NR effect. Mode 0 is brute-force, while Mode 3 can be quite gentle. Mode 2 is the default, which generally does a good job. I think of this like choosing coarse sandpaper versus a very fine sandpaper. The finer stuff often works better, but takes longer to get the job done. :-)

Originally Posted by Glenn Davidson

Why 6db under?

This is the Noise Reduction effect in dB. You are less likely to "bork" an audio recording by correcting things in multiple passes, especially if you make a noise-print of the noise you are trying to get rid of with each pass. Remember the sandpaper thing?

Originally Posted by Glenn Davidson

The key must be the multiple passes with gentle NR each pass. Process time would take a while especially on large file, but the results are fantastic. Good Job.

Yeah, I'm a better audio doctor with a scalpel, than I am with a machete. <g>

Screenshots of the key SoundForge 9 Effects I Used
http://www.mediafire.com/?ctvmtw4999j


- Guy

[Glenn Davidson]

Thanks for the screen shot. So are you normalizing to RMS value to get the overall level up before you begin the the NR process? It is probably better than peak normalization since one spike can really throw they whole level off.

When you are sampling the noise how long is the sample? Typically I will use a fraction of a second.

I know what you mean about the difficulty of how much processing to use. The temptation in the process, at least for me, is to try to go too far in one pass. This creates the distortion or mechanical sound..

I am going to play around with mutiple passes and see if I can get close to your incredible results.

For fun I have included a Waves X-Noise screen shot.

[Dennis Stevens]

...Please don't take this the wrong way, it's possible that it's my equipment isn't reproducing things correctly ( I'm at home now using my Sony 7506's ), but I'm getting some clipping distortion when Beth speaks loudly, like when she says the phrase "Our organization". Visually the top of the sine wave also looks clipped too, so you might want to sort this out.

Also, it sounds like Audition is using a noise-gate which definitely helps to lower the noise floor, but I find it can draw attention to itself unless there's a music or ambience track to hide when the "noise floor drops out". ( i.e. Suddenly it's too quiet )

...Well, I hope this helps to save you from a reshoot. It certainly looks possible to fix your noise in post.

Cool stuff.


- Guy

Yeah, it certainly isn't perfect, but from where it was before, it's much better. Hopefully I can duplicate what you did in my copy of Audition.

[Dennis Stevens]

Originally Posted by Glenn Davidson

When you are sampling the noise how long is the sample? Typically I will use a fraction of a second.

How does the length of the noise sample effect the process? Is it better to do a very short sample, then apply Noise Reduction?

Would a smaller sample take out a smaller piece, so less destructive to the dialog? Then you take another tiny sample, apply NR, and repeat until you get to where you like it?

[Guy McLoughlin]

Originally Posted by Glenn Davidson

So are you normalizing to RMS value to get the overall level up before you begin the the NR process?

Yes. I don't know if it's better to the NR first or the normalizing, but I figured that it would be easier for me to hear the subtleties of the NR effect if I did my usual audio processing first.

Originally Posted by Glenn Davidson

It is probably better than peak normalization since one spike can really throw they whole level off.

In theory yes. RMS normalization is still effected by audio spikes, just to a lesser degree than Peak Norm.

Originally Posted by Glenn Davidson

When you are sampling the noise how long is the sample? Typically I will use a fraction of a second.

I used almost all of the sample that Dennis provided, figuring that it would guarantee a good sampling of all of the noise frequencies that needed to be taken care of by the noise-print process. I think one second would be as effective provided the noise is very consistant. The AC unit seemed to be a steady drone.

Originally Posted by Glenn Davidson

The temptation in the process, at least for me, is to try to go too far in one pass. This creates the distortion or mechanical sound.

Yeah, as soon as I start to hear that metalic "phlanging" I know I've taken things too far.

Originally Posted by Glenn Davidson

I am going to play around with mutiple passes and see if I can get close to your incredible results.

It would be great if you could describe what your process is, so we'll have a recipe on file for Waves Restoration too.

Originally Posted by Glenn Davidson

For fun I have included a Waves X-Noise screen shot.

A lot prettier than Sound Forge, but I think that's a Mac thing. Most Mac apps seem to have better looking UIs.


- Guy

[Seth Bloombaum]

Originally Posted by Dennis Stevens

How does the length of the noise sample effect the process? Is it better to do a very short sample, then apply Noise Reduction?

Would a smaller sample take out a smaller piece, so less destructive to the dialog? Then you take another tiny sample, apply NR, and repeat until you get to where you like it?

Yes, yes, yes and yes.

Add one step - each time you take a noise sample you have some settings you can apply to the NR based on that particular capture. Doing multiple caps and filters is good technique, but you should have a light hand with the NR settings for each sample.

Also, Sony NR offers a checkbox for "keep residual" This is worth selecting and listening to, as it will preview what's being thrown away. If you hear some voice-like sound either back off on the settings or take a new sample until you are hearing and throwing away noise only. (don't forget to uncheck "keep residual" when you're done!)

One more note: Sony NR is a direct-x plugin that also appears in Vegas, and probably in other sound apps that use direct-x plugins (but I've only used it with Sound Forge & Vegas).