How to run match loudness to -16 LUFS without adding background noise

[Isabel5CDC]

Hi all, 

I am editing some audio and trying to use the match loudness function to bring the audio from -32 LUFS to -16 LUFS and when I do so, it amplifies a lot of the fuzz in the background that you could not hear in the original. 

Here is a screenshot of my audio before running Match Loudness:

And here is a screenshot of my audio after:

Should I be making it louder in a different way to avoid this problem, or should I just treat this amplified noise as background noise and try to reduce it using the Noise Reduction tool?

Here is a screenshot of my settings when I run Match Loudness also:

Also included the 2 clips attached.

Thanks in advance!

[SteveG_AudioMasters_ · Accepted solution]

The noise is there anyway. If you amplify the signal by 16dB, you amplify the noise by the same amount. And anything else you do to the signal itself, like compression or limiting which would even the signals out, is going to make the dynamic range worse. And that's what we are looking at here - the dynamic range of the signal. And that isn't affected by the absolute level of it at all.

So if the noise rises to an unacceptable level, then the only thing you can do, as you've surmised, is to use Noise Reduction. This will leave the peaks of the signal alone, but lower the noise floor - thus increasing the dynamic range. Generally the thing to do initially is to normalize the signal to 0dB and then find out what the noise floor value is. Then make adjustments so that you keep the average level comfortable without loads of peaks. On your example, I found that Audition's DeNoise works surprisingly well, although like most NR, it's best run twice at a moderate setting. If you run it twice at 50%, that leaves you with a LUFS value of -15.64 and your side of the conversation at least, pretty clean.

[Isabel5CDC]

Thanks, this is very informative! When you say to normalize signal to 0db, I looked that up and it seems like that process itself makes the audio louder, so should I be doing that to all of my tracks every time instead of using the Match Loudness function and then proceeding to use the NR/DeNoise on that audio or are there some instances where I should use one over the other? Basicaly are there pros and cons to using Match Loudness vs. Normalize to make the audio louder?

[R Neil Haugen]

Steve,

Very informative answer. I especially liked seeing that running things twice at a lower level is often better than once at a higher level. Which is exactly what I've found to occur with most video color correction also.

As it seems that accomplishing a change with two small adjustments often to different tools that affect the same 'thing' has less chance of causing artifacts. It's digital data management, I suppose ... pure and simple.

Neil

[SteveG_AudioMasters_]

I've found over many years that it's generally better to explain process things a bit more than just write a bald 'do this - do that' reply. Being able to touch-type helps as well... 😉

As for getting the best out of NR, we discovered years ago now that with the sample-based NR, the best thing to do is to run it three times, taking only a few dB at a time. The important thing to do between each pass was to alter the FFT size, and use the extremes, plus one value in the centre. What's important about this, and why it works better, is that each time you take a new profile, what resulting artifacts there are will fall at different spots in the result, so the average of all of them sounds smoother. This even works with the 'posh' NR systems like iZotope's RX - it's already a multipass system, but doing more than one pass, with a fresh profile each time, works better on that too. Probably for a slightly refined version of the same reason!

[EuanWilliamson]

An alternative here would be to not amplify the noise.

Use the noise reduction tool purely to find the noise floor.

Then use the dynamics processor to amplify the voice but not the noise.

Something like this perhaps:

This is a snapshot after dynamic processing with the settings as per the green arrows.

All the dynamic increaase in volume is above the noise floor so the signal to noise level is improved.

The green box shows that the noise level has not been amplified.

[SteveG_AudioMasters_]

The trouble with doing it the downward expansion way is that it sounds terrible, because you lose the ambience disproportionately - which is why we never recommend it unless it's in desperation. The general aim is to get the best result you can with minimum degredation of the audio - which is why NR is generally a much better way to go.

[EuanWilliamson]

Thanks Steve, that's an interesting view.

For me, it sounded pretty good.

Perhaps it's one of those by case by case things.

For my own education, I'll try both methods on the file posted and test my ears 🙂

[EuanWilliamson]

A thought came to mind is that since I hear another voice quietly in the background that you may be

creating podcasts or similar.

May I point you towards a YouTube channel by Mike Russell who is a great authority on Audition

and many other audio applications.

Here he is explaining how to clean up a setup where you have cross speaker bleed into you mikes.

https://youtu.be/6HecS03NfFE?t=15

It's worth bookmarking his channel for his great collection of tips.