Constant but loud background noice

I have recently recorded an interview that I am thinking to use in my podcast because it sounds quite interesting. The problem is that I was recording it in a shop in a room with several big wine fridges/displays. As a result, there is strong background noise. I tried to remove it using Noise Reduction but if I remove it completely the voices are quite distorted and if I keep the voices reasonable the background noise is low but quite annoying. I wonder if there is a semi-manual way to reduce the noise to a minimum without distorting the voices?

If fridge noise is constant it may be possible to notch it out without conspicuously distorting the voices.
Look for bold constant horizontal lines on the spectrogram display … https ://

Are the interviews in WAV format (I hope). If they are and they’re in mono (one blue wave), you can post up to 20 seconds of sound sample on the forum. If they’re in stereo, ten seconds.

Scroll down from a forum text window > Attachments > Add files.

Post the raw interview, not something you already tried to fix. Include general conversational words and at least one hole or place where there are no voices, gasps, or mouth noises (pure background sound).


No, the interview is in mp4/3 format, unfortunately. I actually recorded it on my phone using Google Recorder to get transcripts. I didn’t plan to use it in the podcast. But the conversation turned out quite interesting so I am now eager to release it as a podcast episode if possible.

My spectrogram looks like this.

In any case, I have exported 20 sec into wav and I am posting here in case it ok. Thanks a lot for your help.

Thank you for the sound track, but it’s missing the one thing we need for good noise reduction. He never shuts up. We need a good fraction of a second with refrigerator only and no voice or gasping.

Note at 7 seconds, 9 seconds, and about 17 seconds, he gasps for breath between words. Also note at about 13 seconds, the interviewer interrupts the guest “Yeah.” If he does that throughout the interview, then this will be an uphill battle with them stepping on each other’s words.

Noise reduction works in two parts. Drag-select a small portion of clean noise with no performer or desirable sound. Effect > Noise Reduction > Profile. That lets Noise Reduction “sniff” the noise like a dog getting the scent. Then you select the whole performance and run Noise Reduction again with the desired correction values set. Sometimes you can run Noise Reduction more than once.

There are other tricks, too. Noise Gate can remove sounds lower in volume than a set value, and the above mentioned Spectral Edit. Each has its drawbacks. If the noise is complex and has a large number of different tones, then Spectral editing will be a nightmare.

Was this a video?


If it wasn’t a video, you can use a Hollywood trick of writing down the words and have an actor with a similar voice say them in a quiet studio.


No, no video. It was recorded on my Pixel phone using Google Recorder. I have attached another section that has gaps as well as a portion of the noise background from the beginning of the conversation. Yes, the guy was talking nonstop and he mentioned at the end that he was hoping to talk over the noise background. I interjected sometimes with yes etc. but I probably can edit it out.

This is gentle noise reduction as good as I could do with the existing performance and an aggressive Noise Gate.

I’ll look at the second posting.


Much Better.
This is the second posting with two passes of noise reduction and loudness boost.

And another with all that and reduced treble to get rid of that harsh SS sound.

This is a panel with the Noise Reduction settings.

Screen Shot 2021-11-13 at 4.22.28 AM.png
I’ll pick this up again in the morning. One of the problems with this kind of rescue is The Correction With No End. it’s a career move. “Maybe if I increase the bands…”


It’s a bit too difficult for my modest podcast :wink:

Wow! It’s much better than what I got trying to do Noise Reduction. Thanks a lot! I will wait for you to have another go if you don’t mind. I know, the better is the enemy of the good :slight_smile:

OK. Here’s full correction.

And here’s how I got there.

Inerview-Rescue.txt (382 Bytes)
Filter Curve is the Audacity version of the standard 100z rumble filter that appears on every field sound mixer.

Two gentle passes of Noise Reduction using the Profile from 2 Sec.

European (50Hz) hum remover

Redo the profile and two more passes of Noise Reduction.

Loudness Normalization to bring the volume back up.
Limiter to soften the clipped peaks. Note these two have to be used together. Don’t worry about the clipping warnings between them. They clean each other up.

Desibilator to soften the harsh SS sounds.

Some of these have to be downloaded and installed and I don’t remember which ones.


100Hz is the worst offender …

https ://

Filter Curve is hidden.

Effect > Filter Curve > Manage > Factory Presets > Low Rolloff for Speech > Enter

It also doesn’t tell you which curve is loaded. It should look like this. Nothing happens to the right.

Screen Shot 2021-11-13 at 9.08.29 AM.png

100Hz is the worst offender …

I saw that. That and 200Hz vanished in the Hum Remover - European step. Some of it also goes away in Low Rolloff.

You can throw these tools in or take them out and see what happens as you want.


Thanks a lot, guys! Really helpful. I will try the effects and various combinations and see what delivers the best result!

I just wonder what is the best algorithm.

I just wonder what is the best algorithm.

Note: Noise Reduction Profile happens twice. Past the first pair of NR steps, the Profile isn’t valid any more and you have to get a new one.

Most of the tools contrast effectiveness versus time needed. Some of them can do a terrific job, and they’ll be done when the sun cools off.

The combinations and settings affect each other. So there is that. How old are you? How close to retirement?