I made such a rookie mistake. Can I salvage this?

I’ve recorded an interview for a new YT channel and I had promised everyone that I would produce beautiful videos that do justice to the amazing people through outstanding quality…

After locating a small photography store in my home town with a corner in the back that can be used to take portraits or record simple video I asked them if I could rent the place for 1 hour and they agreed for a small price. I couldn’t afford a serious professional studio.
I turned up with my guest and good microphones. We started the session, there was some background noise but it didn’t seem too bad at the time.
I had also done some recording tests beforehand in various environments and I was able to remove any noise with superb results in Audacity 3.1.3, so I was very confident about the whole situation.

Well, the end result is pretty bad… the microphone really picked up everything that was going on through walls, it was pretty rainy and windy outside.

If you take a listen to my audio samples, you can see I have terrible noise in the source audio. If I apply noise reduction it does erase it, but my voice also becomes tinny!
I have good portions of pure noise in my source audio and I made sure to select them carefully to create the noise profile as always, but even though I’ve tried with multiple noise samples (the whole interview is 1hour), my voice always becomes tinny!

Is there anything at all I can do to prevent my voice from changing so much or to enhance it again after noise removal?

Any and all help will be immensely appreciated.

The MP3s have been exported with maximum quality at 320kbps. Original files are wav.

The “Before noise reduction.mp3” (201.02 KiB)" looks/sounds* like a limiter/compressor/(AGC?) has been applied.
Should apply any Noise Reduction before limiting/compression,
as those effects will cause the noise-level to become varied.

[* can hear the background-noise pumping-up in the gaps between words & sentences,
and modulating the speech, e.g. the long “eeeEeeEeee” ].

I haven’t applied anything else in Audacity, it really is the original source audio as-is. It’s “the microphone’s fault”.

The recording-device may have an “AGC” feature, which Automatically turns up the gain when the sound is quiet, and down when it is loud.
It’s usually possible to switch AGC off, but its compression effect is baked-in to the recordings made with it on.

Also if the noise is naturally varying with time, like gusts of wind, intelligent (AI) noise-reduction may be necessary, e.g. https://youtu.be/0-7iUkfcf8s?t=45

Hey, are you kidding me?
What did you is much, much better than my result! The voice remains natural and perfectly understandable!

Can you share the exact steps so I can improve my entire recording?

Thanks so much!

#1. Remove the frequencies below ~50Hz with a high-pass filter, (no voice down there, only noise).
#2. attempt to undo the AGC compression using Dynamic-Mirror plugin to expand dynamic-range.
#3. Apply typical Noise-Reduction settings: 6-6-6 …

Original files are wav.

So right away you’re miles ahead of people trying to do all of their production in MP3. MP3 creates sound distortion and it gets worse as you edit.

maximum quality at 320kbps.

Good call. You may not be able to tell the difference between the higher qualities of MP3, but the lower qualities just kill the production and analysis tools.

Noise Sample.

I would have included a longer sample. Not for the Reduction Tools, but just so we have a good chance of recognizing what made the sound.

Before Noise Reduction

Include, if possible, a place where the guest pauses or takes a breath. The relationship between the guest voice and the background noise is important, but the transition between them is important, too.

background noise but it didn’t seem too bad at the time.

That’s an experience thing. If you can tell your computer is on just by listening, you’re going to have noise reduction problems.

I’m not fond of the default Noise Reduction settings. I designed the Audiobook settings of 6, 6, 6, or Noise Reduction of the Beast. It can reduce background noises and nobody can tell what you did. If you need stiffer reduction, bump up the first number gently up to 12. Past that, you will be able to hear some reduction damage.

We’ll see what Trebor did.

If you do more of these, you may find locations around town that work well for interviews. If you’re shooting video, you automatically have to find someplace that looks good. Those are hard. That’s a studio. People always go for the video first. There’s a YouTube performer that has perfect video, good lighting, entertaining backgrounds, and good content, but he sounds exactly like he’s recording in a bathroom.

I have made good recordings in a large, empty warehouse and my messy garage. Neither one would work for video.


I use Koz’ prefereed 6.6.6 seiings and I finf thay work well for me - most recentlt removing FM carrier hiis fro amlive BBC broadcast of a folky friend of mine.


You even recoded the steps, I cannot thank you enough… I will apply them, listen to the whole audio file, report back!

Reportedly the noise is rain and wind, (apparently made even more intrusive by AGC).
As that type of noise will vary over several seconds, maybe helpful to create a composite noise-profile file from various locations where there is no speech.

Thanks a lot for your input as well Koz.
As regards the 320kbps MP3s, I was only mentioning that to let you guys know that here I am not sharing highly compressed samples that would make things even worse. But I exported to MP3 instead of wav to keep the files smaller on the forum. I won’t use MP3 in my final product.

I don’t have a much longer sample because there wasn’t that much silence between questions and answers.
But I do have new information: I think that my original assumption of rain and wind through thin walls was wrong: I went back to the place and noticed they have an industrial air conditioning unit constantly running, that was it. So, for better or worse, the background noise was pretty consistent.

As regards the location, it’s not easy for a number of reasons… first is budget, if I had serious money I could just go to high-end studios that have amazing sound insulation and no noise whatsoever inside. But the cost is no joke: if I were to go to such a studio I could only produce content once or twice per month and in short sessions. I also evaluated other options such as recording at home or potentially even at other businesses.

Hopefully I can improve things one way or another as I learn and grow.
I can say that the voice of the interviewed, which was recorded with a separate and better mic, sounds near perfect after just simple noise removal with the default settings.
The issue is mainly with my voice, then one you hear in the samples.

With all that said there’s no denying that the first obstacle to getting great results is me: Audacity is amazing overall, but I mustn’t feed it trash. I must work with good equipment and then set it up in controlled conditions…

How do I build a composite noise-profile?

Copy & paste a few noise-only samples, (from, say 3-4 areas), into a separate audio track.
Then select all of that noise-only compilation track as the noise profile in Audacity.
This strategy may help if the noise changes during the interview:
a single noise profile of <1 second may not be representative/optimal if the background noise is variable.

How do I build a composite noise-profile?

Is this for the time saver? Apply one reduction over several segments?

You may not want to do that. Noise Reduction does cause some sound damage even when you make it do what you want. Each damage is unique to the Profile. If you pile up several profiles in one reduction pass, you will also get several damages.

Noise Reduction only works well on noises that do not change. That’s probably why your job worked as well as it did. The air conditioning sounds didn’t change. Had the noise actually been weather and wind sounds, you would be shooting this performance over again. Traffic has the same problem. Constantly changing.

I can save you a step. You would think Noise Gate would work here. Any sounds quieter than a set value either dips in volume or vanishes. It sounds good in the advertisement, but Gates can distort the beginnings and ends of words and the performer can’t move their head or apply theatrical changes or emphasis. Since Gates have no effect inside words, their effect can give you talking air conditioners.

Noise is not fun. Noise kills more audio performances than anything else.


Another AI noise-removal plugin … https://youtu.be/LLI4wPUXBgM?t=122 ,
it won’t work in Audacity though: it’s VST3,
(and it requires a particular graphics-card for the necessary audio-processing grunt-work).

Right, the composite noise profile didn’t do much difference in Audacity.
I am taking a look at the other suggested software as well.

I’m still working on it, but in the meantime I wanted to share a great tip I found in a tutorial.
Noise reduction tip.png

Deliberately Capture

Exactly correct. That’s why it is required that you have some pure background noise (Room Tone) at the beginning and end of each submitted audiobook chapter.

have room tone at the beginning and end and be free of extraneous sounds

That’s also built into the little ten-second voice test for submission to the forum.



Room Tone can also point to other sound problems on your computer. If Room Tone doesn’t sound right or changes over time, that can mean something in your system is “helping you” by changing the sound channel while you work.

Leaving Zoom running in the background by accident can do that.

That can give you a distorted sound recording almost impossible to fix.

There’s another fuzzy rule, too. If you struggle with computer recording longer than three or four days, record on a stand-alone sound recorder and stop trying to use the computer.


Here are my 3 best results, which would you choose?