When the Noise Floor on Compressor is set to max (-20db), all audio experiences negative gain. Why?

After all these years, I am still trying to master my understanding of the compressor. I’ve read how all the settings work time after time. Here’s my question for today:

I have input a track of spoken audio (peaks ~ -15db) with a small amount of ambient noise around -45db on the playback level. So, I set the Compressor threshold to -20db.
*I selected no make-up gain and opted for compressing via RMS, NOT based on peaks.
(Attack time and release time set to minimum levels, 0.2s and 1s respectively)

  • When noise floor setting is low (-45db down to -80db), it works as expected, and only the highs of my track are reduced.
  • When noise floor is set to a higher level such as -30db or -20db (maxed out, but still below the Compressor threshold), everything is reduced, including noise below the “floor.”

Why would audio below the noise floor experience any negative gain/compression? Isn’t the whole point of the setting meant to dictate when the compression is engaged?


Because the fastest attack & release settings of the native Audacity compressor are too slow for speech, e.g. the shortest release-time is one second: the compression takes a full second to wear off, (that’s longer than the interval between words).

There are free compressor plugins which do a better job on speech …

Bertom (simple) vocal compressor works in Audcaity3

TDR Kotelnikov (complex, but has vocal presets) works in Audacity 2 & 3

Maybe you have the wrong goal.

If you start with a plain voice track with no wide swings in volume or presentation, you can apply Audacity Audiobook Mastering Macro and then very gentle noise reduction if needed. If you recorded in a quiet, echo-free room, you may have an audiobook quality standard voice track.

We note that if you can achieve that, you will be able to submit voice tracks to almost any client.

The Macro applies a rumble filter (100Hz Rolloff), RMS (Loudness) volume setting, and then very gentle peak reduction. That’s it.

True, if you have very wide swings in your presentation volume you maybe stuck with full-on compression management, but broadcasters have been struggling with that for centuries. It’s not easy.


Still have that raw, unprocessed WAV track around? Can you post it here? The thick horizontal bar with up arrow atop the text window is the transfer icon. The forum will allow up to 4MB of work.

If that’s too messy, you can use the voice test process I posted.



These answers are great. Let me respond to everyone with one post.


Because the fastest attack & release settings of the native Audacity compressor are too slow for speech, e.g. the shortest release-time is one second: the compression takes a full second to wear off, (that’s longer than the interval between words).

I hadn’t thought about that. Thank you for explaining. I don’t know if this is the case though. I checked again, and it’s adding downward compression on long stretches of audio (~7s) where these is only ambient noise, and the original audio doesn’t even reach -20db. And like I said, this only occurs as I raise the Noise Floor even though my ambient nose is still less than the Noise Floor. So, I’m still perplexed.


you can apply Audacity Audiobook Mastering Macro

Hadn’t tried that. Thanks for the tip. I’ve started using the “Noise Gate” to get rid of my ambient noise (I’m actually removing the noise from cellphone videos to improve the quality on my video projects). I’ve used Noise Reduction a lot in the past. Never the less, I want to master the compressor specifically because it’s so confusing.
(I made another post on it yesterday and discovered the compressor’s example graph is often lying to you, and compression based on peaks always uses upward compression regardless of whether or not you select “make-up gain.” Although, this post is regarding RMS compression for that reason. Why does compressor add gain to things below threshold (Make-up Gain is not selected))

So, I’m still lost as to why a 7s stretch of audio would experience negative compression if it’s below the noise floor, and yet, it does not experience compression if I lower the Noise Floor to its minimal setting of -80 db. Shouldn’t that be the other way around? It should experience compression if it’s above the Noise Floor, not below it, right?.. :man_facepalming: There must be something I’m not understanding correctly here.

And now, to make thing even more confusing- are we all ready for a plot twist?- If I’m using the same compressor setting on ONLY the 7s stretch of ambient noise (~ -45db), the noise selected no longer experiences downward compression at all. If I select the whole track, and I use maximum Noise Floor (still above the actual noise), the ambient noise receives downward compression, but if I only select a chunk of the noise, it behaves as expected, and it doesn’t experience compression at all because it’s below both the Threshold and the Noise Floor.

Boy, this is confusing. Thank you all for the help so far!

The audio has to go above the threshold to trigger the compressor,
it then stays on for at least the release-time which can be set from 1 to 30 seconds.

If the section of audio selected never goes above threshold then no compression is applied.

There are plenty of free compressor plugins which work in realtime in Audacity3,
e.g. The ToneBoosters v3 legacy bundle has three … https://youtu.be/5fffEP5-N5o

You are asking the software to interpret the show content and clean it up as much as it can. You’re always going to find conditions it doesn’t like.

I think you should be getting better at producing the work, not rescuing damaged sound later.

I make it a point to contact and talk to people I know are not techies and yet manage to crank out good quality cellphone videos one after the other.

One example is “Wonderhussy” where Sarah Jane uses a tiny add-on microphone with wind sock (dead kitten) on her phone during recording.

I have seen her turn out good quality voice recordings in the face of desert winds that would make me hide under the bed. She’s been doing this long enough to have a good feel for the conditions.

How are you recording your work and what is it?

I know this is old news, but I did a perfectly good voice recording with the phone on my desk in my quiet, echo-free office.

There are some tricks to it, but no rescuing or patching needed. It came out of the camera/phone ready for mastering and gentle noise reduction.

Looking for Sarah Jane’s microphone…


Makes sense. Thank you!

I recorded some videos using a cellphone, I wanted to clean up the audio using Audacity, and I started this post to better understand its compressor effect. The problem is occurring with a random test piece of audio, not any real work.

I use Audacity for many different things. I know there’s always better equipment out there I could be using, I’ve got a closet filled with gear already, but sometimes I just want to clean up the audio I’ve already got. That’s what I’m here to work on. That’s the goal.

I still don’t know why this is happening with a 7s chunk of ambient noise. When noise floor was set to a higher level such as -30db or -20db (maxed out, but still below the Compressor threshold), everything was reduced, including noise below the “floor.”

Maybe it’s just an unavoidable fluke, but if it’s not, I’d love to know why. Thanks for all the help so far.

Then I’ll sit in the corner and watch. I have no good feel for the compressor and variations. They would seem to be a gift from the angels, when in fact they’re usually a career move. We can wait for more experienced elves to post.

See Trebor’s product research.


When the Noise Floor is set to -20dB, it’s actively reducing the volume of the audio to a level below the specified threshold. If the reduction is too severe, it could result in negative gain, causing all audio to be quieter or even silenced.

The Noise Floor setting in Audacity’s Compressor doesn’t work in isolation, but rather in conjunction with the Threshold setting. When you set the Noise Floor to a level higher than the Threshold, it causes compression to engage even for signals below the Threshold, leading to the reduction of both the ambient noise and the desired audio. This is because the Noise Floor setting is essentially a secondary threshold that, when exceeded, triggers the Compressor’s action.

In your case, setting the Noise Floor to -30db or -20db causes the Compressor to engage even for signals around -45db, resulting in the reduction of the noise and the lowering of the overall volume of the track. To avoid this, you may want to keep the Noise Floor setting lower than the Threshold, or adjust the Compressor’s parameters to better suit your needs.

Thank you so much. This must be what I’m not understanding. Let me clarify a little further.

In this scenario, when I set the Floor EQUAL to Threshold (-20db), you’re saying that I’m basically screwing with the algorithm, and it’s going to compress even for sounds below both settings, because it’s not made for that scenario.
Is that correct?

What about when I set the Noise Floor to -30db? Audio at -45db should not be compressed if Noise Floor is lower than Threshold, and audio is lower than both. Right? But it’s still happening somehow.

Thanks so much

-45dB is lower than -30dB:, cf. -45°F is a lower temperature than -30°F.

Right. I’m trying to specify there that this is still happening when the Noise Floor is below the Threshold as opposed to equal in the other example DigiGod addressed.

My original audio has ambient noise at -45db, the Noise Floor was set to -30db, and both are below the Threshold of -20db. So why would it be compressed? Is it just an unexplainable glitch?

It’s the limiter that lets you choose a sound value and then then doesn’t let anything in a selection get higher/louder than that. How strictly or gently it does that is one of the settings. The Limiter in Audacity Mastering is set for maximum gentleness or softness.

Maximum stiffness or hardness is called clipping for what it does to the blue waves. That can also sound crunchy and buzzy.

A compressor has the effect of turning the volume control up and down as the show progresses. A badly designed compressor can give you breathing or pumping. Every time you pause __sssSSSS the background noise comes up __sssSSSS.

You can prevent that by setting thresholds. If the show scenes have a lower volume than a setting, don’t help it, the scene has to be quiet for a set time before the correction kicks in, etc.

More complicated settings from there are possible. You are simulating the audio engineer. That’s also why compressor design is a career move.

You might say a broadcast sound service might do with a gentle Compressor on the studio feed followed by a Limiter to prevent broadcast peak distortion.

Correct. That’s the Audimax and Volumax used for centuries.

Screen Shot 2024-06-08 at 13.43.15 copy


To clarify the behavior of the compressor in Audacity and address the specific scenarios you’ve mentioned, let’s delve a bit deeper into how the Noise Floor and Threshold settings interact.

The Threshold is the level above which the compressor will start to reduce the gain. When you set the Threshold to -20dB, any audio signal above -20dB will be compressed. The Noise Floor setting helps the compressor distinguish between the noise and the actual signal. Typically, sounds below the Noise Floor are considered noise and can be reduced to avoid amplifying noise during the compression process.

When you set the Noise Floor to -20dB (same as the Threshold), you’re effectively telling the compressor to treat everything at and below -20dB as noise. This can lead to unintended behavior where the compressor might overly reduce the volume of your audio, including parts that you don’t want to be compressed, because it’s trying to lower what it perceives as noise. Setting the Noise Floor to -30dB means that any sound below -30dB is considered noise and will be reduced. Since your audio peaks are at -15dB and your ambient noise is at -45dB, sounds at -45dB should not be affected by compression if the Noise Floor is set to -30dB, assuming that the compressor is only engaging for signals above the Threshold of -20dB.

If you’re experiencing compression of audio at -45dB even with the Noise Floor set to -30dB, there might be a few reasons. One potential factor is the knee setting. If the compressor has a “knee” setting (soft or hard knee), this could affect how aggressively it applies compression around the threshold. Additionally, while you have attack and release times set to minimum levels, very short attack and release times can sometimes create artifacts or unexpected changes in audio levels, especially with fluctuating ambient noise. Moreover, the way the algorithm handles signals close to the Noise Floor and Threshold could result in some overlap or interaction that causes compression to be applied more broadly.

To address these issues, ensure that the Noise Floor is set significantly lower than the Threshold. For example, if your Threshold is -20dB, try setting the Noise Floor to -40dB or lower to create a clear distinction. Experiment with slightly higher attack and release times to see if this stabilizes the compression behavior. If Audacity has a gain reduction meter, use it to observe when and how the compressor is engaging. This can provide insight into whether signals are being compressed as expected. Additionally, consider using noise reduction or gating as a separate process before applying compression to handle ambient noise more effectively.

By adjusting these settings and observing the results, you should be able to fine-tune the compressor’s behavior to achieve the desired outcome without unwanted compression of your audio.

I personally don’t use Audacity’s compressor so I’m not entirely sure if Audacity has a gain reduction meter. This is just a presumption as most compressors have this feature built into it.

Hope this helps clarify some things for you.

Or expansion? Hardware compressors have meters with “Do Nothing” in the middle.

Screen Shot 2024-06-09 at 07.58.02

The Audacity application notes say:

.… lower the volume of loud sounds and raise the volume of quiet sounds.

So they do expect it to go both ways.


Is your goal still to achieve a deep, thorough understanding of the Audacity Compressor, or do you have an actual, real-world application?


I use different compressors for different types of music. Mostly either the standard ones found in Cubase or FL Studio, or the Solid State Logic compressors. I appreciate audacity adding a compressor feature, but it’s definitely not my go-to. My suggestions are based on how most compressors work, from an engineering perspective. It’s how frequencies interact with each other in the compressor plugin…or hardware for that matter. Since, I’m presuming, that audacity bases their algorithm on other compressors, it would be safe to suggest the same cause and effect in their plugin.

Ever try Chris’s Compressor?

He wrote it so he could listen to opera in the car.

The only restriction I know of is an oddity of the look-ahead system. It doesn’t much like falling off the end of a track. Make sure there is more work than you need at the end. Let Chris go nuts briefly and then cut off the damage.