Help ME! Removing Background Noise!


I have a Micro-cassette connected to my MBP (version 10.7.5) and I am exporting it into Audacity 2.0.6.

I am really hoping someone can help me with this. Ive been trying for forever to remove background noise without distorting the voice. I have done everything google says to do by others and maybe I am just not getting it so I am hoping someone in this forum will be patient enough to help me. I get a lot of jobs from cassette recording or phone recordings, that require removing background nose and increasing the voice so you can hear what the person is saying.

I need some help with understanding how to play with the noise removal menu. I already zoomed in, selected a small portion of the background noise, then selected all and applied the noise removal. After I go and listen to it, the voice sounds robotic or distorted somehow. Can someone help me increase the voice and eliminate the background noise so I can hear what is going on?

I attached 4 images so you can see what I am may be doing wrong. Thank you!
System Pref. MBP.png
Audacity Preferences.png

I am exporting it into Audacity 2.0.6.

Step one. Noise Removal in earlier versions of Audacity wasn’t all that good. You should be using Noise Reduction found in Audacity 2.1.0 or 2.1.1.

That’s not background noise. That’s another performer in your show and those are rough to manage.

Select about ten seconds of performance with noise, export it as WAV (Microsoft) and post it here. Try to include at least one or two seconds of noise only. This is the part that usually kills people with very high noise level. How do you tell if the noise is by itself [ahem] with all that noise.

You don’t. If you include any natural sound or desirable performance in the Noise Reduction profile, it stops working. High noise level is very difficult.

Scroll down from the forum text window > Upload Attachment.


Thanks for the response. I actually downloaded the new one (2.1.1) right after I sent the first email so I have been trying to mess with it. I have attached a sample of the audio with partial voice.

This is the WAV format.

This is a good as I can do with the new tools (attached).

It sounds like you were recording in Denver and the performer was in Philadelphia.

Drag-select a bunch of sound you think has nothing but noise for the profile. Then select the whole show by clicking just above MUTE and apply Noise Reduction with the numbers 30, 6, 6. The first number is the amount of reduction or correction. Normal shows come in at 6, 6, 6 (the noise reduction of the beast).

I’m not kidding. You were recording the wrong microphone. This is exactly what it sounds like when you do that.

You might get used to the idea of having no show. Most of the dialog is going to be unrecoverable.


You can use the scratch test to make sure you’re recording the right thing.

We’re all assuming the performer knew he was being recorded. He did, right?


lol, actually these are not performers. These are cases that are recorded by another party, which, did a crappy job. It still sounds really low. You can barely make out what they are saying. Isn’t there some kind of CSI stuff?

What do you mean the noise reductions comes in at 6 6 6? I did the 30, 6, 6 but I don’t see the 6 6 6 that your talking about. I attached a couple pics to see what I am doing wrong.

Anyways, after I did the 30 6 6, i felt like it removed some of the voice.
Screen Shot 2015-09-30 at 8.02.18 AM.png
Screen Shot 2015-09-30 at 8.02.07 AM.png

Not with Audacity. We can’t and won’t help you with legal investigations, if that is what you are doing, because you need a known, qualified audio forensics expert to work on your sounds to make them legally admissible.

So use your favourite search engine to search for “audio forensics”.


The CSI stuff is science-fiction. Digital-processing only rarely causes the sounds to go from unintelligible to become comprehensible. Currently the human-brain is better than any computer at understanding speech : so if you can’t make out most of what is being said then odds-are digital-processing isn’t going to help.

BTW IMO “Test2” sounds like …
" … Terabrain … in front of … are … and mystery … same version … some of … ",
but that could be wrong.