preset for backgrund noise separation


I’m working on old voice records that has extensive background noises (loud conversations, playing and screaming childrens, motorcycle pass and etc. ) I applied to record: normalization, then compression and then low rolloff for speech effects to enhance the voice record. But i need to extract (or enhance or distinct) the specific speaker in this record. Do you know a method to solve this problem? Or is there a compression preset to enhance specific speaker?

thanks in advance…


Dynamic range compression will raise the background noise, so counterproductive to vocal isolation.

Even Artificial Intelligence will confuse simultaneous speakers …
[ but it can exclude non-voice sounds, like engines, fans, mains-hum ]

loud conversations, playing and screaming childrens, motorcycle pass and etc. )

:frowning: Virtually impossible… Pros still record in soundproof studios and radio & TV studios are soundproof. On-location news reporters often pick-up background noise and their best defense is speaking close to the (directional) microphone.

The low roll-off for speech will help a little if there is low frequency noise. A high frequency roll-off can also help if you have high frequency noise but if you go too-far you loose the “T” and “S” sounds and speech gets less intelligible.

The human voice covers most of the audio range and most natural sounds and noises cover the same range so it’s generally difficult or impossible to separate the noise.

Even Artificial Intelligence will confuse simultaneous speakers

It confuses human intelligence too! :stuck_out_tongue: When we listen live there are directional clues that help our brain separate-out different sounds/voices but once it’s recorded everything comes out of the same speaker(s) and it’s more difficult.

You hit number 4.

Screen Shot 2023-01-09 at 12.40.27 PM.png
That’s not to say that a theatrical presentation is dead-flat impossible. If you have many million$ riding on the recordings, you hire either the original performers or someone who sounds just like them and speak the same words in a studio.

Screen Shot 2023-01-09 at 12.48.16 PM.png

Audacity 3.2.3 - Win10

Audacity 3.2.3 - Win10

You’re still stuck.


thanks for the answers!

i wouldnt want to press no4 either but records came to me from somewhere else :smiley: bu yes you are right, thank you.

thanks Trebor! I will try out the Adobe Enhance. My goal is to supress the background or irrelevant sounds as possible. Maybe that helps a bit.

exactly! even we cant understand the conversation precisely in record. But i thought maybe if we enhance the record somehow, we can turn it more understandable.

im going to try the dynamic range compression at first. Then try out the Adobe Enhance. If i wouldnt succed, there is an implementation called speakerbeam which refers to this paper: I’ll try it.

Any other recommendations is highly appreciated.