Freesound Forums

Hello you fantastic people!

SO I have this little personal project, as where I am trying to mix a song for a friend's birthday.
It consists of some background beats, music, and mixed in soundbits.

It's the soundbits that are turning out to be tricky to me.

One of the soundbits is from an episode of Narcos Mexico series on Netflix, and I found the specific part I need in a youtube video.

The small clip is no more than 4-5 seconds..but there is too much background noise, due to them driving in a car with radio etc.

I basically want to simply catch the voice, exclude the background noise and perhaps even the radio chatter (if possible). I tried using Adobe audition, various noise reduction tools.
Though I can't seem to get rid of the humming without losing too much of the voice.

Any ideas, better ways? tips.. tricks?

Sincerly Andy

In this case, there may be enough extraneous audio for you to be able to effectively employ audio separation. Try mvsep.com using a few different algorithms. I usually use Demucs3 Model B, but I don't know which one would work best for your use case.
In the case of Demucs3, some instruments (tambourine, white noise, some hihats, and upper midrange lofi sounds) are still partially determined to be "vocals" and included in the vocal stem, but they're still drastically reduced in prominence and can usually be removed entirely by normal EQ/Denoising techniques. Often simply trimming/fading the intervocal parts of the wave will eliminate them on its own.
I use a free VST called Bertom Denoiser for denoising, and it's usually good at detecting not only "noise" but "extraneous audio" such as low-volume background music. I recommend daisy-chaining a few instances at low settings as opposed to using 100% wet, because the latter will often noticeably darken the sound, and high-shelving it to get the treble back will reintroduce the noise. Some phasing does occur at the more extreme settings as well.

If it's only a few seconds, manual spectral-editing is a feasible solution ...
https://www.youtube.com/results?search_query=spectral+editing

strangehorizon wrote:
In this case, there may be enough extraneous audio for you to be able to effectively employ audio separation. Try mvsep.com using a few different algorithms. I usually use Demucs3 Model B, but I don't know which one would work best for your use case.
In the case of Demucs3, some instruments (tambourine, white noise, some hihats, and upper midrange lofi sounds) are still partially determined to be "vocals" and included in the vocal stem, but they're still drastically reduced in prominence and can usually be removed entirely by normal EQ/Denoising techniques. Often simply trimming/fading the intervocal parts of the wave will eliminate them on its own.
I use a free VST called Bertom Denoiser for denoising, and it's usually good at detecting not only "noise" but "extraneous audio" such as low-volume background music. I recommend daisy-chaining a few instances at low settings as opposed to using 100% wet, because the latter will often noticeably darken the sound, and high-shelving it to get the treble back will reintroduce the noise. Some phasing does occur at the more extreme settings as well.

- Thanks for reply, I will check it out. It doesn't have to be perfect, so I can live with a bit of phasing.

Timbre wrote:
If it's only a few seconds, manual spectral-editing is a feasible solution ...
https://www.youtube.com/results?search_query=spectral+editing

- Thanks for reply, also seems like a viable option worth checking out.

If anything I will just go with what I have, as I mentioned above it does not have to be perfect at all. But just enough so that the vocals stand out and can be heard. Once mixed together with music, it will work better I think.

Thanks both of you