Extracting human vocals from really damaged audio?

Question

My friend received an audio clip from a family member he is concerned about, but the clip is very low quality. He asked me to help clean it up, but I only have a little experience with digital audio. I'm worried that there is not enough information on the clip to extract the vocals.

I've seen tutorials using audacity that allow me to reduce noise and extract vocals. Reducing noise helps a little bit, but if I remove too much noise then I start removing the vocals as well, and I get nasty artifacts in the mp3.

The original file was an amr file, I converted it to mp3 to be able to work with it. Could I have caused issues converting from amr to mp3 that may be going against me?

If anyone has any tips on advanced vocal extraction from audio with a ton of white noise that would be great.

Alternatively if anyone is a true pro and has time to help out a friend I could potentially send you the clip. It might be a real simple job for someone with a lot of experience, but I don't want to share it publicly because it does seem to contain sensitive info.

Answer 1

There are two basic ways I remove noise from an audio:

1. You can use a visual equalizer(e.g- FL studio Parametric EQ2 or fabfilter Pro Q) that show the frequencies of where all sounds in your vocals live then select a potion of the audio where the noise is very audible then filter that frequency range. This video should explain better.

   2. You can use (Fl studio's Edison) de-noiser. It is a more advanced noise reduction tool than Audacity's noise reduction. This video should explain better.

Answer 2

The best tool for audio restoration nowadays seems to be Izotope RX.

If you don't have access to the software you can still try selective eq-ing and stereo cancelation, yet even with the most complex algorithms very little sound data can be actually restored to an acceptable quality if the sample has been recorded poorly.

Answer 3

Firstly mp3 is a LOSSY format - meaning that the compression algorithm will remove information from the incoming file in order to achieve data compression. When doing audio restoration work, you always want to work with a Non-Lossy raw file format, such as WAV or AIF. That way you are working with as close to the original material as possible.

There is certainly a possibility that you have removed useful information from the AMR file when converting to MP3.

My advice would be to go back to the original AMR file - convert it to WAV and then go from there.

Audacity is a nightmare to work with - if you can get hold of anything else - your life will be easier. Try working with Reaper - this is a non-destructive DAW. Izotope RX is a good tool for audio restoration, but there are other very good tools out there as well - it's always horses for courses and the tools you need will always depend on the sort of material you are working with.

IF this is just a transcription job, then recovering the original audio may not be an issue so long as you can hear the words. If you do need to recover the original audio and you have a reasonable budget, look at Izotope, CEDAR, Waves WNS.