[LAU] Improving a bad recording

Kevin Cosgrove kevinc at cosgroves.us
Tue Jun 1 23:59:30 UTC 2010


On 2 June 2010 at 0:10, Philipp <hollunder at lavabit.com> wrote:

> It would be nice if I could at least improve the recording, but
> I don't quite know how to start.  My only 'source' file is a
> single mp3.

Bummer.  But, ya can still learn.

> The major problems of the recording are:  1) quite a lot too
> much bass of the acoustic guitar - this is easy to correct
>
> 2) a very loud noise, possibly from some part of a laptop. To
> my knowledge the recording was made in two takes, guitar and
> voice, on a laptop with a cheap mic of some sort. The noise
> seems to be located mainly somewhere around 12kHz, a region
> which seems to be important for the clarity of the voice. Any
> attempts to just pull down with an eq (4-band parametric for
> example is what I tried today) had a significant impact on the
> voice.
>
> 3) a constant level of noise, probably from some part of the
> cheap equipment used. Due to the two takes it's 'doubled' for
> the most part of the song.

Is there any part of the sound file that has only the noise
present, without the music playing?  The more the better; 5-10
seconds would be good, and 1-2 seconds might be adequate.  If
so, then you could try the noise removal effect in Audacity.
I use it often and I'm quite happy with it.  I would try the
noise removal first, before compounding the recording with other
effects.

Audacity at version 1.3.11 (probably before then, but certainly
after 1.3.0) has a good graphic equalizer in it.

Parametric EQ might help to really "zoom in" on some things.

> To tip it all off, I can't even make out all lyrics. It might
> be that some significant hints are simply masked by the noise,
> or they simply are hard to make out.

After the noise reduction, then you might try compression or
expansion to isolate various sounds.  Frequency sensitive
compression or expansion might also help, but is more difficult
to use.

> I'd really love to somehow get this recording to an enjoyable
> state, no matter how involving it is. Given the source it seems
> next to impossible, but it has to be possible to make at least
> some improvements.

One thing you might try, which I haven't tried, is to make
several passes at the file.  On each pass you try to isolate one
instrument or vocal, making that sound as good as you can, while
trying to eliminate all the other sounds.  Saving a collection
of those kinds of sound files might allow you to mix those
intermediate files back together to produce something useful.

Is this recording stereo, with things panned somewhat left and
other things panned somewhat right, and yet other things panned
to center?  If so, then you might be able to get at yet more
aspects of the recording by subtracting right from left, or visa
versa.

Hmm, you might also websearch for "audio forensics" or "audio
restoration" or similar topics as I'm sure there are folks who've
been farther down this path.

Please report back on how it goes.  I'll look for your report in 7 
years.  ;-)

Good luck!

--
Kevin




More information about the Linux-audio-user mailing list