PDA

View Full Version : Residual birdies after noise reduction


Paddy
September 28th 03, 03:46 AM
I have been digitising some of my treasured vinyl and cassette collection.
The cleanup steps that I use are:
1 Manually remove large transients.
2 DC removal
3 Click and crackle removal (a Sonic Foundry DX plugin)
4 Normalise the whole album
5 Noise reduction (a Sonic Foundry DX plugin)
6 Listening and manual touch-ups.
7 Maybe employ some DFX for bass and treble recreation.
8 Cut into tracks.

The noise reduction involves finding a suitably long
silence, either between tracks or from the beginning
or end of the file and sampling the noise spectrum,
then applying the noise reduction to the whole album.
Sometimes it is necessary to do the left and right
channels separately, and sometimes there are problems
with poor source material having earlier noise reduction
that results in noise being 'pumped' by the signal.
But those are not my main problems.

My main problem is with low level recordings where
the signal fades into the background noise. In these
cases after noise reduction I will be able to hear
'birdies' or tinkelling noises as the signal fades
away. I cannot figure out how to do the noise
reduction differently so that the residual noise
has a white-noise spectrum or is made to not
be audable when the signal is very low.

Squelching the audio at levels below where the
birdies become audable is not always possible.
In some cases a great deal of the album is in
this category.

I've tried other NR plugins, but the SF one seems
the best, and morover, it is the one that I have.

Any advice from the more experienced?

Paddy

Arny Krueger
September 28th 03, 12:11 PM
"Paddy" > wrote in message
om
>
> My main problem is with low level recordings where
> the signal fades into the background noise. In these
> cases after noise reduction I will be able to hear
> 'birdies' or tinkelling noises as the signal fades
> away.

Sounds like truncation error due to a digital attenuation without proper
dither or something like it.

> I cannot figure out how to do the noise
> reduction differently so that the residual noise
> has a white-noise spectrum or is made to not
> be audible when the signal is very low.

You probably won't, since this is one of those major unsolved issues in
audio, if my diagnosis is correct. The working solution is to dither
attenuation properly in the first place. The noises that are generated are
predictable if you know all the details about the failing attenuation, but
complex in nature regardless.