> What you did here is a simple noise gate that removes or attenuates
> any sound components below a certain (frequency-dependent)
> threshold.
Raimund,
As noise gates go, the algorithm in Audacity 3.1 is one of the more
sophisticated. It does more that a simple expansion with a threshhold.
It seems to subtract the sample spectrum even over the threshhold
level.
> In theory, this mechanism will however introduce artifacts! Imagine
> you had recorded a distant bird call that is partially buried in the
> system noise. Then
In practice all digital manipulation introduces artifacts, as with mp3
compression which we have got used to. The trick is to smother them
and the easiest way is to add back some of the original track. I
"noise remove" to -24 and add back about -12, fine adjusted by ear.
What seems to be happening with my mic hiss experiments is that the
level is so low that the artefacts are inaudible, or that they are
hiss-like and inaudible in practice, or both. From my listening tests
the hiss is removed from the whole recording, including a recording of
hiss. As I said, I checked out the hiss removal at slower speeds so I
could clearly hear the ultrasonic hiss go. On straight a-b listening
checks, I could hear no difference in quality but there was a
fractional waveform change.
What is essential is that the noise sample must be taken from the
quietest part of the track or the artefacts and "gravel" are bad. One
second of background (without any of the object sounds) is enough. It
is also important to filter out as much of the lower frequencies as
you can get away with before taking a sample.
I record at a constant level setting about -15 below peak, so I'm
going to try to de-hiss some existing recordings using my "underbed"
sample. If if doesn't work, I'll let you know. :-)
David
David Brinicombe
North Devon, UK
Cogito cogito ergo cogito sum - Ambrose Bierce
|