In case I didn't make myself clear, normalizing is the final step in the editing and restoration process. The entire file is normalized after the individual tracks have been matched so they have the same apparent loudness. I assume that, like me, when you do an initial transfer, you do not set record levels so the peaks hit 0 dBFS. I leave myself 6 dB or so of wiggle room to allow for changes in level due to various signal processing that may be needed.
When I was editing my live concert recordings, I would generally leave the levels of the individual pieces alone, unless one was way out of whack relative to the others. Sometimes our wind ensemble will begin their concert with a chamber work, which is much softer than the remainder of the program. In that case I would raise the level of that work to make it more reasonable relative to the rest of the concert. But, generally, I liked to leave the changes from one piece to the next, in a live concert, unaltered.
Audio Engineer Emeritus
The Crane School of Music
SUNY at Potsdam, NY 13676
"Great art presupposes the alert mind of the educated listener."
"A true artist doesn't want to be admired, he wants to be believed."
From: Association for Recorded Sound Discussion List [mailto:[log in to unmask]] On Behalf Of John Haley
Sent: Sunday, March 04, 2018 4:15 PM
To: [log in to unmask]
Subject: Re: [ARSCLIST] RX5, etc.
OK. I stand corrected. But let me "reinterpret" here. There is nothing "normal" about raising the level of every track "to the max" and thinking you have accomplished a good thing. A sound file really needs to sound well in relation to to its neighbors, for listening purposes (as opposed to simple archiving). Where the tracks are parts of the same piece of music, we usually want the level to remain consistent for that piece of music, meaning softer tracks will sound softer, etc. That is, unless the goal is to achieve some kind of "overall compression," where everything is at the same level (for a disco party?). I don't think this issue is much of a problem, as it is so fast and easy to raise or lower the level of a track to where you want it. For unrelated tracks, I always listen to them in context, so one does not jump out or seem to disappear. You have to hear this--looking at the computer screen doesn't really tell you what you need to know about the "apparent" loudness, particularly where any compression has been applied. A compressed track will always sound louder than it looks.
On Sun, Mar 4, 2018 at 2:24 PM, Gary A. Galo <[log in to unmask]> wrote:
> I lean towards Lou's definition rather than John's. When a file or
> selection is "normalized", the program scans the file or selection and
> then raises the level of the entire file or selection to bring the
> loudest peak or peaks to maximum level. Most engineers I know do not
> bring the loudest peaks to 0 dB, but rather leave a little headroom.
> Some normalize to -0.1 dB. I'm a bit more conservative - I normalize to -0.5 dB.
> When you go through a file and match apparent levels from one track to
> the next, not every track will be normalized to achieve this. Under
> the Tools menu, Sound Forge has a useful gadget called Statistics,
> which give you a slew of level information on your file or selection.
> I find that RMS level is useful for matching levels from one track to
> the next, but the ear is the final arbiter, of course. But, remember,
> level matching is not normalizing.
> I don't understand Paul's view that normalizing is bad practice.
> Benchmark makes a big deal about "intersample overs", which they
> eliminate in their
> DAC2 and DAC3 converters by reducing level of the incoming datastream
> 3.5 dB. But, they also point out that intersample overs are not likely
> to be a problem at sampling rates above 48 kHz, so if you're recording
> at 96 kHz, you should not have to worry about them. All normalizing
> should be done with your file still at high res - 96/24 or higher.
> When ALL work is done, resample to 44.1 if you're going to make CDs,
> and convert to 16 bits as the very last step. That way everything you
> do will have been done with
> 24 bits of resolution. Following these practices, I fail to see how
> normalizing is bad.
> Gary Galo
> Audio Engineer Emeritus
> The Crane School of Music
> SUNY at Potsdam, NY 13676
> "Great art presupposes the alert mind of the educated listener."
> Arnold Schoenberg
> "A true artist doesn't want to be admired, he wants to be believed."
> Igor Markevitch
> -----Original Message-----
> From: Association for Recorded Sound Discussion List [mailto:
> [log in to unmask]] On Behalf Of John Haley
> Sent: Sunday, March 04, 2018 1:58 PM
> To: [log in to unmask]
> Subject: Re: [ARSCLIST] RX5, etc.
> Sorry, Lou, not the way I have seen "normalized" used. Perhaps that
> is related to what you are saying, if one views the exercise as
> raising all the tracks to their "max." It is about balancing the
> relative level among a number of tracks,not just raising one track to its max level.
> On Sun, Mar 4, 2018 at 1:40 PM, Tim Gillett
> <[log in to unmask]>
> > The "swish" energy is possibly full of highs and the loudest thing
> > in the recording, which is why a standard denoiser wont touch it.
> > It's looking to reduce low level sounds.
> > The swish will also probably contain frequencies way above that of
> > the wanted program, as well as above human audibility.
> > For access, I'd declick and then probably subjectively filter out a
> > lot of those highs, and even lows, but without an audio sample hard
> > to
> be sure.
> > Tim
> > ----- Original Message ----- From: "Paul Stamler" <[log in to unmask]>
> > To: <[log in to unmask]>
> > Sent: Monday, March 05, 2018 2:18 AM
> > Subject: Re: [ARSCLIST] RX5, etc.
> > On 3/4/2018 12:10 PM, Lou Judson wrote:
> >>> It is also possible that the clipping sounds are from overloading
> >>> the D/A, whilst the waveform is okay. It is called ï¿½intersample
> >>> peaksï¿½ and one reason I avoid normalizing. Try normalizing to -1
> >>> or -2 and see if it still sounds bad. Or, as I said, use a
> >>> look-ahead limiter, again instead of normalizing!
> >> Yet another reason why normalizing is generally a bad policy.
> >> Peace,
> >> Paul
> >> <L>
> >>> Lou Judson
> >>> Intuitive Audio
> >>> 415-883-2689
> >>> On Mar 4, 2018, at 10:02 AM, Tim Gillett
> >>> <[log in to unmask]>
> >>> wrote:
> >>> The crackling noises after normalising sound like clipping. You
> >>> could
> >>>> visually inspect (by magnifying) the waveform peaks both before
> >>>> and after normalising. Have you tried normalising but minus a few db's?
> >>>> Tim
> >> ---
> >> This email has been checked for viruses by Avast antivirus software.
> >> https://www.avast.com/antivirus