Topic: prefered dB level for Normalizing

[Denes]

I am using Audacity to process my recordings.  The default Normalization level is -0.5dB.  Should I be using this, or just use 0dB.  I am also normalizing my left and right channel independently.

[robeti]

I use -0,1

[page]

normalize together (otherwise it adjusts your balance between channels).

I bring it up to -0.3 or -0.4 and have a plugin that detects intersample clipping. There is some DAC equipment that will clip at lower than 0dbfs is my reasoning (The first Grace Designs headamp is an example I'm familiar with but there are others at various price points). Plus, there isn't any realistic benefit in pushing all the way to 0.

[greenone]

I bring it up to -0.3 or -0.4 and have a plugin that detects intersample clipping. There is some DAC equipment that will clip at lower than 0dbfs is my reasoning (The first Grace Designs headamp is an example I'm familiar with but there are others at various price points). Plus, there isn't any realistic benefit in pushing all the way to 0.

This is the first time I have ever heard a rational reason for anything less than 0. THANK YOU. I think I've asked at least a half-dozen times in other threads and haven't heard anything like this...

[page]

I bring it up to -0.3 or -0.4 and have a plugin that detects intersample clipping. There is some DAC equipment that will clip at lower than 0dbfs is my reasoning (The first Grace Designs headamp is an example I'm familiar with but there are others at various price points). Plus, there isn't any realistic benefit in pushing all the way to 0.

This is the first time I have ever heard a rational reason for anything less than 0. THANK YOU. I think I've asked at least a half-dozen times in other threads and haven't heard anything like this...

courtesy of GS, they pointed it out and then I started playing with the various DAC chips I had accessable. I found that my 901 (when I had one) wanted to generated a clip sound around -0.2 or -0.1. I'd have to look for the thread on GS to see just how low some folks had to go before their stuff didn't clip, but it seemed to even out above -0.5 so I keep it in that range.

The catch is it's actually possible to generate a signal higher than what your dbfs reading is. I remember reading it on the Izotope Ozone manual as to why they have a "detect intersample overs" option. I'm not sure how you would figure that stuff out manually though.

[Gutbucket]

The catch is it's actually possible to generate a signal higher than what your dbfs reading is. I remember reading it on the Izotope Ozone manual as to why they have a "detect intersample overs" option. I'm not sure how you would figure that stuff out manually though.

I don't think you can figure it out manually, at least not simply, which is why normalizing to something like .5 dB below full scale is good practice since that leaves sufficient headroom to accomodate possible intersample overs without compromising loudness in any significant way.  Which is what you just said. 

Techie speculation, it's been a while since I looked into this- But as I recall, intersample overs occur on energetic transients near the bottom edge of the transition band where there are minimal sample points per cycle and those samples land to either side of the to-be-reconstructed analog waveform peak.  I'd guess that intersample-over detectors first oversample the signal, then check the new samples for values over 0dBFS, indicating analog peaks larger than the DAC may be able to reconstruct but which go undetected at the actual sample rate being used as the samples on either side of the analog peak which are present at that rate are indeed =< 0dBFS.

[alpine85]

A couple more reasons NOT to go all the way to 0dB:

1) You want to do all your editing (including normalization) in the highest quality sample rate and bitrate, and levels can fluctuate in the resampling and bit depth conversion processes.  If you normalize to 0 and then resample/convert bit depth you may end up with some clipping. 

2) The MP3 encoding process can change the levels too, so if your recording is ever going to be converted (via the LMA or whatever), you will probably end up with clipping if you go all the way to 0dB.

Usually normalizing to anywhere from -0.2 to -0.5 will leave enough headroom for all this stuff, and like the others have said, won't really affect the perceived loudness

[capnhook]

thanks alpine85....i will put that into practice.

[Denes]

I actually notcied some clipping.  Also, even though I normalized to 0db, I did notcie when playing on audacity, it showed those red bars when it went over 0db.

Thanks for the input.

[runonce]

2) The MP3 encoding process can change the levels too, so if your recording is ever going to be converted (via the LMA or whatever), you will probably end up with clipping if you go all the way to 0dB.

Wouldn't the codec be smarter than than that? I would be surprised if this is the case...

[Todd R]

A couple more reasons NOT to go all the way to 0dB:

1) You want to do all your editing (including normalization) in the highest quality sample rate and bitrate, and levels can fluctuate in the resampling and bit depth conversion processes.  If you normalize to 0 and then resample/convert bit depth you may end up with some clipping. 

2) The MP3 encoding process can change the levels too, so if your recording is ever going to be converted (via the LMA or whatever), you will probably end up with clipping if you go all the way to 0dB.

Usually normalizing to anywhere from -0.2 to -0.5 will leave enough headroom for all this stuff, and like the others have said, won't really affect the perceived loudness

I've been normalizing to -0.2dbFS for both of these reasons, but hadn't heard of the DAC issue.  Reading more on it here though, I think I might start using -0.4db.

I did have problems with clipping when encoding to mp3 when I used to normalize to 0dbFS.  It was using the LAME encoder, but I don't know what version it was back then.

[JasonSobel]

There's a good discussion of inter-sample peaks on pages 24-26 pf the AD2K+ manual:
http://www.benchmarkmedia.com/sites/default/files/documents/ad2402-96-manual.pdf

As that manual is for the AD2K, it focuses on the A/D side of things, rather than the D/A during playback.  But it has a nice graph and the general concept is the same, so I think it's applicable to this discussion.

[silentmark]

I've been normalizing to 0dB for most of my recordings and have never gotten any clipping, shrug ...

[newplanet7]

Come to think of it I don't think I have a set threshold that I always go to.
What's great in Wavelab and I am sure many other programs is, scanning for the highest peak.
I then find those particular peaks and depending on the severity either reduce gain or just bust
out ye ole pencil tool and redraw.
After that I peak normalize the difference from the peak scan to a safe level.

[vanark]

I've been normalizing to 0dB for most of my recordings and have never gotten any clipping, shrug ...

I think the conversation here has been that it may clip on certain playback equipment, not all.  I've never heard an audible clip on my playback, but am willing to use a less than 0 dB peak if it helps out some others with more sophisticated playback set ups.

[silentmark]

Interesting, thanks !

[runonce]

Sounds like I need to update my Audacity workflow...

In the early versions - Normalize stopped at -3db...so I always used Amplify (which defaulted to 0)

Sounds like they've revised the normalize function - and with a safe preset...

Note to self...

This is almost merits an alert somewhere - I think a lot of us who upload to the LMA probably still go to 0 (former conventional wisdom)

[cashandkerouac]

i set my peak at -0.5db for the reasons already stated.  however, i prefer using the hard limiter over the normalization function.