Normalization methods for LP digitization

[sarumbear]

Essentially, we're talking about 2 alternatives.

1. After recording, shift the amplitude of the digital recording (the entire LP, not track by track) so it peaks at -1 dB. (or -0.5). Safe the shifted digital file.

2. After recording, scan the digital recording to find the average & peak levels, then create metadata for replay gain, so it can shift the recording level on playback.

Method 1 is guaranteed transparent but is more work. Method 2 is less work, but may or may not be transparent, depending on the implementation of replay gain used during playback.

You can’t have metadata on a WAV file. Even if you attached an ID3 tag to the file, it will be lost when re-saved by most editors.

 

[Lttlwing16]

Normalisation of the whole album is volume control. Loudness matching track to track is process. Allow me to eloborate.

Streaming and broadcast has limited dynamic range capacity to work on suboptimum listening environments. A classical piece has a crest factor around 17dB. Quite passages are 17dB lower than the high passages. Imagine that level of difference in a car. You have to constantly adjust the volume to hear every passage at the same perceived level. At a domestic audio system noise is much less and such level difference works OK. Piano parts play at a low level, forte parts at a high level. We feel the difference in level as the composer wanted.

On streaming and broadcast those different parts must be loudness matched. If you do the same while archiving you are effectively changing the master. Naturally, this doesn't apply to modern pop music but as we are talking about vinyl digitising I expect modern pop is not the criteria here.

It is important to differentiate between broadcast, recording and mastering.

Hey, thanks for taking the time to elaborate.

I would consider any digital process that changes the initial data, processing. In Audacity at least, normalization does just this. It takes the digital data and adjusts it to a new data set, different than the original. With any adjustment of the data set, errors can be introduced.

I would consider volume control, the act of increasing or decreasing the capture amplitude. With this process, the risk of introducing error is essentially null, as long as the level was set appropriately.

So perhaps just a difference of vernacular.

I do think you raise a good point. Loudness normalization is a tool for the mastering process, not the recapturing/archival process. Question is does the vinyl mastering process (if done separate from the digital master) include loudness normalization? I would think not, certainly for older releases, but perhaps the mastering engineer for modern releases wants to have the perceived loudness match that of human hearing as laid out in the Fletchner-Munson curves with the tools available in the EBU-R128 standards.

However, when I look back at the tests I ran last night, it sure seemed the loudness normalized track was a much closer match to the peak/non-normalized captured track.

Either way I think it's safe to say if capturing at as close to peak as possible, with no other adjustments apart from track separation and exporting as FLAC, would offer the most digitally transparent process, barring the ADC isn't introducing audible noise/distortion.

 

[MRC01]

... Loudness normalization is a tool for the mastering process, not the recapturing/archival process. Question is does the vinyl mastering process (if done separate from the digital master) include loudness normalization? I would think not, certainly for older releases, but perhaps the mastering engineer for modern releases wants to have the perceived loudness match that of human hearing

Even old vinyl from the 60s was mastered with dynamic range compression in order to fit wide dynamic range material, like orchestras, into the limited range of vinyl, to lift the quiet parts above the surface noise while avoiding distortion on the loudest parts, which are often near the lead-out groove where vinyl has the worst dynamic range. They did other processing to optimize for the limitations of vinyl, like summing bass to mono. And of course the RIAA emphasis curve. The difference is, back then they used a much lighter hand with dynamic compression than is often used today with digital recordings.

Either way I think it's safe to say if capturing at as close to peak as possible, with no other adjustments apart from track separation and exporting as FLAC, would offer the most digitally transparent process, barring the ADC isn't introducing audible noise/distortion.

Yep. However, digitally recording vinyl at a lower level like -6 to -12 dB, to avoid overload, using 24-bit to avoid loss of resolution, then shifting it in post-processing up so the peaks are around -1 dB, is completely transparent. I use the word "shifting' carefully, to mean literally shifting the amplitude up, like a digital volume control, no dynamic compression. After doing all that, you can even convert it to 16-bit without loss of information, which is convenient since then you can make a CD from it.

 

[solderdude]

normalization does just this. It takes the digital data and adjusts it to a new data set, different than the original. With any adjustment of the data set, errors can be introduced.

normalization = automated volume control.
The program looks at the max. peak it can find and then calculates each sample to the same amplitude difference.
It is no different than volume control and does not change anything but the volume to a preset peak level.
That is ... if you set it to detect peak level, not RMS level.

 

[levimax]

I sometimes think I should digitize my LP collection but besides all the work involved, especially splitting and tagging the tracks, I don't do it because once "digitized" I am stuck forever listening to the TT and Cart and set-up I had at the time. Part of the fun of LP's is getting a new cart or stylus or TT and listening to your LP collection with a different perspective.

 

[sarumbear]

I think it's safe to say if capturing at as close to peak as possible, with no other adjustments apart from track separation and exporting as FLAC, would offer the most digitally transparent process, barring the ADC isn't introducing audible noise/distortion.

What about the clicks and pops which are 6+dB over the music peak?

 

[sarumbear]

Even old vinyl from the 60s was mastered with dynamic range compression in order to fit wide dynamic range material, like orchestras, into the limited range of vinyl, to lift the quiet parts above the surface noise while avoiding distortion on the loudest parts, which are often near the lead-out groove where vinyl has the worst dynamic range. They did other processing to optimize for the limitations of vinyl, like summing bass to mono. And of course the RIAA emphasis curve. The difference is, back then they used a much lighter hand with dynamic compression than is often used today with digital recordings.

That recording I mentioned elsewhere, Harry Belafonte Live at Carnegie Hall, was recorded in 1959 before the first studio quality compressors (UA-1765 & 176) were available. No compression, no Dolby, a simple mixer feeding directly to an Ampex 601.

 

[Lttlwing16]

What about the clicks and pops which are 6+dB over the music peak?

I think that's a fair point against peak recording. My records get a thorough clean before playback and recording, so I typically don't run into any that loud. I have some crystal quiet LP's, and some not so quiet LP's. Selective utilization of peak recording would have to be used. Since one would have to screen the LP for peak transient to peak record, simply record around -6dbFS, and if there are pops and clicks that prohibit peak recording, you have the capture for normalization method.

Perhaps @daftcombo could comment how he handles it since I believe he prefers peak recording.

normalization = automated volume control.
The program looks at the max. peak it can find and then calculates each sample to the same amplitude difference.
It is no different than volume control and does not change anything but the volume to a preset peak level.
That is ... if you set it to detect peak level, not RMS level.

certainly a well understood process and obviously utilized and suggested by many reputable folks.. that said, I probably didn't make my point properly..it's still a digital modification to the original bit data. Which means you're at the mercy of the program doing the normalization, just as the quality of the capture is dependent on the quality of the ADC that's capturing. It's a process, void of A/B comparison may or may not be hurting the capture. In A/B comparisons of a couple different tracks from different LP's, I can tell the difference between the peak capture and the normalized file.

Let me elaborate on my setup to help shed some light.

I record LP's in Manjaro Linux, and edit using Audacity. Unfortunately, the fully functional version of Audacity in my distro is stuck at 1.4.2. The current release of Audacity is 3.x, which I cannot use on my pc. Who knows if the normalization algorithm for 3.x has been improved since version 1.4.2? Perhaps it's not changed over two full version updates, but perhaps the software as better all together, including the normalization.

I also have a Windows XP partition on the same PC which I use for Pro Tools for recording my own music. However, this is also stuck at version 8 (i.e. ancient) as that was the last version of software for my original Mbox. I'm yet to try to normalize the same file in Pro Tools and compare it to the Audacity version, but that is something I'll do today. I've heard Pro Tools has it's own "sound" so there's that too.

 

[daftcombo]

Perhaps @daftcombo could comment how he handles it since I believe he prefers peak recording.

I wrote exactly the contraty: I indicated that I set the ADC input gain so that the loudest record in my collection peaks at -3dBFS. So, most of the time, with other records, I am around -9dBFS or -12 dBFS.

Anyway, I remove the pops and clicks manually. Even if there's clipping on some of them, it can be corrected the same way.

 

[rdenney]

normalization = automated volume control.
The program looks at the max. peak it can find and then calculates each sample to the same amplitude difference.
It is no different than volume control and does not change anything but the volume to a preset peak level.
That is ... if you set it to detect peak level, not RMS level.

Be careful--"automatic volume control" sounds a lot like "automatic gain control" as used on cheap tape and MP3 recorders back in the day. Those were gain-riding strategies that resulted in pumping and loudness compression--just what we don't want.

And it's not automatic, at least not in Vinyl Studio. After recording the needledrop, I can set the normalization ceiling, check the box that keeps both sides of an album at the same level, and press the "normalize" button. It then moves everything up equally until the peak in the file reaches digital -0.2 dB (or whatever the ceiling is). I don't have to adjust a volume slider myself, so in that sense it's automatic. But really it happens at the data level.

Rick "gain-riding: the original Loudness War strategy" Denney

 

[Lttlwing16]

I wrote exactly the contraty: I indicated that I set the ADC input gain so that the loudest record in my collection peaks at -3dBFS. So, most of the time, with other records, I am around -9dBFS or -12 dBFS.

Anyway, I remove the pops and clicks manually. Even if there's clipping on some of them, it can be corrected the same way.

Apology, I guess I was thrown off by your statement on the first page

"I think the best practice is: "capture at as close to -1.0 to -0.5 dBFS, no normalization", adjusting the volume analogically at the output of the preamp or at the input of the ADC, or both"

 

[rdenney]

What about the clicks and pops which are 6+dB over the music peak?

Vinyl Studio will allow some tiny percentage (it's a setting) to exceed the ceiling. But I do try to deal with clicks and pops before normalization.

Rick "power tools, but like all power tools, can be wayward if mishandled" Denney

 

[rdenney]

Apology, I guess I was thrown off by your statement on the first page

"I think the best practice is: "capture at as close to -1.0 to -0.5 dBFS, no normalization", adjusting the volume analogically at the output of the preamp or at the input of the ADC, or both"

The consequences of error in identifying the peaks can be really hard to listen to--digital clipping is hard. Even if I listen all the way through to find the peaks, I still might miss the very loudest bit. When we were recording tapes, the occasional missed peak would be smoothly compressed and probably not that audible, but the challenge to getting above the noise floor was more important with tape, so it was worth the risk. With a noise floor at -96 dB or better, that's no worry, so some safety margin above the peaks seems much less time-consuming and risky.

Rick "best practice = no mistakes for the least effort" Denney

 

[daftcombo]

"I think the best practice is: "capture at as close to -1.0 to -0.5 dBFS, no normalization", adjusting the volume analogically at the output of the preamp or at the input of the ADC, or both"

Best practice but definitely not the most convenient IMHO, though I know someone who does that and has ripped 1000+ records already.

 

[danadam]

Who knows if the normalization algorithm for 3.x has been improved since version 1.4.2? Perhaps it's not changed over two full version updates,

1.4.2? According to https://wiki.audacityteam.org/wiki/Audacity_Versions they went from 1.3.14 to 2.0.0.

 

[MRC01]

Best practice but definitely not the most convenient IMHO, though I know someone who does that and has ripped 1000+ records already.

If that wasn't me, then I resemble that!

 

[solderdude]

I think that's a fair point against peak recording. My records get a thorough clean before playback and recording, so I typically don't run into any that loud. I have some crystal quiet LP's, and some not so quiet LP's. Selective utilization of peak recording would have to be used. Since one would have to screen the LP for peak transient to peak record, simply record around -6dbFS, and if there are pops and clicks that prohibit peak recording, you have the capture for normalization method.

Perhaps @daftcombo could comment how he handles it since I believe he prefers peak recording.

certainly a well understood process and obviously utilized and suggested by many reputable folks.. that said, I probably didn't make my point properly..it's still a digital modification to the original bit data. Which means you're at the mercy of the program doing the normalization, just as the quality of the capture is dependent on the quality of the ADC that's capturing. It's a process, void of A/B comparison may or may not be hurting the capture. In A/B comparisons of a couple different tracks from different LP's, I can tell the difference between the peak capture and the normalized file.

Let me elaborate on my setup to help shed some light.

I record LP's in Manjaro Linux, and edit using Audacity. Unfortunately, the fully functional version of Audacity in my distro is stuck at 1.4.2. The current release of Audacity is 3.x, which I cannot use on my pc. Who knows if the normalization algorithm for 3.x has been improved since version 1.4.2? Perhaps it's not changed over two full version updates, but perhaps the software as better all together, including the normalization.

I also have a Windows XP partition on the same PC which I use for Pro Tools for recording my own music. However, this is also stuck at version 8 (i.e. ancient) as that was the last version of software for my original Mbox. I'm yet to try to normalize the same file in Pro Tools and compare it to the Audacity version, but that is something I'll do today. I've heard Pro Tools has it's own "sound" so there's that too.

Normalization is an extremely simple and straight forward process that cannot change the sound. Only the amplitude in a digital way.

The technical process is very simple. The file is scanned for the highest amplitude (both negative and positive).
That peak is higher or lower than the target value.
The target value is the set peak value.
Say the found peak value is -5.72dB and the target value is -1.03dB.
Where 0dB = factor 1.
-5.72dB = 0.517507
-1.03dB = 0.888178
This means that peak sample needs to be amplified by 1.716263 and so do all other samples in the entire file with the exact same value.
This is super easy for any program. Of course there will be some rounding-off of values when it needs to be in a 16 bit or (much smaller deviations) in 24 bit or 32 bit.
NO sound will ever be changed during this extremely simple bit value calculations. There will also not be any filtering or looking at previous or next values involved. It only appears that way because the overall volume changed.

The effect (and rounding off errors) will be exactly the same as when digital volume is used or the actual recording was done at a slightly higher or lower level. Calculations can even be done in 64 bit and then truncated (or dithered) to 16 or 24 bit.

 

[daftcombo]

Normalization is an extremely simple and straight forward process that cannot change the sound. Only the amplitude in a digital way.

The technical process is very simple. The file is scanned for the highest amplitude (both negative and positive).
That peak is higher or lower than the target value.
The target value is the set peak value.
Say the found peak value is -5.72dB and the target value is -1.03dB.
Where 0dB = factor 1.
-5.72dB = 0.517507
-1.03dB = 0.888178
This means that peak sample needs to be amplified by 1.716263 and so do all other samples in the entire file with the exact same value.
This is super easy for any program. Of course there will be some rounding-off of values when it needs to be in a 16 bit or (much smaller deviations) in 24 bit or 32 bit.
NO sound will ever be changed during this extremely simple bit value calculations. There will also not be any filtering or looking at previous or next values involved. It only appears that way because the overall volume changed.

The effect (and rounding off errors) will be exactly the same as when digital volume is used or the actual recording was done at a slightly higher or lower level. Calculations can even be done in 64 bit and then truncated (or dithered) to 16 or 24 bit.

Like those who want "bit-perfect", some "purists" want records to be digitalized with the gain adjusted digitally and no further processing. Even track splitting should leave blanks between tracks as is.

 

[MRC01]

... there will be some rounding-off of values when it needs to be in a 16 bit or (much smaller deviations) in 24 bit or 32 bit. ... Calculations can even be done in 64 bit and then truncated (or dithered) to 16 or 24 bit.

Like those who want "bit-perfect", some "purists" want records to be digitalized with the gain adjusted digitally and no further processing. Even track splitting should leave blanks between tracks as is.

Those people may not realize that this rounding off and dithering is transparent, as these tiny differences in amplitude are far smaller than the resolution of the vinyl being recorded. Way below the level of tape hiss, groove noise, etc. They might as well worry about a piece of dust floating in the room landing on top of the tonearm while it was playing.

 

[sarumbear]

I probably didn't make my point properly..it's still a digital modification to the original bit data. Which means you're at the mercy of the program doing the normalization, just as the quality of the capture is dependent on the quality of the ADC that's capturing. It's a process, void of A/B comparison may or may not be hurting the capture. In A/B comparisons of a couple different tracks from different LP's, I can tell the difference between the peak capture and the normalized file.

Let me say it very clearly: Digital gain is not an audible process.

You cannot hear difference between digital gain and analogue gain if you are not changing the dynamic range of the captured signal. If you are, then you are either a) not matching gains correctly between digitally applied gain version and analogue gain version when comparing them or b) you are not capturing the full dynamic range of the programme.