Volume Normalization

[David W]

For the last few months, I have begun a project of ripping ~1000 LPs. Here is the process that I developed:

• Clean the record using Spin Clean
• Rip the record using VinylStudio saved as AIFF, 192kHz
• Add album artwork largely from Album Art Exchange
• Mildly declick the tracks using Click Repair following the settings under post *13
• Convert the AIFF format to FLAC using dBpoweramp Music Converter

I recently ripped a clean copy of Abbey Road from the LP as a hi-res rip (MFSL version) and then compared it to an earlier CD rip. The LP rip appears to be more muted while the CD has a crisper sound. I've also experienced the same when comparing other LP to CD rips. Abbey Road was a real clean copy and that is the one I'm using for this comparison.

I have been advised to normaize the volume in the ripped copy to give it more punch. I can do that for each album in VinylStudio and then re-run each of the following steps above, or I can do so in bulk using "Volume Normalize" in dBPoweramp (much more attactive). I ran on test (Peak to Peak) but it did not boast it enough. But really at this point, I'm somewhat flying blind as I'm not sure about the other features. Any gudiance on optimal settings?

Thanks.

[Spoon]

Re: Volume Normalization

Look at the dsp effect 'dynamic range compression'

[David W]

Re: Volume Normalization

Thanks, Spoon.

I may have a no win here. I've been researching this over the course of the day and any post-rip corrections I do will, as I understand it, negativly impact my recordings. All of my tracks were done as lossless FLAC rips from LPs at 192K/24 bit. Are there any corrections I can do in boosting the volume without any negative impacts? My review and test of using dynamic range compression out of the box further compresses the sample track. If this is the right solution, what are the settings I need?

I attached some Audacity screenshots comparing the options (so far).

[garym]

Re: Volume Normalization

Have you considered using a ReplayGain aware player for these files (there are many...all the players I use, LMS, squeezeboxes, foobar2000, Roon Labs). This way you could simply add ReplayGain tags (album and track) to the files and the player would use these to adjust playback volume, but the actual audio in the files would remain untouched. (ReplayGain tags are simply tags like any other metadata tag such as ARTIST, ALBUM, etc.). You can add RG tags in a bulk manner using dbpa and the ReplayGain utility DSP.

[David W]

Re: Volume Normalization

I'm not familiar with ReplayGain. Where can I get more info on options? What is the difference between ReplayGain and ReplayGain (Apply)?

Are there no other solutions to boost the recording level without impacting (as I understand it) the dynamic range?

[schmidj]

Re: Volume Normalization

Hello David,

It would be worthwhile to educate yourself concerning dynamic range as applied to digital recording and as applied to analog vinyl recordings, the term "headroom", the term "loudness", the term "normalization", how the "replay gain" tag is supposed to function, how dBpa's "Replay Gain (apply) works, and the basics of dynamics processing.

I'll attempt to explain in a not too technical manner, but it is quite a bit to chew on all in one post. And your source material, transfers from vinyl, probably deserve different treatment than bit-accurate rips of (digital) CD's.

To begin with, the dynamic range capability of media is the difference between the loudest sound that the media can store and the "noise floor" of the media. For digital media, this is dependent on the number of bits used to quantize the waveform. The upper limit is defined as 0 dBFS (FS stands for Full Scale). Think of it as a binary word with all "1"'s. The lower limit is determined by the number of bits in the sample, each additional bit adds 6 dB. CDs have 16 bits, common audio files may have 16, 24 or more bits per sample. In theory, that sets the noise floor for a 16 bit recording at -96 dB. It is actually more complicated because of the probable application of dither, the possibility of source identification by adding "inaudible" digital patterns to the source audio, the possible application of HDCD processing, etc. But all of that is way beyond the scope of this post.

Analog audio is more complicated. In the case of vinyl, the upper limit can vary widely, and is controlled in the cutting lathe depending on the spacing between the grooves (which may have been varied within the recording), the frequency being recording, as a recording equalization curve (think RIAA EQ) is used which especially boosts the loudness of the high frequencies during the mastering of the record, and decreases the loudness in your turntable preamp. This decreases the perceived noise, but significantly decreases the loudest recordible high frequency audio. Furthermore, the location on the record (the inner grooves have issues) and for bass, whether the signal is mono (equal in both left and right channels, or only present in one channel.

The noise floor is also hard to define. What kind of noise? Hiss, rumble, pops and clicks, and how is it measured? Also, are you concerned about the noise floor when the record is newly pressed, or after it has been played, and quite possibly damaged. Whatever the definition, most people would consider the potential dynamic range of records,particularly used records to be less than that of CDs.

Now we have to look at the dynamic range of the source material. Much pop material has limited dynamic range, jazz and classical usually more. Why is that? because of the use of dynamics processing of the audio content. In the days of records, almost every record had quite a bit of dynamics processing. The source material generally exceeded the difference between what on the loudest side would make the record unplayable on the consumer's turntable, and on the quiet end would be impaired by the noise of the vinyl (or of the analog tape used in the production of the music). And in the case of pop music, many producers wanted their recordings to sound "loud" because they were (are?) of the opinion that they would sell more records if the record was "loud". So they used a lot of compression and limiting (and a few other techniques) to reduce the dynamic range of their recording and cut the master with as loud a sound as they thought would still play without skipping or excessive distortion on people's turntables. But even jazz and classical vinyl recordings typically required a fair amount of dynamics processing to produce a pleasing product for the consumer.

In digital recording, things start out simpler. The loudest part of the recording cannot exceed 0 dBFS without distortion. And the quietest parts cannot go below the lowest level captured by the number of bits in the samples (an over simplification, but enough for now.) In many cases when CDs are made, a technique called normalization is used. In a computer, all the samples of a file are examined, to find the very loudest sample(s). Equal volume is added (or possibly subtracted) to every sample to make the loudest samples equal to a particular volume level, historically often 0 dBFS or some level slightly below this. Normalization like this does not change the dynamic range of the recording at all, it simply changes the overall volume during playback (assuming the playback device is not turned up into distortion or down into the noise floor.) Essentially, the same as adjusting the volume control on your playback device, but recorded into the media.

Now most digital recordings, particularly of pop music have still had a lot of dynamic processing before the normalization. Pop producers wanted their CDs or downloads to be loud, hopefully louder than the competition. Even many jazz and classical digital recordings have some dynamic processing, either on particular instruments prior to mixing or on the mixed results, to suit the tastes of the producer. And dynamic processing is by no means all bad, it can often make the results more pleasing to listen to, not just loud.

Now lets talk about "loudness". Loudness, as perceived by our ears is not determined by the peak audio level, as was used to determine the normalization point. It is perceived from the average level as presented in a much greater number of samples, the number of samples and amplitude varying depending on the frequency of the audio. While loudness processing has been discussed for generations, the real impetus came from complaints about "loud" commercials, particularly on TV. The producers of the commercials used the same processing the record producers used to make their pop records loud, and sometimes more. Since the audio processing used by most TV stations to comply with government modulation requirements for their transmitters had similar analog effects to normalization of digital material, the heavily compressed/limited commercials were perceived by our ears as much louder that the surrounding, less processed programming.

So various organizations (the EBU in Europe, the ASTC in the USA and some government agencies) came up with ways of measuring the "loudness" of digital audio, and many government television licensing agencies put limits on the loudness of commercials and sometimes programs that could legally be broadcast.

Now the loudness standards apply to TV and not (at least in the USA) to radio, or to recordings. But the same standards, to a degree, can be applied to recordings to make them play back at a consistent level on the consumer's equipment.

Now lets talk about replay gain. Replay gain is nothing more than a tag stored (usually) with the audio data the same as the title, the artist name, the composer and sometimes the artwork. It doesn't change the actual recorded audio samples in the stored media at all. What it may do, if (and it is a fairly big if) your playback device is replay gain compliant, is to automatically adjust the playback volume by the number of decibels (up or down) stored in the replay gain tag. It does not effect the dynamic range of the playback audio, only the volume. But there are some issues and potential problems. Since the replay gain adjustment is almost always applied in the playback device digitally, before conversion to the analog output, it would be quite possible to have a positive replay gain number drive the peaks of the audio "above" 0 dBFS. Since for the fixed point audio on a CD and most download files, this is impossible, the audio is "clipped", some peaks are shaved off. This may or may not be audible, but violates the definition of replay gain, that the dynamics of the source are not changed, only the volume. Therefore most replay gain calculators, including dBpa's, will not store a replay gain tag that would increase the level of the loudest samples into clipping.

Now we need to discuss how to calculate the replay gain. dBpa gives you several choices. Historically,it was an automatically "normalization" process. The software simply scanned every file and looked for the "loudest" sample, and calculated the number of dB which either increased or lowered the volume (gain) of the entire file to the target number of dB below 0 dBFS and stored that number in the tag for the playback device to deal with if so equipped. But while this was a first attempt at constant perceived listening levels, it wasn't very effective, as (as we described above) the human ear doesn't perceive loudness by momentary audio sample peaks.

With the development of the TV "loudness" algorithms (and all the variations are quite similar although the terminology and actual calculated numbers may be different) people realized that the loudness algorithm designed for TV could do a much better job of calculating the replay gain number than the old "loudest sample" method. And that in fact is the case. But there is a fly in the ointment if you actually read the standard and try to apply it as written. That is because digital audio levels are handled much differently in TV plants than on typical consumer CDs. Your TV production facility allows much more "headroom" of unused bits above the loudest sample than most CDs, where the headroom is zero or very small. CDs are much louder, on average, than the digital audio in a TV plant prior to broadcast. You will see a loudness number of -23 or -24 dB in the various TV standards. If you try to use that with consumer playback equipment, you may well find that you can't turn the volume up enough to make the playback comfortably loud. And if you can with your car radio, and then switch from the digital playback to the radio, you will be blasted with audio from the radio.

So various people have experimented with an appropriate loudness number for recorded files to play back on consumer equipment Various parties suggest numbers from -14 to -18 dB. dBpa's default value (visible in the advanced settings for the replay gain DSP) is a conservative -18.

But there is another issue, if using -18 on a particular file would calculate a replay gain number that would create clipping in the playback device, dBpa reduces the number to avoid possible clipping, making the playback softer than the target loudness.

"Replay Gain apply" in dBpa essentially applies the replay gain tags on the files to modify the gain (volume) of the stored audio file the same as a properly functioning playback device would, but in the file before playback. It is useful if you have a playback device that is incapable of applying the replay gain tags. Again, it doesn't change the dynamic range of the file, only the overall volume. People typically don't like to use this DSP on original CD rips because there is no way to determine if the file became corrupted after the DSP is applied, (The corruption determination can only be done for CDs in accuraterip when the "apply" DSP has not been used because the actual audio is modified by the "apply" DSP, the volume has been changed. So most people suggest if you are going to use this DSP, make a copy of the files and apply it to the copy, save the original as a backup. But in your case, this doesn't apply, you didn't rip CDs, so there is no accuraterip to compare it to (well, unless you burn a CD of the transferred recording, then rip the CD, creating an accuraterip entry).

And now, why I've spent so much time describing all this: It is quite possible that the loudest samples in your transfers of the phonograph records is in fact pops and clicks that your "mild" declicking did not remove. Therefore the dBpa replay gain calculation will fail as I described above, it won't calculate the correct number because the pops and clicks would be driven into clipping. Now, the clicks probably don't sound that loud, because they aren't anywhere long enough to be perceived as very loud. But they are messing up dBpa's algorithm.

So what to do? Apply a little dynamics processing to shave off the pops and clicks without audibly affecting the music. I don't know any way to do this directly in dBpa (in fact I once asked Spoon about this and got a negative reply). You didn't say what software you used to capture the transfers from vinyl. Perhaps that has a limiter plugin to apply. You should be easily able to see the difference between the clicks and the music peaks if your software has a graphic interface. If that doesn't work in your case, look for a free or inexpensive audio editing program, like Audacity. Use the limiter function to shave off the clicks. Then the dBpa replay gain DSP should work for you. Unfortunately, there is no way to do this automatically, you have to deal with each file individually.

I use software that includes a loudness processor that will automatically apply a pretty transparent limiter if needed to meet my loudness goal for every file I create when transferring vinyl, after declicking and any other restoration. But it is very expensive software, more than you want to spend for your use. (I use it for a lot of other purposes, such as cleaning up live recordings.) But the limiters in Audacity and similar editing programs are perfectly adequate. You should experiment on copies of the files you transferred from the vinyl.

[David W]

Re: Volume Normalization

schmidj,

Many thanks for the extended reply. I'll review it over the next day or so and be far better equipped to make some decisions on the best path forward.

David

[garym]

Re: Volume Normalization

Thanks @schmidj. Very useful explanation.

[McLion]

Re: Volume Normalization

Many thanks @schmidj! Great posting and explanation.

[larrydonline]

Re: Volume Normalization

The instructions still refer to "volume normalization", however I can't find any setting in options that mentions "volume normalization". Has "replay gain" replaced volume normalization but it's not noted in the intructions?

[garym]

Re: Volume Normalization

The instructions still refer to "volume normalization", however I can't find any setting in options that mentions "volume normalization". Has "replay gain" replaced volume normalization but it's not noted in the intructions?

No, there's still a "volume normalize" DSP. For example, open CD Ripper, go to the ADD DSP section (bottom center), then pretend to add a DSP to get the dropdown list, and you'll see Volume Normalize as a DSP.

[Mark S]

Re: Volume Normalization

Sorry to hijack this thread. The forum has been so helpful assisting me to understand Normalization vs Replay Gain. I have Normalized all my files and I have to say, the consistency is amazing (using the EBU R128 I believe). I have seen comment that the default is very low and I have found the same. What I can't work out is how to increase the volume leveel of the normalization. I can see the "Desired Volume" slider on the right of the Settings window. And a dB value below it. What is the link between the slider and the value that can be entered?

[garym]

Re: Volume Normalization

what did you normalize them at with EBU R128. LUFT of -18 is default (and the default used to be -21). If you want the files louder, this number should be smaller. So instead of -18, try -16 or -14, etc. I think -18 is trying to normalize to 89dB (and the older ReplayGain algorithm used 89dB I believe..

p.s. I've never found my files using RG (EBU R128) with -18 to be too low in volume. But I'm playing on systems that I can turn way, way louder than I could stand to listen to.....

[Mark S]

Re: Volume Normalization

Thank you so much. First time I have understood that I need to use negative values. I was trying to increase volume and thus used "+". Thanks again.

[garym]

Re: Volume Normalization

Thank you so much. First time I have understood that I need to use negative values. I was trying to increase volume and thus used "+". Thanks again.

wow. Cannot imagine what a POSITIVE value would do! Super low volume at best.