Tools for Analyzing the Quality of Mastering

[Strangelove424]

We’ve read about the loudness war, we can hear it in our music, but I started wondering if there was an easy way to analyze the dynamic range compression in a mix (the loudness) by looking at measurements of the music itself. Usually this is done with a peak meter, but the music doesn’t always have to peak over 0db to be loud or have its dynamics squashed. The peak meter is just showing us the tip of the iceberg, and there’s a lot we don’t see with a peak meter alone.  I recently tried looking at the spectrogram in Foobar, and found this pretty interesting. A spectrogram analyzes the frequency distribution, and displays density of sound information at any given frequency with black. So white in the following images is quiet, black is loud. The vertical axis is the freq range, horizontal axis time.  What you are looking at below is a comparison of Bob Dylan’s The Times They Are A Changin’ on 1964 mono vinyl release* vs the modern CD Remaster for a Greatest Hits downmixed to mono. Both sources were ripped into digital file formats and compared in Foobar.   
 

(click to expand: http://cdn.head-fi.org/9/95/953a19b1_1964monovsCDremaster_441downconvert.jpeg)
 
The CD remaster doesn’t ever get so bad that it peaks over 0db, but the quietness of the vinyl release is demonstrable just by looking at the spectrogram. There is also less emphasis on the midrange on the CD, making it sound thinner. The CD gets loud and screechy at times, the vinyl in a comparison balanced and polite, with better dynamics when Dylan really does pick up energy. 
 
While the audio industry wants to sell us more bits and bytes, they seem to be avoiding the essential problem of quality mastering.  So while one struggles to hear any difference in quality between the $15 CD he bought or the $30 HD track he DLed, the quality in difference between mastering jobs can be clearly demonstrated and measured. And I also believe it's sorely obvious to our ears, unlike larger formats/file sizes. 
 
I created this thread because I was hoping to compare experiences, methods, or software tools people have run across which would allow music listeners to see with their own eyes the quality of a master, and the amount of dynamic range compression or loudness that went into the mix. Or conversely, how well balanced and dynamic it is. Perhaps by exchanging information with each other about different albums, we can find out what the best mastered version is, and use objective information to arrive at a conclusion of what’s the best release version is to purchase.  
 
* Edit 9/21/2104: MindsMirror clarified an important point about the use of spectrograms: higher sample rates will place more white area in the diagram due to the FFT size used to generate the graph, and does not have anything to do with the actual sound. "It is the reason why a perfect sine wave does not appear as a perfectly thin line, it is blurred, especially in the lower frequencies when it is displayed on a log scale like that." This is an important aspect of using spectrograms, so sample rates must be downconverted before comparing masters with this technique. Images have been updated with downconverted sample rates.  
 
___________________________________________________________________

 
LINKS 
Links provided throughout the discussion will be posted here as a centralized resource. 
 
 
Bob Katz, Loudness, War & Peace:   http://www.youtube.com/watch?v=u9Fb3rWNWDA
 
Thomas Lund on Loudness Wars:     www.youtube.com/watch?v=BhA7Vy3OPbc (If anyone one can find part2 let me know)
 
Bob Katz on the Loudness War I:      www.youtube.com/watch?v=XCd6MHlo_iA
Bob Katz on the Loudness War II:     www.youtube.com/watch?v=5FVF9vRLyVc
Bob Katz on the Loudness War III:   www.youtube.com/watch?v=tPwni7yvSQc
 
Friedemann Tischmeyer w/ collected statements:  www.youtube.com/watch?v=qUcgg2vMX_s
 
Downmixing to mono in Audacity: http://www.youtube.com/watch?v=ZJ-Nq-os2YI
 
Best free software for audio: http://www.audioholics.com/how-to-shop/best-free-audio-software
 
Audacity: http://www.audacity.sourceforge.net/
 
Dynamic Range Album List: http://dr.loudness-war.info/
 
Foobar Dynamic Range Plugin: http://dr.loudness-war.info/downloads/foo_dynamic_range_1.1.1.zip
 
___________________________________________________________________________
 
TOOLBELT
 
 
Peak Meter

There are rules to handling audio, and the most basic rule is that no signal should ever go over line level (0db). From Wikipedia: “Clipping is a form of waveform distortion that occurs when an amplifier is overdriven and attempts to deliver an output voltage or current beyond its maximum capability. Driving an amplifier into clipping may cause it to output power in excess of its published ratings.” It is the job of the amplifier to receive line level input and amplify that signal. When the line level signal is over 0db, it is impeding on the amp’s design limits for headroom.  If an album violates the 0db rules, it is in technical violation of the rules for audio distribution/broadcast, and can lead to premature equipment failure. A clip or two in an otherwise well-mastered album is not necessarily a sign of a “bad” master. But with excessive clipping, especially in combination with squashed dynamics, it is a major indication of a badly produced master. Peak meters are available for free in Foobar, but can also be found within the freeware program Audacity (for all OSes).  The Foobar peak meter looks like a typical analogue peak meter, and displays the current peak levels in the music, leaving a faint ghost line for a second or two. There is also a real time peak meter in Audacity, but by analyzing the waveform in Audacity (see next entry) we can see all the peaks in a song at once, marked either by a red line or specific clipping analysis. After importing a song into Audacity, goto View->ShowClipping. To perform an in-depth clipping analysis, goto Analyze->FindClipping. The lower the threshold samples, the longer the scan will take (it takes a fraction of a second to scan) but the more clipping it will catch. Lower start and stop threshold samples will catch the most incidents of clipping.   
 
Foobar: http://www.foobar2000.org/download
Audacity: http://www.audacity.sourceforge.net/

Analysis for clipping in Lady Gaga’s Poker Face:

Foobar real time peak meter (clipping both channels at time of capture)

 
 
Audacity “Show Clipping” Option checked

 

Audacity w/ “Show Clipping” checked and “FindClipping” scan with sample threshold of 1

 
 
 
 
Waveform Analyzer

A waveform, like a peak meter, is a measure of signal amplitude, but it changes over time and is designed to highlight dynamics in a signal’s amplitude/loudness. Large changes in amplitude indicate a dynamic track, while the flat-line plateau we see on many new records indicates a loud, compressed mix. When a mix is always loud, it is really always quiet. Perception in dynamics and energetic shifts in music require us to perceive a loudness shift, and an increase in the signal. In the words of Bob Katz, “things need to get quiet before they can get loud, you need a decrescendo after every crescendo”. But if the source is always loud, always hovering near it’s peak, and there are no “quiet” moments, our ears adapt with their own kind of sound normalization, they literally tighten up their anvils and stirrups and shut off to the constantly high amplitude in the signal. The nuances of the track become hidden, it becomes hard for the brain to pick out details, so what do we do? Turn up the volume! Then our ears adapt again. What do we do next? Turn up the volume again! The dog keeps chasing its own tail until we eventually need to sit in a quiet room with a Tylenol.   

Waveforms can be analyzed in any professional audio program, or the professional-grade freeware Audacity available here:
 
http://www.audacity.sourceforge.net/

This a sample waveform analysis of some rock'n'roll through the years... the notoriously loud and dynamically squashed Red Hot Chili Peppers’ Californication (CD version), and below it a dynamic version of Hotel California (on both vinyl and CD). 
 

 
 
 
Loudness/Dynamic Range Meter

These wonderful tools make it very easy to measure peak, loudness (the average level of the waveform), and dynamic range (min/max of the waveform) all in one meter. It is made by DR, which is also the website that hosts dynamic range list for a large assortment of popular albums (link below). There is both a Foobar plugin and stand-alone app available for download.

FoobarPlugin: http://dr.loudness-war.info/downloads/foo_dynamic_range_1.1.1.zip
 
Windows App: http://dr.loudness-war.info/downloads/DR-Meter-1_4a.zip
 
OSX App: http://dr.loudness-war.info/downloads/DR-Offline-Meter-for-MAC.zip
 
Album list: http://dr.loudness-war.info/
 
A huge thanks to member RRod, who recompiled a DR meter for Linux (successful with Ubuntu 14.04). Copied from post #59 of this thread:
 
"There's a github project for a Linux-native dynamic range meter:
https://github.com/adiblol/dr_meter
It is, however, horribly out of date (last update was 3 years ago).  I went ahead and did what I could in terms of updating the function definitions, and I got it to compile if I left in one deprecated function call.  Here are the files (they compile for me on Ubuntu 14.04 with the latest libav file installed):
https://drive.google.com/folderview?id=0BwmVtb5IwniEalgtWVhCNFFJTGs&usp=sharing" - RRod 
 
 
These are sample Dynamic Range readings from popular music, roughly a generation apart:

Awolnation’s Sail (2011)

 
Awolnation’s Burn It Down (2011)

 

Beastie Boys’ Intergalactic  (1998)

 

Beastie Boys’ Song For the Man (1998)

 
Michael Jackson's Smooth Criminal (1987 - this version released on CD early 90's)

 
 
Michael Jackson's Thriller (1982 - this version released on CD early 90's)

 
 
 
 
Spectral Analysis 
 
We often analyze FR graphs for headphones to death, but are willing to pick up and listen to any master we find. Often times the frequency colorations we obsessively avoid (such as V-curves) are far more emphasized in the track than even between two sonically opposed headphones. A spectrum analyzer will only tell us what frequency we hear at any given sample, and makes it almost impossible to compare two masters. A spectrogram on the other hand, makes that possible by graphically depicting the frequency range over time. 
 
Although I started this thread discussing the spectrogram, this form of analysis has its downsides. First off, it’s hard to read a spectrogram and apply that in context to an actual song. At best, we can compare two master to confirm a difference, or specific trait (like a heavy V-curve) in one master. Otherwise, on its own, there’s not much we can gain. Mastering is a creative process, you cannot look at the frequency balance alone and make a decision about whether one master sounds a particular way, or better than the other. Use discretion when analyzing the frequency spectrum of a track. A headphone or an amp can be neutral, but a song can never be (nor should). When it comes to frequency analysis, always compare things in relation to one another, i.e. the same song between two master and realize that your ears are the ultimate judge, the spectrogram can only hint at differences your ears should already be hearing. A free to use spectrogram is included in both Foobar and Audacity. 
 
This video is a helpful guide to exploring Audacity's Spectrogram: 
https://www.youtube.com/watch?v=7WYw3qoTdU4 
 
Options for setting Audacity's spectrogram:
http://manual.audacityteam.org/o/man/spectrograms_preferences.html
 
 
In the following sample is a comparison between Bob Dylan's The Times Are A Changin' in vinyl and CD Greatest Hits remaster. If you were to take these songs into Audacity, they would both show relatively dynamic waveforms, yet the CD Greatest Hits version is hard for me to listen to. It's not a loudness issue though. Once I looked at it in the spectrogram, I realized it was due to the tone. Keeping in mind black=harmonic activty and white=lack of activity, the CD shows more harmonics in the lows and highs compared to vinyl, leaving a relatively sucked out midrange.
 
This is a common technique used (often but not always in combo with dynamic range compression) to make songs "stick out more" on the iTunes store or on radio, thus leading to more sales. The songs that sound the most "bassy and detailed" will most likely sell the most. Producers know this and push engineers to make the sound more "engaging" to digital store window shoppers. However, after long listening sessions this V-shaped emphasis can become fatiguing, or in the case of ancient recordings of a nasally singer such as Dylan, just torture from the start (with all respect to Dylan, the master poet that he was)...
 

 
 
 
Diffmaker/Null Testing
 
Note: I was in the process of creating audio samples of null/difference tracks for a number of different masters of the same song, but am having issues due to timing differences. This method may not be a reliable way to analyze differences in masters. If I can find a way around these timing differences, and this becomes fruitful, I will update the post. In the meantime, I thought it would be a good idea to leave a link to Diffmaker and the Audacity tutorial here since null testing is extremely powerful, and at the very least, it can be used to pinpoint the exact differences (or lack thereof) between a hi-res/lossless file and a down-converted copy.   
 
Diffmaker is a free program that allows you to perform null comparisons quickly and with ease. Diffmaker can match the waveforms of two files, match the levels, and then output the differences between the two tracks (in addition to creating ABX tests with those two tracks). Diffmaker can be used to objectively and incisively compare all sorts of changes, including physical components in a signal chain (such interconnects and wire), DSP, and converted formats. It is available here: www.libinst.com/Audio%20DiffMaker.htm Use with caution, may destroy prejudices.  
 
Screenshot of the easy-to-use Diffmaker UI:

 
 
This tutorial (http://sdk.bongiovidps.com/2013/09/26/audio-null-test/) demonstrates the process for null testing tracks in Audacity (remember to match levels when using this technique. Changes in gain between tracks will not be adjusted for as they are in Diffmaker, but this can be helpful in case you are trying to find gain differences). Thanks to Roly1650 for the link.
 
 
In Conclusion

I started this thread after having a few drinks and playing with the spectrogram in Foobar while listening to different masters. What started out as a random impulse has evolved into a consolidated list and a multi-pronged approach to analyzing a master without any cost to the listener. You just need a computer. The effort could possibly save us all a lot of time and money in the long run, especially if we share results. Save your fellow Head-Fiers a few bucks, and let them know what masters to stay away from! I intend to do that continually, and will use this thread to highlight both the dissapointing masters I run across as well as the rewarding ones that I think are worth listening to. Well mastered music should be respected, applauded, and elevated above the loud fray with our own adoration. That said, I don't see the trend for the loudness war getting any better, despite the software advancements in loudness normilzation. Hopefully threads like this, in due time, will center less around pointing out badly engineered music, and more about honoring well engineered music.  

Happy listening!

 

[bigshot]

Mastering and mixing are creative processes that require compromise and decision making on the part of the engineer. I think the only thing you can measure is whether two masterings are the same, not necessarily whether one is better than another.

 

[Strangelove424]

I don't think the graphic analysis can give us a thumbs up or thumbs down all on its own, but perhaps mixed with subjective impressions it can help us narrow down certain traits in a master we may or may not find appealing. You do raise a good point about differences, and I imagine the only way to do this kind of analysis would be to compare two masters next to each other. At the very least, I think it could help to identity when a new HD or Vinyl release some people expect to be mastered better is the exact same master as the CD.  

 

[bigshot]

I'm finding a LOT of HD audio that is identical to the CD. SACDs too.

 

[Strangelove424]

Me too. If the HD release is the same as the CD, we should know what we are paying for. Especially if the CD has loudness issues, and the HD release has the same exact issues.

 

[MindsMirror]

In the spectrograms you have shown, the magnitude of the frequency is shown in grey scale, with white being loudest and black being silence. The reason the vinyl spectrogram may appear more white is because the sample rate is higher, or the FFT size is smaller, resulting in lower frequency resolution. It has nothing to do with the loudness or the dynamic compression of those samples of audio.

 

[Strangelove424]

Back to this again, are we? Here's lossless vs. 128kbps, same track, they are nearly identical. Mastering makes a far greater impact than resolution. 
 

 
Edit:
 
I got an even tougher one... lossless vs. 320... totally identical. I reiterate, mastering trumps resolution or rate. 
 

 

[Strangelove424]

  In the spectrograms you have shown, the magnitude of the frequency is shown in grey scale, with white being loudest and black being silence. The reason the vinyl spectrogram may appear more white is because the sample rate is higher, or the FFT size is smaller, resulting in lower frequency resolution. It has nothing to do with the loudness or the dynamic compression of those samples of audio.

Ok, I looked at your post closer and realized you were talking about sample rate specifically, not bits. Sorry about that, and sorry for dismissing your comments so quickly. I re-converted the Mono Vinyl to 44.1, just like the Greatest Hits and this is the outcome:
 

 
You are correct that a high sample rate introduces more white to the spectrogram (but not more silence to the music), and this raises a very important point that sample rates need to be brought to the lowest common denominator before comparing masters using this method. Once this is done, I still think the quietness of the Mono Vinyl is obvious in the spectrogram, but I apologize for my dismissive comments.  

 

[MindsMirror]

  Ok, I looked at your post closer and realized you were talking about sample rate specifically, not bits. Sorry about that, and sorry for dismissing your comments so quickly. I re-converted the Mono Vinyl to 44.1, just like the Greatest Hits and this is the outcome:
 

 
You are correct that a high sample rate introduces more white to the spectrogram (but not more silence to the music), and this raises a very important point that sample rates need to be brought to the lowest common denominator before comparing masters using this method. Once this is done, I still think the quietness of the Mono Vinyl is obvious in the spectrogram, but I apologize for my dismissive comments.  

Yes, to clarify, frequency resolution is a result of the FFT used to generate the spectrogram, not anything to do with the audio. It is the reason why a perfect sine wave does not appear as a perfectly thin line, it is blurred, especially in the lower frequencies when it is displayed on a log scale like that.
 
I'm curious, what part of the spectrogram makes the vinyl seem obviously quieter? The only differences I see is the CD is a slightly different time interval, the vinyl has some more noise at the lowest frequencies, and the CD appears to be cut off at the top.
 
To compare the loudness using the two graphs, you have to look at the whiteness of the lines, which looks pretty much identical to me. That makes sense, since I'm guessing you got those spectrograms from the foobar visualisation, with replaygain enabled.
 
The spectrogram doesn't really give any obvious indication of dynamic compression due to loud mastering. Your ears are far better than your eyes when it comes to analyzing the frequency domain. I think the best way to see dynamic range compression is by looking at the waveform.
 
Lossless and lossy spectrograms do have an easily noticeable difference. The 128kbps mp3 has a frequency cutoff above ~16kHz, which you can see more clearly on a spectrogram if you switch to a linear scale. You may also be able to easily see the results of mp3s limited time resolution in a spectrogram with a small enough time window.

 

[BlindInOneEar]

The spectrogram doesn't really give any obvious indication of dynamic compression due to loud mastering. Your ears are far better than your eyes when it comes to analyzing the frequency domain. I think the best way to see dynamic range compression is by looking at the waveform.
 

I'm not really sure I follow what you are saying here.
 
Have you ever fiddled with Audacity?  Simply playing back a file in it gives you a pretty good visual indication of the amount of compression applied.  Further, using "Analyze" > "Plot Spectrum..." will tell you at a glance just how much "Hi-Rez" you actually get in those expensive, and gigantic, Hi-Rez files people offer to sell.  

 

[BlindInOneEar]

Originally Posted by Strangelove424 /img/forum/go_quote.gif
 
Mastering makes a far greater impact than resolution. 
 

This.
 
However, I wonder how much of the "Loudness Wars" is being driven by the market move away from big stereos in quiet rooms to ear buds crammed into ears and played back on buses or while wandering city streets or other areas with high ambient noise levels.  As much as we audiophiles enjoy dynamics, I suspect a lot of the low level dynamics that we miss would simply be inaudible to a modern urbanite enjoying music on his iPod.  

 

[MindsMirror]

  I'm not really sure I follow what you are saying here.
 
Have you ever fiddled with Audacity?  Simply playing back a file in it gives you a pretty good visual indication of the amount of compression applied.  Further, using "Analyze" > "Plot Spectrum..." will tell you at a glance just how much "Hi-Rez" you actually get in those expensive, and gigantic, Hi-Rez files people offer to sell.  

Dynamic range compression is trying to increase the loudness while keeping the signal level at or below the clipping level. In the waveform you can directly see this level. Look at the waveform for any loud CD, the worse they get, the more they just look like rectangles. You can almost directly see the dynamic range compression. Then look at the frequency spectrum, the clipping limit does not translate to any visible limit in the frequency domain, so there is not a good visual indication of dynamic range compression in a spectrum analyzer.

 

[BlindInOneEar]

I think we might be saying the same thing in two different ways.  Not the first time that has happened on the internet! 

 

 

[Strangelove424]

  Yes, to clarify, frequency resolution is a result of the FFT used to generate the spectrogram, not anything to do with the audio. It is the reason why a perfect sine wave does not appear as a perfectly thin line, it is blurred, especially in the lower frequencies when it is displayed on a log scale like that.

I will update the original post tomorrow morning to reflect more accurate visuals (a bit lazy to open Photoshop now), and information for spectrogram generation, and thank you for the clarification.
 
  

  I'm curious, what part of the spectrogram makes the vinyl seem obviously quieter? The only differences I see is the CD is a slightly different time interval, the vinyl has some more noise at the lowest frequencies, and the CD appears to be cut off at the top.

In post #9, with Vinyl on the left, I see less emphasis on bass and highs, more mid range fullness, and this seems evident while listening. The vinyl has more balance. The CD remaster, on the other hand has a lighter colored bass area, lighter colored highs, and deflated mid-frequency compared to the vinyl. It's a typical u-shaped master (that also happens to have louder signal levels) that seems to be produced for the low-fi market, where these frequencies are boosted to pop out and make sales. Again, the point of the spectrogram was to analyze something other than peak level, because peaks or waveforms really say nothing about frequency distribution or density at certain bands. Bad mastering I don't think just means loud mastering, there are other things that can make the sound thin and hollow as well.
 

  The spectrogram doesn't really give any obvious indication of dynamic compression due to loud mastering. Your ears are far better than your eyes when it comes to analyzing the frequency domain. I think the best way to see dynamic range compression is by looking at the waveform.
 
Lossless and lossy spectrograms do have an easily noticeable difference. The 128kbps mp3 has a frequency cutoff above ~16kHz, which you can see more clearly on a spectrogram if you switch to a linear scale. You may also be able to easily see the results of mp3s limited time resolution in a spectrogram with a small enough time window.

 
Regarding lossy vs. losless spectrograms, the 128kbps graph did have a cutoff, but otherwise I think it was vastly the same, And as I illustrated, 320kbps became indistinguishable (mirroring data that shows 256kbps is a safe cutoff range for audibility). Even streaming is @320kbps nowadays, I think we can put this bit depth thing aside as a reason a master doesn't sound good to someone.    
 
Spectrogram analysis to me have the benefit of indicating frequnecy balance. A spectrum analyzer doesn't track frequency change over time, so it makes it hard to compare data. Waveform analysis, although the goto tool for analyzing levels, doesn't give out spectrum information. That being said, the purpose of this thread was to amass a number of tools to help understand the quality of a master in a number of different ways. If you have a suggestion for a particular form of analysis, please suggest one, I am all ears. I have a couple of audio programs here like Soundbooth and Apple ST Pro, and am willing to download and learn open source software available on the internet if there are more things available out there to help this process along. Original post will be updated with any new software links, tutorials, etc
 

  This.
 
However, I wonder how much of the "Loudness Wars" is being driven by the market move away from big stereos in quiet rooms to ear buds crammed into ears and played back on buses or while wandering city streets or other areas with high ambient noise levels.  As much as we audiophiles enjoy dynamics, I suspect a lot of the low level dynamics that we miss would simply be inaudible to a modern urbanite enjoying music on his iPod.  

 
Yes, the songs are designed to pop out on low-fi equipment, to scream "buy me" during the 30 second sample someone hears it on their ear buds. But music should not be created for the lowest common denominator of enjoyment, To be pimped out in a musical 99 cent store.

 

[bigshot]

How do you know the vinyl isn't an upside down U instead of the CD being an upside one? The only way to know is to compare it to something calibrated to be flat, like for instance a classical recording. And the only way to do that would be by ear.