"Loudness War" Dynamic Range Compression & The DR Database - Observations
[h=Preface]%2[/h]
The "Loudness War" of popular music has been ongoing for more than 20 years. In recent years, an online database of dynamic range measurements of both digital and vinyl (ripped to digital) ratings for a large number of music titles (in excess of 50K database records) has been compiled by users. The DR Offline Meter and its analog plugin for foobar2000 can rapidly provide a measure of all music tracks and the aggregate album dynamic range (DR) ratings.
Any user can query this database for their favorite CDs, SACDs, and vinyl rips to disk to see the scored dynamic range of their particular music disc or media files, to see if their music files have been subjected to varying degrees Loudness War dynamic range compression techniques. In order for a user to to do a database query, they enter in an artist name and/or album into one or both of the supplied database search fields.
If the exact disc title/version title cannot be found, the user can then choose whether or not to analyze their own music media and then conveniently and quickly upload that information back to the DR Database, thus growing database utility for the entire user community.
Perhaps not realized by most users, all the DR database records are also available for download--albeit using screen-by-screen cut-and-paste of database records. It was the entirety of the ~38,000 DR Database that were downloaded in April 2013 and analyzed for observations here. It is this source of data which has been used to generate the plots shown in this wiki article.
[h=Initial Observations]%2[/h]
The following top-level observations were gleaned from initial inspection:
[h=Analyzing the Data]%2[/h]
The first area investigated was album vinyl vs. digital DR average values spread by year of release.
The blue bars coincide with the average of ripped vinyl albums, and the green bars indicate population means for digital media (CDs, SACDs, DVD-As, mp3s, etc.). The short bars are the standard deviation of each population. Note that even the CD format has at least 30 dB more signal-to-noise ratio (SNR) capability than does vinyl, but as can be seen here, no attempt to use this added capability by most record companies, with some notable exceptions from small recording firms such as Sheffield Labs, DMP, etc..
Next a plot was generated using only reported maximums for each album/CD/digital track, by year of release. The following plot also indicates that even the "uncompressed" tracks relative to vinyl are compressed or recorded in the far field relative to live performance dynamic levels:
[comment]Note: It was noted by one former recording engineer that about the time that CDs came out in late 1984, the large studios studios began to use near-field monitors for recording and mixing. These speakers have a sensitivity of around 86 dB or so, so therefore they can't reproduce the dynamic range of the soffit mounted monitors that they replaced. This seems to coincide with a general reduction of DR on all consumer formats as studio use of near-field monitors became prevalent.[/comment]
[h=More observations...]%2[/h]
From observation of inividual records, it was determined by inspection that digitial recordings ought to be recorded no higher than the -20 dB(FS) average level or below for virtually all music and speech genres in order to capture natural music peaks. However, most popular titles nowadays are being released at -5 to -7 dBFS average levels. It is also noted that recording industry standards do not limit or regulate average recording loudness levels as does the movie/video industry's DVD/BD formatted disks via THX and other video/film recording standards.
[h=Vinyl Vs. Digital]%2[/h]
The following plot was made of total DR Database ratings counts of both vinyl and digital formats. This plot shows bi-modal data characteristics characteristics in the digital data, implying that there are actually two types of digital disks in terms of dynamic range:
Another slice was performed on the data to find the differences in DR ratings between vinyl and digital copies of the same album/CD in those instances where both vinyl and digital DR ratings were found in the DR Database. All of these data values were then filtered for only those released in 1992 or later, which by inspection is about the starting year of serious Loudness War compression for many pop and rock artists. The resultant data showed that 86% of the time, the vinyl version had higher average and maximum DR values.
Since the standard deviation of the differences in DR ratings between vinyl and the best digital values for this example was fairly large, a plot of the DR rating differences vinyl-digital was made and found the following:
The bottom line on the above plot is: if you have a choice between vinyl and digital for a new release album, 86% of the time, the vinyl record will have a significantly higher DR rating. How much higher? By the plot, this translates into 2-3 dB extra dynamic range on average--which is actually a large difference if the music is compressed as on typical digital releases nowadays.
[EDIT 27 Sept. 2014] See http://forums.stevehoffman.tv/threa...ccurate-for-vinyl.330706/page-11#post-9481216 for an alternative explanation of this phenomenon.
[h=Two Types of CDs]%2[/h]
Another cut on the data, this time plotting the histograms for digital disks only, but split into two populations: one population below an album DR rating of 10, and the other population with DR ratings of 10 and higher (plot below). The idea is to see the relative count and value of these "two types of CDs". What was found was that for the data listed in the DR database, there are approximately equal numbers of "loud" and "quiet" CDs. Why did I pick a DR value of 10? Because the shape of the average curves for the two populations started to look like normal distributions instead of double hump/bi-modal curves.
The notion that "all CDs are loud" doesn't exactly hold water since there seems to be a bit more CDs/digital disks with average album DR ratings of 10 or higher as there are below 10. DR ratings of 10 correspond roughly to average-to-max dB ratios on each track of -13 dB. This means that the average dB level on each track has to be -13 dBFS or lower in order to reproduce at a DR rating of 10. If you are looking for an average DR rating of, say, 20, you would have to have an average dBFS level on your disc of -25 dBFS or less. That's a dramatic difference from the loudest CDs, which might have average dBFS values of -8 dB or even louder, such as -6 dB. (Remember that Red Book CDs can have average dBFS levels of -35 dBFS or lower and still not audibly affect low levels, or "quantization effects" at the low end of the scale. This means that the dynamic range of CDs is barely being used, even for discs with average DR ratings of 20 or higher.
It's been mentioned that certain types of music just don't have a great dynamic range. But note that even acoustic guitar DR ratings typically have very high DR ratings. In an example of a men's A Capella choir ("Cantus") with no apparent compression used during recording/mixing/mastering, maximum DR ratings of 17 are typical.
[h=Subjective Observations]%2[/h]
[h=The Darker Part of the Story...]%2[/h]
Why would these typically giant record companies consistently place better, more dynamic recordings on vinyl--which is more expensive to produce at high quality, but also having a significantly lower inherent performance of the format itself relative to digital? Each dB of extra dynamic range really adds to the sense of realism and involvement with the produced recording, especially at the compressed levels found in most popular music titles of the last 20+ years.
One answer is that it is a way of charging customers more for their music, but another more sinister aspect arises: the digitally recorded source material is first converted into an analog signal, then impressed onto the vinyl, then this copy is sold to the consumer, which typically does not then rip the vinyl analog music back to digital. Why? Consider an "anti-digital lie" memeplex and aura of vinyl: consumers might believe that their "analog recordings" (which is of course a misnomer) are superior. They also will be less inclined to reconvert their vinyl-based recordings back into digital files to share with their friends.
Since the record giants almost do not use dynamic range at all, it may be that vinyl is "good enough" to reproduce music using only a few dBFS of dynamic range, even though the signal-noise (SNR) dynamic range used is at least 45 dB less than Compact Disk (CD) "red book" format capabilities. Apparently, for many folks listening to these compressed audio genres, this is acceptable. However, this practice has resulted in an uproar by factions within the recording community and by consumers demanding better quality recordings--notably the "audiophile community".
There may be a technical reason for the downhill slide in dynamic range since digital recording and playback came into being. In the analog days there was a limit on how much compression and limiting could be used before audibly objectionable consequences would result. Too much compression and/or limiting would allow the bass notes to affect the high notes resulting in an effect known a "pumping" or "breathing". Then someone invented digital multi-band compressors. These units separate the audio signal into 3 or more frequency ranges so that those audibly objectionable results would not occur even if brick wall limiting was used. So the recording and mastering engineers began to produce recordings with limited dynamic range that would sound "better" on iPods and car stereos that are used in areas with more ambient noise than a quiet listening room.
[h=The Last Word...]%2[/h]
Now that we've discussed the DR ratings scale, data observations, correlations to our music preferences, and how Loudness War CDs occur, the next subject is at least as important: what the DR ratings don't tell us. I'll enumerate some obvious ones.
DR ratings don't tell us about:
This list is by no means exhaustive. Note that the DR database wasn't generated as a panacea for audiophile issues, but merely a reaction to the dynamic range compression increasingly used in pop/rock/mass market recordings.
[h=Preface]%2[/h]
The "Loudness War" of popular music has been ongoing for more than 20 years. In recent years, an online database of dynamic range measurements of both digital and vinyl (ripped to digital) ratings for a large number of music titles (in excess of 50K database records) has been compiled by users. The DR Offline Meter and its analog plugin for foobar2000 can rapidly provide a measure of all music tracks and the aggregate album dynamic range (DR) ratings.
Any user can query this database for their favorite CDs, SACDs, and vinyl rips to disk to see the scored dynamic range of their particular music disc or media files, to see if their music files have been subjected to varying degrees Loudness War dynamic range compression techniques. In order for a user to to do a database query, they enter in an artist name and/or album into one or both of the supplied database search fields.
If the exact disc title/version title cannot be found, the user can then choose whether or not to analyze their own music media and then conveniently and quickly upload that information back to the DR Database, thus growing database utility for the entire user community.
Perhaps not realized by most users, all the DR database records are also available for download--albeit using screen-by-screen cut-and-paste of database records. It was the entirety of the ~38,000 DR Database that were downloaded in April 2013 and analyzed for observations here. It is this source of data which has been used to generate the plots shown in this wiki article.
[h=Initial Observations]%2[/h]
The following top-level observations were gleaned from initial inspection:
- Early CDs produced in the 1980s through 1990 categorically have much more dynamic range than those produced/remastered since 1991, and particularly in the last ~15 years. It is noted that 1980s-vintage CDs perhaps contrary to common belief have the highest average levels of audio dynamic range of any period of time in music recordings.
- The phrase "remastered" was actually a another way of saying "dynamically compressed" for virtually all titles/discs with this phrase attached - relative to original titles available (typically in the 1980s era discs or even pre-CD time--in vinyl or analog tape format).
- Highly dynamic titles are recorded at what is called today "quiet" levels in order to provide the necessary headroom for dynamic transients at the top of the scale (loudness). These same discs sound dull if played back at levels typical for "loud" or dynamically compressed disks: i.e. these "loud discs" are literally louder when played back at preamplifier gains appropriate for quiet discs. However, when the preamp gain is increased to the levels appropriate for the quiet discs, highly dynamic recordings come alive in terms of subjective listening experience.
[h=Analyzing the Data]%2[/h]
The first area investigated was album vinyl vs. digital DR average values spread by year of release.
The blue bars coincide with the average of ripped vinyl albums, and the green bars indicate population means for digital media (CDs, SACDs, DVD-As, mp3s, etc.). The short bars are the standard deviation of each population. Note that even the CD format has at least 30 dB more signal-to-noise ratio (SNR) capability than does vinyl, but as can be seen here, no attempt to use this added capability by most record companies, with some notable exceptions from small recording firms such as Sheffield Labs, DMP, etc..
Next a plot was generated using only reported maximums for each album/CD/digital track, by year of release. The following plot also indicates that even the "uncompressed" tracks relative to vinyl are compressed or recorded in the far field relative to live performance dynamic levels:
[comment]Note: It was noted by one former recording engineer that about the time that CDs came out in late 1984, the large studios studios began to use near-field monitors for recording and mixing. These speakers have a sensitivity of around 86 dB or so, so therefore they can't reproduce the dynamic range of the soffit mounted monitors that they replaced. This seems to coincide with a general reduction of DR on all consumer formats as studio use of near-field monitors became prevalent.[/comment]
[h=More observations...]%2[/h]
From observation of inividual records, it was determined by inspection that digitial recordings ought to be recorded no higher than the -20 dB(FS) average level or below for virtually all music and speech genres in order to capture natural music peaks. However, most popular titles nowadays are being released at -5 to -7 dBFS average levels. It is also noted that recording industry standards do not limit or regulate average recording loudness levels as does the movie/video industry's DVD/BD formatted disks via THX and other video/film recording standards.
[h=Vinyl Vs. Digital]%2[/h]
The following plot was made of total DR Database ratings counts of both vinyl and digital formats. This plot shows bi-modal data characteristics characteristics in the digital data, implying that there are actually two types of digital disks in terms of dynamic range:
Another slice was performed on the data to find the differences in DR ratings between vinyl and digital copies of the same album/CD in those instances where both vinyl and digital DR ratings were found in the DR Database. All of these data values were then filtered for only those released in 1992 or later, which by inspection is about the starting year of serious Loudness War compression for many pop and rock artists. The resultant data showed that 86% of the time, the vinyl version had higher average and maximum DR values.
Since the standard deviation of the differences in DR ratings between vinyl and the best digital values for this example was fairly large, a plot of the DR rating differences vinyl-digital was made and found the following:
The bottom line on the above plot is: if you have a choice between vinyl and digital for a new release album, 86% of the time, the vinyl record will have a significantly higher DR rating. How much higher? By the plot, this translates into 2-3 dB extra dynamic range on average--which is actually a large difference if the music is compressed as on typical digital releases nowadays.
[EDIT 27 Sept. 2014] See http://forums.stevehoffman.tv/threa...ccurate-for-vinyl.330706/page-11#post-9481216 for an alternative explanation of this phenomenon.
[h=Two Types of CDs]%2[/h]
Another cut on the data, this time plotting the histograms for digital disks only, but split into two populations: one population below an album DR rating of 10, and the other population with DR ratings of 10 and higher (plot below). The idea is to see the relative count and value of these "two types of CDs". What was found was that for the data listed in the DR database, there are approximately equal numbers of "loud" and "quiet" CDs. Why did I pick a DR value of 10? Because the shape of the average curves for the two populations started to look like normal distributions instead of double hump/bi-modal curves.
The notion that "all CDs are loud" doesn't exactly hold water since there seems to be a bit more CDs/digital disks with average album DR ratings of 10 or higher as there are below 10. DR ratings of 10 correspond roughly to average-to-max dB ratios on each track of -13 dB. This means that the average dB level on each track has to be -13 dBFS or lower in order to reproduce at a DR rating of 10. If you are looking for an average DR rating of, say, 20, you would have to have an average dBFS level on your disc of -25 dBFS or less. That's a dramatic difference from the loudest CDs, which might have average dBFS values of -8 dB or even louder, such as -6 dB. (Remember that Red Book CDs can have average dBFS levels of -35 dBFS or lower and still not audibly affect low levels, or "quantization effects" at the low end of the scale. This means that the dynamic range of CDs is barely being used, even for discs with average DR ratings of 20 or higher.
It's been mentioned that certain types of music just don't have a great dynamic range. But note that even acoustic guitar DR ratings typically have very high DR ratings. In an example of a men's A Capella choir ("Cantus") with no apparent compression used during recording/mixing/mastering, maximum DR ratings of 17 are typical.
[h=Subjective Observations]%2[/h]
- Initial observations that there is a correlation between disks that I like to listen to and their track-level DR ratings have borne out, even more than I was expecting. I've found that certain disks were louder than others from the same artists, and I found that these are the disks that typically stay on the shelf. This is a significant finding--perhaps the most significant of all.
- The same expectations of DR on images and video also apply to audio. But while video and images usually enjoy at least 10 stops of dynamic range (i.e., -30 dBFS, or a DR rating of about 40) with 14 stops being the dynamic range of the human eye (i.e., -42 dBFS, or a DR rating of 51), audio tracks typically have maximum DR on the order of 7 "stops" (20 dBFS, or a DR Database reading of about 16-17). The latest loudness-war CDs can have an average of less than 2 "stops" of dynamic range (i.e., -6 dBFS or a DR Database rating of ~3).
- I've found that I cannot enjoyably listen to CDs with a DR rating of less than ~9 (-6.6 dBFS, or about 2.2 "stops") without being aware of the dynamic compression at times. Probably no music recordings should have a DR rating below 10 or 11 unless dynamic compression has been used on the tracks, and a average disc DR rating of 12 should be used as the clear dividing line between loud and non-compressed CDs.
- Another observation is that the loud CD maximum values (i.e., the max DR rating for the best track on the disk) shows a low and flat curve, indicating that all tracks on these loud discs are being uniformly squashed in their dynamic range, with few or no tracks above the disc average value, while the high DR population of discs shows a very healthy max DR curve, indicating that few if any tracks are being artificially reduced in dynamic range.
- Of the 786 instances of vinyl vs. digital recordings compared above, this author personally owns only 17 of those recordings, mostly in digital format. Of those 17 instances, more than 2/3s of the digital versions of each owned were at least as good or better as the vinyl version in terms of DR ratings.
[h=The Darker Part of the Story...]%2[/h]
Why would these typically giant record companies consistently place better, more dynamic recordings on vinyl--which is more expensive to produce at high quality, but also having a significantly lower inherent performance of the format itself relative to digital? Each dB of extra dynamic range really adds to the sense of realism and involvement with the produced recording, especially at the compressed levels found in most popular music titles of the last 20+ years.
One answer is that it is a way of charging customers more for their music, but another more sinister aspect arises: the digitally recorded source material is first converted into an analog signal, then impressed onto the vinyl, then this copy is sold to the consumer, which typically does not then rip the vinyl analog music back to digital. Why? Consider an "anti-digital lie" memeplex and aura of vinyl: consumers might believe that their "analog recordings" (which is of course a misnomer) are superior. They also will be less inclined to reconvert their vinyl-based recordings back into digital files to share with their friends.
Since the record giants almost do not use dynamic range at all, it may be that vinyl is "good enough" to reproduce music using only a few dBFS of dynamic range, even though the signal-noise (SNR) dynamic range used is at least 45 dB less than Compact Disk (CD) "red book" format capabilities. Apparently, for many folks listening to these compressed audio genres, this is acceptable. However, this practice has resulted in an uproar by factions within the recording community and by consumers demanding better quality recordings--notably the "audiophile community".
There may be a technical reason for the downhill slide in dynamic range since digital recording and playback came into being. In the analog days there was a limit on how much compression and limiting could be used before audibly objectionable consequences would result. Too much compression and/or limiting would allow the bass notes to affect the high notes resulting in an effect known a "pumping" or "breathing". Then someone invented digital multi-band compressors. These units separate the audio signal into 3 or more frequency ranges so that those audibly objectionable results would not occur even if brick wall limiting was used. So the recording and mastering engineers began to produce recordings with limited dynamic range that would sound "better" on iPods and car stereos that are used in areas with more ambient noise than a quiet listening room.
[h=The Last Word...]%2[/h]
Now that we've discussed the DR ratings scale, data observations, correlations to our music preferences, and how Loudness War CDs occur, the next subject is at least as important: what the DR ratings don't tell us. I'll enumerate some obvious ones.
DR ratings don't tell us about:
- Poor quality recording and editing equipment that introduces noise and degrades the sound quality, including high levels of analog tape hiss
- Poor quality transfer to the recorded medium, e.g., noise or wow/flutter of vinyl records, and off-center or off-balance disks/discs.
- Poorly EQed recordings during mixing or mastering
- Pops, ticks, warps, and other vinyl issues
- Uninteresting mixing or mastering of produced recordings
- Poor musicianship or composition
- Poor reproduction of either extreme low frequencies or extreme high frequencies, effects of monaural bass lines on vinyl recordings, effects of analog recorder limitations on capturing accurate percussion transients such as cymbals, triangles, small bells, glockenspiel, etc.
- Vinyl FR loss toward the inside tracks of the disk
- Differing sensitivity of some listeners to analog recording defects
- Constraints in listener tastes for certain music types, i.e., preferences for acoustic music or for amplified music, preference or dislikes for certain music genres, preference for a limited set of music of a certain historical time period (to the exclusion of other types and vintage music)
- Use of "enhancement" techniques such as aural exciters, over-processing, autotune, electronic synthesis of acoustic instruments, electronic percussion, electronic voices, synthesizer effects, etc.
This list is by no means exhaustive. Note that the DR database wasn't generated as a panacea for audiophile issues, but merely a reaction to the dynamic range compression increasingly used in pop/rock/mass market recordings.
very nice statistical work you did... again be thanked in the name of "dynamophile" music listeners ... 
why not have a look at my own modest database using the Ber-SD ClippingAnalyzer, which is a very fine, imho more useful tool for analysing clipping and dynamic content of audio files...
Clipping and Loudness Database
general information on fighting the loudness war in my own way:
the music disease
All the best!
Felix
why not have a look at my own modest database using the Ber-SD ClippingAnalyzer, which is a very fine, imho more useful tool for analysing clipping and dynamic content of audio files...
Clipping and Loudness Database
general information on fighting the loudness war in my own way:
the music disease
All the best!
Felix
Felix,
Thanks for your interest and comments. I'm also digesting your applications and links to plugins for DR analysis and processing.
The simple statistical analyses that I performed for this article were meant to be self-explanatory and easy to defend since I find the tendency for skeptics to hold pre-formed opinions to be fact very strongly in this area. The histogram is one example of simple statistical tool that I find that works well for this subject.
Regards,
Chris
Thanks for your interest and comments. I'm also digesting your applications and links to plugins for DR analysis and processing.
The simple statistical analyses that I performed for this article were meant to be self-explanatory and easy to defend since I find the tendency for skeptics to hold pre-formed opinions to be fact very strongly in this area. The histogram is one example of simple statistical tool that I find that works well for this subject.
Regards,
Chris
Hi Chris!
Yes, this is really very nice work, there are NOT too many people doing things like that for non-profit actually... if you need that one *.dll, PN me...
one can nicely see hao the discrepancy between CD / LP developed ... you could even make a scientific paper out of this (believe me I'm a natural scientist ;-)) hehe...
after all I came back to this topic after a (personally related) break for almost two years...
I just want to show you that appreciation for your effort... and so on... we will see, what comes next...
thanks a lot for your reply!
Felix
I updated the section on "Vinyl vs. Digital" to reflect the results of an interesting experiment by a noted audio mastering engineer in the U.K. on the phenomenon of higher DR values for vinyl record tracks:
"[EDIT 27 Sept. 2014] See http://forums.stevehoffman.tv/thread...1#post-9481216 for an alternative explanation of this phenomenon." based on the following video made by that same mastering engineer: Is the DR database really accurate for vinyl? | Steve Hoffman Music Forums
"[EDIT 27 Sept. 2014] See http://forums.stevehoffman.tv/thread...1#post-9481216 for an alternative explanation of this phenomenon." based on the following video made by that same mastering engineer: Is the DR database really accurate for vinyl? | Steve Hoffman Music Forums
;-) oh... thanks! Dennis and me were also quite busy... lol... wait I'll get the links quickly...
Lana Del Rey, soundcheck orig vs vinyl & declipped - YouTube
so nice of him making that video... and inventing the term "SnowWhite edition" hohoho ;-)
1) CDA 2) Vinyl 3) CDA declipped 4) SnowWhite edition .... enjoy !!
Lana Del Rey, soundcheck orig vs vinyl & declipped - YouTube
so nice of him making that video... and inventing the term "SnowWhite edition" hohoho ;-)
1) CDA 2) Vinyl 3) CDA declipped 4) SnowWhite edition .... enjoy !!
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.