The Missing Octave(s) - Audacity Remastering to Restore Tracks

[Chris A]

I've rediscovered some facts that I learned early on when starting off learning to unmaster (Jan-Feb 2005).

 

When unmastering tracks, the SPL at your listening position needs to be as close to concert volume (or as loud as you will ever find yourself playing the tracks) in order to avoid defects in the unmastering process.  Notable unmastering defects in my older unmastered tracks include:

 

a) too much mid-bass in the 80-210 Hz band, resulting in a honky, midrangey, or too-full overall sound

 

b ) too much low bass below 50 Hz, mostly due to kick drum beats resulting in true deep bass transients that beat around the room's boundaries (walls), even if not apparent at the listen position while unmastering

 

c) too little 5-10 kHz energy, resulting in dull sound when turned up higher

 

etc.

 

I'm currently going through a number of my earliest unmastered tracks from early 2015 and re-doing them at a higher overall SPL (e.g., 90-95 dBC at the my listening position for rock-type music).  I've found a some things that I believe that stumps most beginning novices, along with some supporting observations, which I will describe in successive posts here.

 

I first start the un-EQing process by "seeing" the mastering EQ patterns used on the original tracks.  In the past, my approach was to successively apply mostly negative EQ haphazardly to the most visibly largest humps and peaks in the "plot spectrum" (cumulative EQ across several minutes of the track) until I got a mostly straight line decreasing EQ curve from 40 Hz--12 kHz.  This resulted in very complex and fragmented unmastering curves, resulting in fragmented unmastering track "plot spectrum" curves.

 

Now I find myself looking at the starting track plot spectrum view, and trying to visualize how the mastering engineer arranged the 1/3 octave sliders on his/her two track mastering board.  Most of the time, what appears to be very complex EQ is actually quite simple to undo.  An example...

 

Here is the "plot spectrum" view from a typical Rock music track.  Notice the hump from 400 Hz to 5 kHz:

 

 

In the past, I'd start by putting in a negative EQ dip anchored at 400-5000 Hz, with a trough centered at 1000 Hz.  I'd then proceed to add additional negative EQ dips in the correction curve for each remaining "hump" in the resulting spectrum, going back and forth, trying and undoing until I got an approximately straight-decreasing slope resulting curve. 

 

Nowadays I don't do that.  Here's what I do: I look carefully and simply add two points in my anchored endpoint starting EQ correction curve--one at 400 Hz and another at 1 kHz, pulling the EQ curve down from the second point:

 

 

...and the resulting "Plot Spectrum"...

 

 

This saves time and gets me much closer to the final EQ curve than the prior way of adding negative EQ dips in the curve.  Now all that is needed is to add a point at 5 kHz and add slight negative dip at 20 kHz to tilt the curve downward.  While this looks equivalent to the prior method, it turns out that the second approach converges more rapidly, requiring far fewer corrections to achieve a flat, decreasing slope "Plot Spectrum" curve and more closely approximating what the mastering engineer actually did when applying his/her original overall EQ to the mixdown two-track.

 

Chris

Edited July 10, 2016 by Chris A

[Chris A]

Just in case anyone cares, here is the final EQ curve that I used on this track...Take the Highway (LIve) from the album Where We All Belong (Remastered) by the Marshall Tucker Band (original 1974):

 

 

and the final "Plot Spectrum" view:

 

 

Additionally, here are the beginning and ending "Spectrogram log(f)" views showing the differences in SPL vs. frequency vs. time.  The difference in listening pleasure is like the difference between the living and the dead:

 

Beginning spectrogram:

 

 

and final (unmastered) spectrogram:

 

 

Chris

[Chris A]

Continuing with "lessons learned...yet again", Dr. Floyd Toole, formerly VP of Acoustical Engineering for Harman International, has said two things about loudspeakers and sound reproduction that significantly affect our unmastering efforts:

 

1) When talking about loudspeaker frequency response, he said...

 

"Other manufacturers quoting +/-3 dB of loudspeaker accuracy isn't good enough. +/-2 dB is much better, but at +/-1 dB, you're starting to get my attention."

 

Toole typically spends a lot of effort emphasizing flat FR, but in my view, this is an attribute that is easy to dial-in during installation.  Toole himself admits this. Independently of Toole, I've found that his comment is correct; we need +/-1 dB flatness between 400 Hz and 7-10 kHz at our listening position in order to achieve true hi-fi performance.  Additionally we need at least +/-3 dB of sound reproduction accuracy below 400 Hz...subject to the limitations of small room acoustics. This is actually fairly easy to achieve using digital crossovers and room correction software.

 

The comment that he made, though, still has enormous implications to our unmastering efforts, regardless of marketing of Harman products. Typical mastering practices significantly degrade the "hi-fi" in our hi-fi setups, but is mostly invisible to most "audiophiles" that aren't paying much attention to their source music. I've found that you cannot really have hi-fi without unmastering your recordings. Not even close...because almost all commercially bought recordings have been affected by mastering equalization significantly greater than +/-1 dB above 400 Hz, and +/-3 dB below that frequency. 

 

Toole's comment also tells us how much effort that should be put into correcting our music tracks (i.e., "a lot"), because the more that our music tracks conform to Toole's criterion (+/- 1 dB above 400 Hz, +/- 3 dB below), the higher the perceived hi-fi performance. 

 

2) Toole also said:

 

"Because of Fletcher-Munson effects of human hearing at low frequencies, very small changes in reproduced SPL at these low frequencies has a much greater effect on the perceived loudness at these frequencies." 

 

If you consider the common mastering practice of REALLY rolling off LF energy on most recordings in order to accommodate sound reproduction equipment that can't handle these frequencies without causing buzzing and poor overall performance, it becomes apparent how much these industry-standard mastering practices significantly alter recorded music fidelity..

 

This is one area where we need a fair amount of dialogue to understand how to unmaster current music track bass roll-off curves, and how to assure that the bass restoration is done to sufficient accuracy to achieve hi-fi performance.  More on this subject to come, but note that the unmastering to achieve realistic bass reproduction is actually quite sensitive to doing it well.

[WMcD]

Chris, I love reading about your work.

Excuse me if I missed it, but. Have you run across any recordings which are proper and not needing your work?

WMcD

[Chris A]

https://community.klipsch.com/index.php?/topic/155096-the-missing-octaves-audacity-remastering-to-restore-tracks/?p=1832549

 

Note that most of the discs in the list linked just above really do have issues, just not nearly as severe as all other discs on average. 

 

I see deviations from flat EQ (put there by mastering engineers) on the order of 10s of dB, not 1-3 dB variations that you'd expect for "enhanced" hi-fi playback.  It's like using a sledge hammer to make all consumers' playback systems conform to a single "one size fits all" (except every mastering engineer is using his/her own custom EQs that are different from all other mastering engineers, and a different set of EQ filters for each track...figure that one out--sometimes more than one set of EQs on each channel...

 

!

 

Chris

Edited July 10, 2016 by Chris A

[Chris A]

I'll add two more "good" discs requiring no changes that I've identified recently:

 

Redbird - Redbird

The Leinsdorf Sessions Vol. 1, Erick Leinsdorf and the LA Philharmonic (Sheffield Labs)

 

Both are on CD format.  I recommend both of these very highly.

 

Chris

[Chris A]

More "lessons learned-yet again"...

 

I've learned  that having a system with flat frequency response (±1 dB from 400 Hz-10 kHz, ±3 dB below that point except at major room mode frequencies) is required in order to demaster tracks and to hear the results plainly if you intend to use your demastered music on anything other than the system that you're unmastering them on.  An example of the measured frequency response in my room using REW and a calibration microphone.  Note the left-hand axis scale (usually plotted at ±5 dB or even ±10 dB per division--here plotted at ±2 dB/division):

 

 

EDIT (23 Aug. 2019):  The concept of "continuous improvement is applicable to this effort as it is to almost any repeating effort.  For the above loudspeaker, in the same room, the following update in SPL and phase response is noted to have basically annulled many of the posted demastering curves found in this thread:

 

 

Along with flat frequency response, I find that corner bass traps are required to control room modes in the 90-210 Hz frequency band that a produce a boomy sound. An example of measured "RT60" reverberation times in my room using REW and a calibration microphone:

 

 

If your system doesn't have flat response, when you listen to demastered tracks, you'll likely hear a more muffled, bass-heavy presentation, especially if you've added any "house curve" to your setup (decreased highs and increased bass). The solution is to undo the room curve that you might have used with your setup in order to hear balanced demastered tracks.

 

[The alternative is that everyone must use a standard "house curve" to listen to tracks which have been specifically demastered using whatever "standard" house curve that everyone can agree to. I find, however, that everyone understands "flat response", and that's what most Jubilee owners use from Roy's settings. Even small tilting from flat response from 20 Hz to 20 kHz is extremely audible--as little as 1 dB net tilt.]

 

When comparing demastered tracks to the originals, it's not difficult to hear the thin, table-radio-type tinny sound of the original tracks to the full, rich and balanced sound with deep bass and natural highs (...as in...not "ear splitting" and strident).  One of the most distinguishing features of demastered tracks is that, when played back at lower loudness levels, they sound very smooth and balanced and all instruments/voices are much easier to hear.  In fact when listening to some tracks that had been poorly mastered with boosted highs and attenuated lows, after unmastering you can hear parts that you might not have been aware of before especially for music tracks that you've heard over the years or perhaps decades. 

 

When played back at higher levels, the demastered tracks gradually sound better and better, until you're aware that the SPL is pretty much at concert volume, and perhaps even a little higher than "concert volume in the audience", rather "concert volume on stage"...for acoustic recordings, not amplified instruments. 

 

For albums that proclaim something like:

 

"...this music must be played back loudly..."

 

or words to that effect, I find that this is actually code for:

 

"We unbalanced the equalization of these music tracks so that they can only be listened to at one loudness level, on one set of studio monitors that we used in our studio (with likely poor performance), and that level will likely hurt at high frequencies when you listen to them".

 

The solution to resurrecting these tracks is to simply demaster them.  You'll thank yourself each time you play them again. 

 

Chris

[Chris A]

The bottom line is that your setup needs to be calibrated to flat response if you're unmastering music.  Knowing the effort put into creating really good unmastered tracks, I want my finished tracks to be able to be used on any setup in the future, not just just the setup that I'm using presently to unmaster.  I think a lot of people don't take the time to do this part well. 

 

It doesn't cost much at all except a little effort and EQ/room correction capabilities, which you get in the package at no extra cost if you use active digital crossovers, and is even easier if you also use an AVP or some device with room correction firmware that can compensate for channel-to-channel time delays.

 

Chris

Edited July 13, 2016 by Chris A

[Don Richard]

So, recapping the crossovers and using a tube amp isn't enough, LOL.

[Chris A]

More lessons learned:

 

During the entire time that I've learned to unmaster albums (~3-4Khours to date), I've noticed that the most important skills that lead to successful outcomes are:

 

1) Looking at the "plot spectrum" and "spectrogram log(f)" views extremely carefully to see not only the initial state of each track's EQ and noise, but also being able to look again, carefully, after a correction is made to see if smaller updates to the EQ unmastering curve(s) are needed.

 

2) Listening carefully to tell me when I'm converging on good result, and when things aren't converging well.  I've found that simply taking a break and coming back later will make all the difference in the world in seeing the underlying issue that's preventing the track from sounding natural.  I found this aspect of unmastering to be key to the "quality control" of the unmastered tracks.  I find that I have to wait a day and listen again to make sure that I wasn't drifting from a good result.

 

3) Patience in using the work already done on an EQ curve or saved in a track to continue with corrections, instead of dumping the track/EQ curve and starting over.  Once you get a track back to nearly a 1/f "plot spectrum" curve, I've found that makes future fine tuning much easier and much more sensitive to corrections.  Once you start getting close to a 1/f curve, very small corrections have proportionately large effects on the overall sound--sometimes as little as 1 dB at some frequencies. 

 

The only times that I dump tracks that I work on is when the initial trial with an EQ curve winds up sound significantly worse than the original track.  Then I step back and look more carefully at the overall picture and what my ears tell me is occurring.

 

Sometimes when I go back to the beginning, to redo tracks when I was just starting out, I'll go back to the original CD tracks and begin again, but this is only when I find that my beginning efforts led me far away from what the tracks sounded like before the mastering engineer(s) got a hold on them.

 

Sometimes, the answer is to come back to the track(s) at a later date, after you've forgotten working on them, then proceed using the corrections that you've already applied to the tracks.  I've found that very small changes in unmastered tracks that didn't quite "click" are the magic that I was looking for.  To start over again would be to start that entire process over again.

 

3) There are certain times when the 1/f "purple mountain" curve (as my wife calls it) isn't really a straight line.  Those instances include the following:

 

a) solo piano

b ) solo string instrument (violin, viola, cello, double bass, acoustic guitar)

c) string orchestra, including the standard orchestral wind instruments such as flute, clarinet, oboe, bassoon, trumpet, french horn, trombone, tenor or alto horn, and tuba

d) recordings of solo voice and one accompanying acoustic instrument, such as acoustic guitar or piano

 

In these cases, the "plot spectrum" view looks like an elliptical shape, such as the following "plot spectrum" curve:

 

 

The above curve is for solo violin and piano.  In this case, it's Nadja Salerno-Sonnenberg playing violin on the disc It Ain't Necessarily So, playing Fritz Kreisler's Liebesleid.  It took me two or three tries at unmastering the same tracks before my unmastering skills were fully up to the task, and I was successful at producing extremely natural sounding unmastered tracks for this album.  It was, clearly, one of the most difficult albums that I've unmastered to date...taking me several start/stop trials and re-listening the day after.  I'm very happy with these tracks now.

 

 

Chris

Edited July 19, 2016 by Chris A

[Chris A]

Every once in a while, I run across an album of some note--so called audiophile-quality CDs.  While some of these CDs are notable for their excellent recording quality, requiring little or no mastering, others of this group of CDs have been mastered with significant limiting (i.e., "clipping") and creative equalization applied. 

 

One of those notable "well produced" (versus "well recorded") audiophile favorite CDs is Flight of the Cosmic Hippo (1991) by Bela Fleck and the Flecktones:

 

 

This CD suffers from both clipping and creative applied mastering EQ to "enhance" its audiophile splashiness and to cut off the very lowest frequencies of the bass guitar (Victor Wooten) and synthesized kick drum (Roy "Future Man" Wooten) below 50 Hz.   The following EQ curves are provided for each track, along with a combined XML export file of the unmastering curves for your convenience.  Make sure to run "clip fix" before each EQ curve is applied, and normalize the tracks after each curve is applied to re-level each output track back to -0.3 dB peak.

 

Here are the unmastering EQ curves for each track of this CD:  Flight of the Cosmic Hippo album EQ curves.XML

 

These corrections will not sound very good if you pre-apply a "house curve" to attenuate highs and accentuate lows below 100 Hz. 

[Chris A]

Note that the above EQ curves can be adjusted for taste using a simple curve to increase the relative level of highs or lows, like the one posted earlier in this thread:

 

 

This is the type of salt-and-pepper adjustment that I make to tracks the day after doing the big corrections to undo the typically heavy-handed mastering EQ curves that are applied to tracks.  In the case of Flight of the Cosmic Hippo, the above EQ curves aren't really that dramatic, but serve to show that even "audiophile quality" CDs are affected by those forces that seek to "make it sound better" at the distinct expense of lowering fidelity.

 

Today, I'm adjusting the unmastered tracks formed by the above EQ unmastering curves.  In general, having a fresh listen to all unmastered tracks usually results in very small adjustments to the tracks in order to bring out their musical character--but without significantly altering the overall balance of the track.  Very small changes to the unmastered tracks produce disproportionately greater effects, because the unmastered tracks have been returned to something much closer to the natural recording first picked up by the microphones in the recording space (before the recording is altered). 

 

Chris

[mustang guy]

I like the term salt-and-pepper adjustment. In a way, on a stereo the bass and treble are the salt and pepper just like above...

[Chris A]

-

 

[Chris A]

 

Here's an example of a CD produced in 2002 with extremely repairable mastering artifacts...namely clipping and creative EQ applied, including bass roll-off below ~100 Hz.  The results were so spectacular that I felt this one deserved a detailed posting of the EQ curves.  Note the following process using Audacity to unmaster:

 

1) run "Clip Fix..." at the 95% level, then run "Normalize..." at -0.3 dB to first remove the clipping used on almost all the tracks

 

2) run the respective EQ curve found the following XML import file: Molly Johnson - Another Day EQ curves.XML

 

3) run "Normalize..." once more to re-normalize the track, then save using export (ctrl-shift-E) back to disc.

 

I think that you'll find this CD to have superior sonic qualities.  The following screen shots will give you an idea of the unmastering EQ curves used for this album.  The "DR" dynamic range rating for this disc improved almost 3 points -- from 11 to almost 14.

 

Chris

[mikebse2a3]

Chris I just wanted to say I appreciate the time and effort you put into your threads/post and always a lot to think about with good source links in many.

 

miketn

[ClaudeJ1]

Chris I just wanted to say I appreciate the time and effort you put into your threads/post and always a lot to think about with good source links in many.

 

miketn

I concur. I just ordered the two CD's that he recommended. Redbird and the Sheffield Prokofieff one. Can't wait to try.

[mikebse2a3]

The bottom line is that your setup needs to be calibrated to flat response if you're unmastering music. Knowing the effort put into creating really good unmastered tracks, I want my finished tracks to be able to be used on any setup in the future, not just just the setup that I'm using presently to unmaster. I think a lot of people don't take the time to do this part well.

 

I would emphasize also IMO that smooth, consistent, controlled polars from the loudspeaker would be a vital quality adding to neutrality from the reference system when unmastering tracks because any loudspeaker that doesn't have that quality is adding it's own coloration to the reproduction from the recordings.

 

 

miketn

 

 

miketn

[Chris A]

Part of the "salt and pepper" comment, above, is meant to parallel the "salt-and-pepper EQ" practices that Roy has demonstrated in our presence using Jubilees in different configurations, i.e., the K-510 vs. the K-402 with the various compression drivers. 

 

Note that the polars of the Jub bass bin get progressively narrower above about 210 Hz up to its cutoff above the bass bin low pass frequency (Mike: you know this...it's stated again here for those that don't).  This will affect the 210-~550 Hz band, which is right in the lower middle of the midrange.  This is about the only polar consideration for two-way Jubilees. 

 

However, other Klipsch loudspeakers have even more polar considerations:

 

1) Khorn bass bins begin to experience polar narrowing/lobing about an octave below 210 Hz and even greater lobing than the Jubilee bass bin above that point due to the wide-set horn mouths, as documented in Roy's and Paul's JAES Jubilee bass bin article.  The K-400 midrange loses vertical polar control at about 1.8 kHz, and horizontal polar control at about its high pass crossover point (~400 Hz), which leads to a large polar mismatch at the midrange-bass bin crossover point.  Similarly, there is a vertical polar mismatch between the K-77 tweeter and the K-400/K-55 midrange, and likely a horizontal mismatch the crossover point (with the tweeter polars exceeding the K-400 midrange horizontal polars).

 

Vertical separation of the bass bin, midrange and tweeter horns of greater than 1/4 wavelength at the crossover frequency also causes polar distortions at the crossover regions, especially the extended crossover band (no low pass electrical crossover on the midrange circuit). 

 

2) The La Scala and Belle have fewer issues with the bass bin polars and vertical separation with the K-400 midrange, but have no real output between 30-60 Hz compared to the Khorn, which is a huge issue when EQing.  The other existing tweeter-midrange polar distortions are basically the same as Khorns. 

 

3) The Cornwall has narrowing polars of the bass bin up to 800-1000 Hz, which will also form a disturbance with the midrange horn horizontal and vertical polars, in addition to the time misalignment of the woofer--midrange horn.  Its midrange-tweeter polar misalignments are similar to Khorns, La Scalas, and Belles.

 

Etc., etc.

 

All of these issues need to be dialed-in using what I would call salt-and-pepper EQ adjustments after the major corrections.   But the major EQ unmastering corrections shown in this thread would all be necessary to correct for the heavy-handed mastering practices documented above. 

 

Chris

[Chris A]

One note on the use of Audacity to run batches of files through the same EQ curve:

 

The "Edit Chain..." and "Apply Chain..." commands under the File menu allow you to run the same commands on a group of files in quick succession, saving each updated file as it completes the tracks one by one.  Below is an example "EQ" batch file available from the "Edit Chain..." command under the File menu, that runs the default EQ file last used within Audacity to edit and then save one or more music tracks:

 

 

The commands to edit this chain macro are found under the "Insert" button, then you can specify whatever parameters for each command step after you double click the command from the list of available commands.  Here, the command file applies the "unnamed" (i.e., the last EQ file used), then saves the file, closes it, and moves to the next music track in your list.

 

Once you've created your chain macro and saved it, you can then go to the File menu again to run it, using the "Apply Chain..." command, then select the music files that you wish to convert.  Once you select the music tracks to edit and press "open" on the pop-up menu, the batch sequence runs. 

 

The point that I wished to make...is that Audacity will save all the edited music tracks under a sub-directory it automatically creates called "cleaned".  In order for you to replace the old music tracks with the newly updated/edited batch-produced tracks, simply highlight the tracks in the "cleaned" directory and pull them into the parent directory which holds your original music tracks, replacing the old tracks with the new ones.  I found this step to be a "gotcha" when using the chain command for the first time, then going back and using foobar2000 to play the tracks.  Foobar2000 doesn't know about the new edited tracks in the newly created "cleaned" sub-directory until you either copy-and-replace them into the parent directory, or re-run the search command within foobar2000 to pick up the newly edited music tracks within its database view, thus duplicating the tracks in the foobar2000 database, one old file and one new edited file.

 

Chris