DolbyA decoding feedback -- 'feral' examples (yes/no)

[pkane2001]

23 hours ago, sandyk said:

HI Lucretius 

 Just for one example .

 I find that the decoded version of "ONJ-Take Me Home Country Roads" has the chorus sounding much cleaner (less harsh sounding) and clearly defined. When her voice kicks in it is a little softer and purer sounding , with noticeably improved separation between elements for the rest of the track. This version sees me wanting to hear the whole track.

(Admittedly though, I have had quite a bit of practice listening to John's various examples)

 Regards

Alex

 

There appear to be a couple of notch filters, in addition to a slightly lower level above 8.7kHz and a couple of unexpected 2dB dips at 8.9k and 9.8k:

 

Here's the same plot with the undecoded file on top:

[pkane2001]

6 hours ago, John Dyson said:

 

 

Okay -- the demos with the 'K1' and 'K1A' suffixes do not have the bumps that you are talking about.  Also, I spent more time in setting the calibration level -- anyone listening should get the sense that every last bit of detail on the recording is being reproduced (in the flac version.)   All of the errors were in my sox script and the calibration setting, not in the decoder.

 

In the K1* versions, I applied no sibilance correction (there is often a bump in the 9kHz range which is artificial sibilance -- in the leaked DolbyA material.)   I don't know if the 9kHz bump is intentional, or it results from enhancement causing a 'friction' distortion between the HF0/HF1 bands in the DolbyA HW encoding.  (The HF0/HF1 bands actively interact in the 9kHz range -- the dance must be precise.)

 

The calibration change brings out a lot more detail also.

 

The detail in the results are closer to what a master tape will provide, but still the leaked DolbyA consumer stuff is not 100% trustworthy, so sometimes there is a little excessive 'love' added.   Significant phase errors can be troublesome even though the DHNRDS uses phase insenitive detectors, still too much EQ done by the 'mastering' can cause phase shifts with untoward effects.

 

Listen to 'please mr please' or 'Let me be there' -- very closely.  Methinks that every last bit of detail in the recording (modulo the 16 bits and 48k) is reproduced there.  (There is a certain 'sweet' sound that happens when the DA decoder is set correctly.)  When the HF0/HF1 dance lines up perfectly -- the sound is astonishingly good.  (It is sometimes tricky to get that precision on feral material.)

 

https://www.dropbox.com/sh/d2dyuonx8fk3pwv/AAArjQgYoo_Skpy8YSiOuxpha?dl=0

 

John

 

 

Hi John,

 

K1A Country Road is a different portion of the track, makes it harder to compare to the undecoded and earlier versions.

K1 version does look much cleaner now (blue is undecoded):

 

Can't tell yet if I can hear the difference, as I only listened to these on my Mac built-in audio/speakers.

 

[pkane2001]

6 minutes ago, John Dyson said:

Maybe I should explain the defects in the original ABBA version -- and what was done to fix it.

 

If you listen carefully to the highs in a lot of the original ABBA versions -- the higs are at least partially existant, but they happen AFTER the initial impact producing the transient.   So, it is something like a group delay, where the highs are significantly delayed on the original versions.  A good example is that the sibilance in the original vocals is kind of 'sweet', but should have happend some milliseconds before it should actually appear.   Ring Ring is an especially egregious example, but the original might also have had a few dB of HF suppression.  By doing a flat and more time-correct remastering -- the signal brightens up (maybe too much.)

 

* to me, the original 'S' sounds are 'almost' like a "thuhs" instead of a "sthuh"...  That sure sounds like a group delay in the original signal.

Tretow might have been trying to compensate for some suppression in brightness somewhere, and that is probably where the strong brightness comes from in my versions.  He pushes it so far that the 'S' finally comes through, but only after the vocalization had happened.

 

I don't think that the high frequency content is delayed as severely as the envelope (technical term) of the highs are delayed, probably SEEMING to produce a kind of group delay, but not really.  This might result from 'gain control distortion' in the original.  (or modulation distortion -- I am not sure.)

 

The 'remastered' versions resynchronize the impacts and the highs -- therefore brightening up the sound.   The result might be a little too bright sounding, but it comes from the original recording once the 'S' or other high frequencies are happening at the correct time. That is where a good technical description of what might seem too intense, so itcan help direct me to bring the signal and the envelop back into a more subjectively desirable sound.

 

John

 

 

John, if you have an example of a DolbyA processed recording and the original, not messed up version, it would be interesting to run it through some analysis to see the variable group delay/phase differences and dynamic compression effects. 

[pkane2001]

9 minutes ago, John Dyson said:

I can do a pure decode -- but they appear to have cut the middle frequencies.  I'll do a totally raw decode with zero modifications other than to EQ the input to the decoder so that there wiil be no other modications.   The decode has to be presented a correct DolbyA signal or really sounds bad.   I have never had access to the before-DolbyA versions -- because they probably don't exist anymore, but I can do a very raw approximation -- and will post it in about an hour.   I'll do 'RingRing' and one other.

 

For me -- the original 'RingRing' is horribly messed up.  I even have an ORIIGNAL vinyl version -- it is as bad or worse than the one that I demoed!!!

 

Listen 'carefull'y to the orginal 'RingRing' -- I have CD versions and ripped versions, they all have that odd sound.  The vinyl version is typical.  The RAW decode is temporally lined up, but still has an overly tight sound to me -- maybe just my own taste.  But what I have included here is as clean and simple as possible.

 

John

 

01. ABBA - Ring Ring-RAWdecode.mp3 2.1 MB · 0 downloads

 

I was thinking of comparing the true original to the encoded, and the a decoded file. Unfortunately, comparing an encoded version to a decoded one doesn’t tell me how close the decode is to the original. Also, MP3 encoded files will not be good for precise phase analysis, as the compression will also mess with phase and frequencies.

[pkane2001]

1 minute ago, rando said:

@pkane2001

Again I feel dumb for making you add mp3 to DW while lightheartedly exploring typical audiophile meandering through a rough patch on the way to betterment.

 

That was a good suggestion, even if I took your suggestion seriously    I’ve added support for a few other formats since then.

[pkane2001]

1 minute ago, John Dyson said:

I can provide full flac versions of the results.

The 'originals' wont' do you much good, unless I give you the DolbyA input.  I can give you those also.

 

Just tell me, and I'll put together a package.   (The source material MUST be corrected to be DolbyA -- very little consumer released material hasn't been damaged in one way or another.)  I do run into pure DolbyA from time to time -- I do have some ABBA that IS pure DolbyA, but it is damaged by generation loss, so what I do is bring the best copies up to the standards of the 'damaged' copies, therefore getting the best quality of each.  (Proper EQ and best sound/least generation loss together.)

 

My process for finding clean material that CAN be corrected is very tedious.   Out of approx 10 CDs of each album, I have picked out the least damaged DolbyA, and then applied the corrections that cause the DolbyA decoder to be maximally happy.   ABBA is the most difficult group because they have purposefully suppressed the midrange, which then makes the DolbyA decoder do weird things.  On most recordings, they do things like a simple 3 or 6dB at 3kHz (sometimes a few variations of that.)   ABBA is NOT like that -- as I have proven by the actual, rough sounding DolbyA copies.

 

John

 

 

I’m  not sure I can get anything useful from an encoded file compared to the one you decoded. I can measure the differences, but there is no way for me to decide if these changes bring the sound closer to the original or taking it further away.

 

Assuming it’s just a subjective preference test, then yes, I liked some of the latest files you decoded compared to their encoded version. But I tend to not trust that my personal preference will always be for the more transparent, authentic reproduction.

[pkane2001]

6 minutes ago, John Dyson said:

Specifically, the original on 'RingRing' is so astoundingly poor -- the improvement in the raw decoded version doesn't need much commentary.  Since I did a RAW decode, and the DolbyA is apparerntly destined for vinyl, it might a good idea to do a 1.5dB cut at 9kHz (q=0.707.)   The results that I posted are totally absolutely not modified (other than as I specified above.)

 

On DolbyA -- phase is pretty much meaningless -- however, the DHNRDS is 100% linear phase except for below 200Hz.  It must be a standard IIR filter at the 74Hz frequency because the Q needs to be 1.070, and it causes a needed peak and phase is important in the 200Hz region and above.

 

If there is variable group delay, then phase change is not linear.

[pkane2001]

7 minutes ago, John Dyson said:

The goal is to MATCH the DolbyA so that the temporal relationships are maintained.   On a true DolbyA 2.8kHz and 8.8kHz filters are near linear phase, so linear phase FIR filters (carefully crafted to match the response curves) are a valid replacement.   The 74Hz filter is NOT linear phase, and must have a Q of 1.070 to match a DolbyA.   These are one aspect of a DolbyA that I had to replicate, or the results/phases will NOT be correct.   Above about 200Hz, the phases straighten out, and the time relationships are coherent (don't have offsets.)

 

The DHNRDS is totally phase stable and has no unintended delays causing decoding problems -- however, the initial/basic phases must match -- and I did that.

 

The DolbyA DECODING doesn't very accurately match the dynamics of the ENCODING -- so there are weird little phase shifts that wobble all over the place, thereby causing decoding fog.

 

The DHRNDS has no such 'wobbling'.   It is dead-on accurate (really!!!)

 

John

 

 

Got it. Interestingly I’ve seen a phase distortion produced by some DACs and ADCs that appears like an oscillating waveform. Could be some error in the filter implementation or an actual clock being modulated by a frequency. Weird looking!

[pkane2001]

2 hours ago, PeterSt said:

Frank may make a fool of himself everywhere throughout, but not in this thread he is. You, on the other hand, twist things slightly around by making a fool of him..

 

This is way more difficult for Frank, who just received my very first upvote. It was hard to do but I did it with my heart anyway.

 

Despite the incessant posts by Frank about his way being the only right way to audio bliss, he has a really good understanding of audio and digital processing. I wouldn't discount out of hand what he hears or finds, even if he is using a laptop.

 

Personally, I heard some improvements on some of John's decoded tracks, and no improvement or even slight worsening on others. In the most recent decodes, I like the decoded result a bit more, but the differences are not that significant for me to say definitively.

 

What concerns me more is that this seems to be an ad-hoc decoding process, with John deciding what sounds good on a track by track basis, when to apply EQ and when not to, etc., rather than having a single, demonstrably accurate decoding mechanism that works on all tracks (likely I'm just misunderstanding the process).

[pkane2001]

5 minutes ago, John Dyson said:

Absolutely NO processing is being done on a track by track basis.  That violates the rules.  There is one track that I have been tempted 'Supertrouper/Supertrouper', but I have not demoed it, and I think that I have the  problem solved.

 

This is NOT an ad-hoc process -- it is tedious, but not ad-hoc.

 

The decoding is definitely not ad-hoc, however ALL mastering is by-ear.  Frankly, I have a process and a set of algorithms that I use based upon certain problems.  For example,  have phase twisting algorithms/etc, intended to undo the damage of Aphex 'distorter' type messes.   I have vinyl-emulation EQ for material that was mastered for vinyl, but being played on CD -- these are procedures, not so much 'tweaking.'

 

Here is a before/after on ABBA -- and shows the cr*p that I am trying to get rid of...

sos-orig.flac 1.56 MB · 0 downloads sos-remaster.flac 1.31 MB · 0 downloads

 

Sorry if I misunderstood. I thought that was part of the process. I highlighted what I thought was the 'tweaking' part of the process in your previous post:

 

Quote

The biggest skill on using the decoder is understanding the EQ being used to 'fake decode' and to find the correct calibration levels.  Finding the correct calibration levels is a combination of knowing what the typical tape calibration levels are on the DHNRDS scale of measurement and also knowing the problems to listen for.   The 'listening' problem is also an issue with actual DolbyA units.

 

Dolby claimed that 1dB accuracy is all that was needed -- but that isn't really true when trying to get results as desired today.  I have heard significant DolbyA HW decoder surging on commercial material.  Even on a true DolbyA unit, 0.25dB of accuracy seems about correct.  However, since the DolbyA HW doesn't have a super-accurate means for balancing the meter, for the best quality, DolbyA HW DOES require 'tweaking' and listening.

 

So, how much of the process just a mathematical transform, and how much tweaking it until it sounds good to you? I don't mean to push back on this, but I'm pretty much confused by what is being attempted here.

[pkane2001]

7 minutes ago, John Dyson said:

There is no real criteria that is based upon 'sounds good'.   There IS a judgement call, and the attempt is to remove what 'sounds bad'.

On ABBA, there appear to have been two levels of signal manipulation, and I am only looking at the manipulation that can be reversed with minor temporal shifts (or in really desperate cases, EQ.)

 

In essence, I see my effort as a problem solver, not creative musical talent.

 

Let's do a boost at 6kHz to gets some more presense -- BZZZTT..  That kind of thing AINT GONNA HAPPEN.

 

This is more the process:

I am hearing some strange temporal problems -- well, they are usually at 3kHz or 9kHz (sometimes in between), so maybe a bit of a phase twist will help.  (usually some kind of incorrect arrival of various sound components.)

Or

There is a sibilance problem, lets do a twist at 9kHz and maybe one at 12kHz if it still needs help.  (most people would just do a brutal HF null of some kind, but often it is a temporal problem, not a peak.)  Shortening the sibilance is usually safer, and it doesn't take much.

Or

The vocals have a messed up relations in the vocal tone....  Ahhh -- the ABBA trick of a 'somewhere below 3kHz LF cut' to fake-brighten the sound...

 

Some of these remedies are made possible because the DHNRDS is temporally dead-on accurate, so when there are little errors because of brutal EQ technqiues in the past, they can be corrected without variations from cut to cut.   The little dithers in timing dont' get magnified when using the DHNRDS, but instead there is actually some cancellation of the effects of original DolbyA encoding.  (That is something that I do not advertise, but is evident in the decoding quality.)

 

If using an original DolbyA -- many of the temporal problems are fuzzed away, and then the recording has a bunch of blobs or over-enhanced components because of some kind of sluggish dynamics overshoot.  That kind of thing DOES NOT happen on the DHNRDS.

 

THE SOS EXAMPLE ABOVE shows the extreme sloppiness that I am trying to remedy.

 

 

John

 

Alright, John. I still see this as subjectively driven by what you perceive to be some problems in the original track. You may be right that these are real problems, but there is no way for me to know, other than 'it sounds bad without decoding, and sounds better with'. In effect, you are performing a remastering job with the intent to make it sound better. That's fine, but that's different than saying you are correcting some very specific distortions introduced in the original mastering process. For this, you'd need to demonstrate the nature of these distortions and that your correction eliminates them.

[pkane2001]

4 minutes ago, John Dyson said:

 

There will never be an auto-corrector without some kind of amazing AI, and I am not making that representation (yet.)  It would be a good idea, but even an automatic DolbyA detector is something that seems algorithmically impossible to me (and I know everything mathematically -- in DETAIL --  that a DolbyA does.)   Maybe could detect on a restricted kind of signal, but in general, I don't think that any kind of problem like that is ALGORITHMICALLY solvable.  My techniques COULD be documented, but still require listening or the development of tools that could automatically detect the individual impairments.  We aren't THERE yet, as I am learning the various impairments in recordings.

 

Yes -- I am correcting distortions 'somewhere', whether they are from mastering or elsewhere.  Again, PLEASE refer to the flac files that I posted earlier.  My example isn't perfect either, but I tend to focus on correcting problems rather than creating new ones.  That has been a major portion of my career over the last 40+yrs -- correcting/solving problems.

 

It might be hard to understand unless you listen to the discrete example that I provided.  By listening to that -- it will give you a snapshot of my approach.  Information/context free discussion is very difficult (I cannot draw pictures in the sky for everyone reading this message to see.)

 

John

 

I get it now. It was my misunderstanding that you were trying to detect and/or fix this algorithmically.

[pkane2001]

9 minutes ago, fas42 said:

 

Perhaps incessant , but I wouldn't use the term "the only right way" to describe my methods - getting "audio bliss" has always been available for people who have enough money to throw around, and plenty of time to spend fooling around with all the bits; there have always been accounts of people with "Everest" rigs, that produce magic for the listeners.

 

My interest is in what's going on - because if one understands that, then huge shortcuts to doing the Magical Mystery Tour to "nirvana" are then available; my posts are about the steps, IME, which are highly effective, have excellent cost/benefit ratios, in creating that shortcut.

 

What's your conversion ratio here, on AS, Frank? How many people did you actually convince compared to the number of posts on the subject? Just curious