Jump to content
IGNORED

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


Recommended Posts

2 hours ago, John Dyson said:

I am so sorry if I didn't answer the question, I don't remember seeing it -- various things have been going on.

 

Encoded but no decode actually seldom happens, but instead they do an 'Encoded', but extra 'EQ' that hides the nasty, intense DolbyA sound.  So most commonly, we have these (in order);

 

A0) Encoded but EQed to sound okay.

A1) Encoded and correctly decoded.

B) Encoded but not decoded.

C) Decoded but not encoded.

 

C would seldom happen, because it usually sounds really bad for a DolbyA equivalent decoder to 'decode' uncompressed material.  Sometimes material compressed with other than DolbyA can sound listenable when decoded, but 'C' is a serious mistake.  I included a copy of 'C' above because I wouldn't have to carefully review the results. (Unreliable hearing right now.)

 

B sounds pretty bad.  It is harsh!  The worst album that I have is 99Red Balloons from Nena which has both CD pre-emphasis and DolbyA encoding.  With just the raw DolbyA, is pretty ugly.

 

How do you detect feralA or DolbyA?   It is tricky -- a practiced ear for the HF compression and that isn't very reliable based using my own hearing.  Also, you can check for hiss at the lead-in, intersyllable or fade-out -- sometimes that is a pointer to too much hiss, therefore possiibly undecoded DolbyA (in feralA form or native DolbyA.)   Also, looking at the spectogram, bring the background hiss up so that you can just barely see it for the 3k-10k range.  Sometimes, DolbyA where it was actually used for NR purposes will show a small increase in hiss above about 12kHz.  (That shows the extra 5dB of  compressionn gain for the 9k+ range.)

 

My hearing is corrupted right now, so making examples for overt comparisons might be unreliable right now ( could easily botch it.)

 

Here is the previously posted snippet/song "its gonna take some time', but decoding the alrleady decoded material -- but I used the pristine flac version for the source instead of the mp3.   Also, the material is alread compressed, so the excess expansion isn't so terrible, and the middle frequencies are essentially saturated (the gain is the maximum 0dB at apprpox -25dB input, so is basically pinned on.)   Notice that the worst of the loss is in the highs and stereo image damage.  * A true DolbyA would be worse sounding.

 

(Note that my previous decode might not have been 100% accurate, my hearing was starting to fail earlier today.  I keep on trying to bounce between two versions -- but the sibilance is better on the version that I previously uploaded, but has a bit too much 'growl'.  When cleaning up the 'growl', then the sibilance gets a 'hitch' in it.  So, I uploaded the best of several attempts -- and when listening to the entire album, the results are pretty good -- just that one song bothers me)

 

 

.

 

 

04 - It's Going To Take Some Time-decoded-doubledecoded.mp3 1.68 MB · 0 downloads

 

Thanks so much for that breakdown of the matter!

 

And no, neither you nor anybody else missed the first time I attempted to ask my question.  It simply never made it on here, as often happens with Facebook and other correspondence sites.  Like I said, I have felt the impact of the 2018 repeal of Net Neutrality - slower overall speed, websites not fully displaying, just hanging there with a spinning wheel in upper LH corner, posts not posting, etc.  Happens to me both when out using my cellular, and at home on Wi Fi.

 

People, watch who you vote for!!!

Link to comment
  • 2 weeks later...

FeralA decoding chart.

 

I have been doing some documentation, perhaps the least well known aspect of the feralA -- what is it?   I have created a simple document describing the feralA filter for the Carpenters 1971 album.  This is part of a larger document describing the parameters and possible choices in general.   The complete document gives FeralA decoding formulas for several albums - giving a more thorough pictorial idea of what is going on.   ALL feralA conversions are identical in shape, it is just the frequencies and levels.  Also, there is a limited set of choices, so 'tweaking' is not really onerous as there is a limited set of choices in the values.  (For example, some of the filters are almost all the same attenuation.)

 

One important comment:  the 250Hz offsets between the M and N groups are critical.  It seems like that offset might have been used to make it difficult to cleanly decode the material.  The only reason why I could determine that offset is that the DHNRDS is scrupulously clean.

 

John

 

 

FeralDocB-Carpenters1971.pdf

Link to comment

The feralA thing is pretty much proven, and finally I think that I understand all of the various filters.  Implementing the feralA version of the DHNRDS is mostly a job of accounting for all of the modes needed, and I think that I have a handle on them.

 

I just put together a coherent set of decoding modes for the acutal DolbyA product and will be shipping the latest/cleaned up version for an upcoming speech/convention.

 

After the new release is done, I'll be adding some command line switches along with some defaults for feralA decoding -- gotta simplify decoding feralA.   Been trying to figure out ways to make the DHNRDS work well without it being so tweaky and dependent on SOX.

 

First, 44.1kHz CDs cannot be processed with high quality at an internal 44.1kHz rate.  There are zero quality limitations when running the decoder at 60kHz through 74kHz, and since 66.15kHz is a good sample rate with integer related ratios, I'll use that.  The output will be selectable rates of 44.1k, 66.15k and 88.2kHz.  I will not be doing any in-between rates -- keeping all of the conversion simple.  Because of its architecture and some mathematical behaviors, the very cleanest/least filtering kind of conversion can be done by the DHNRDS, and there will be absolutely no benefit in using external rate conversion.   For example, normal rate conversion has to be good for all kinds of sources with all kinds of nonsense in the audio.  The output of the DHNRDS is just as pure as the input, so assumptions can be made about spectral content.

 

The DHNRDS only does .wav files, and will NEVER output 16bit .wav files even though there is an internal option to do that.  The DHNRDS will accept the somewhat normal 16bit signed, 24bit signed and floating point as input.  I might consider adding 32bit signed for symmetry, but it is seldom needed, and floating point is usually better because of dynamic range.   For output, the DHNRDS will continue to support 32bit floating point or 24bit signed integer.

 

Outside of the deep technical issues, the default user quality modes available will not include the anti-MD capability.  That limitation will be in place for market differentiation.  Also, there will be no specialty controls for out-of-spec DolbyA units as source.  For released, feralA material, it is unlikely that there will be an eccentric, special purpose A301 encoded recording anyway.

 

ADDED:  there will be 48k input for 48k/96k output, and 88.2k input/output and 96k input/output.   Note that the freq response is limited to 40kHz in all of the higher sample rate inputs (88.2k and 96k input.)  The DHNRDS is more flat up to that frequency than an actual DolbyA unit.

 

Everything will be KIS, and all of the advanced, super quality production modes will be available on the full DHNRDS product.

 

Also, this version of the decoder won't directly work with normal DolbyA files, but can be twerked into doing it -- just not as convenient.

 

No matter what, every kind of FeralA that I have encountered will be decodeable.  The results will sound SUPER good.  This upcoming  version will not be payware.

 

I don't know when it will be ready, at worst you might lose your $$$ deposit :-).   (that is, $0.00 :-)).

 

John

 

Link to comment

I have something that might be fun to compare -- two versions of the beginnings of a Carpenters song -- one DA decoded with anti-MD/IMD and one without.  When listen -- notice the difference in seperation between the sound sources.  On the raw version, the bonnn bonnn - bonn bonn -- bonn bonn, and then the bass merges in...   On the raw version, the bass merging in sounds like a single sound with the original bonn bonn -- bonn bonn....  On the no-imd version, the bass sound keeps seperated.

 

This is a simple pre recorded example of what the anti-md and anti-imd do.  No DolbyA unit can maintain the distinctness.  (In fact, a normal gain*signal type implementation and no anti-IMD can NOT keep the signals from intermodulating.).  Interestingly, the spectrums superficially look so similar as to be indistinguishable.  It is those little intermodulation sidebands probably sitting too close to the main peaks that make the difference.  It aint easy to get rid of the intermod...

 

* The raw demo provides close to a true DolbyA sound.  The demo is created by using --basic mode of the decoder, which bypasses the anti-IMD front end processing, and reverts to a simple gain*signal gain multiplication.   The --xpp mode, which is the more clean no-imd version, there is anti-IMD processing in the detector, and the gain*signal is more of a special, very complex operation that mimicks a gain*signal, but compresses the distortion sidebands into a place where they are less audible.   Basocally, much of the modulation is suppressed.

 

*  We have been hearing that IMD since day one, and is a portion of the 'DolbyA fog' that kept the perfectionist recording engineers from being happy about using DolbyA.

 

John

 

carp-noimdormd.flac carp-raw.flac

Link to comment

Have an additional announcement, I guess I can make publically.   The feralA mode will be an official part of the DHNRDS DA.   It has been cleared by the other team member, and in fact it was by his initiative.   I had expected to keep it as a cute little feature for consumers, but apparently it is being looked at by pros now.

 

John

 

Link to comment

About the two comparisons between modulation and no modulation distortion.   I did a difference operation, and the variations in the gain can be heard.  Interestingly, I didn't bother doing the comparison like that before, but what is heard, however 'normal', are actually modulation effects which distort the relationships in the sound.

Because the 'raw' version allowed inter-modulation in the sound, the sounds modulated each others levels.  There are specific frequency ranges where the modulation is worse on DolbyA attack/release, but they are still prominent enough to cause problems.

 

Looking at the spectrum of the differences -- the bands are pretty much what I expect.   Around 1kHz because of the fastest attack/release speeds, and 9kHz because of the two bands (3k-20k and 9k-20k) that intermix because of all kinds of delays and stablization that happen in the feedback loop.  Also, at the lowest frequencies because of the attack/release.   The spectrum of the differences pretty much match what I'd expect.

Screenshot from 2020-02-03 08-57-07.png

Link to comment

Good news -- the FeralA decoder is now an 'official' product.  Instead of it being a 'hidden' add-on for my jollies, it will actually be an official feature.  My project partner asked for it -- even though I own/write the software, he is a co-equal on the project and makes the strategic decisions.  He asked for the feature, recognizing that it *REALLY* works very well.

 

This is a first step for feralA being fully legit and decoding the material will be within reach of ANYONE with an Windows box or Linux box.   I have also heard that the decoder works on Windows emulators on the MAC.   I will probably make the feralA specification public in some ways.  Early drafts (super early drafts) have already been passed around, but the early drafts both have errors and need better artistry.   Additionally, I might make the source code for the important parts of the --basic level decoder freely available for other developers.   This --basic level of quality is close to that of DolbyA, but 3X faster, and without needing HW interface.   The anti-MD/anti-IMD -- which makes the sound smooth and pretty, will probably be kept as a trade secret until the subject of an AES talk or something similar.

 

So, I am planning to make the *raw* technology openly available for doing a TRUE DolbyA compatible decode.  Replicating the technology is not a feasable  weekend project for anyone, and having the technology up-front would bypass at least several months of work/testing/etc for plug-in developers.   Basically, would make feralA decoding easy to implement.   Using my source code,  basic DolbyA compatible decoder can function approx 3-4X faster than realtime on one Haswell core, and probably be made to run 10X faster with careful optmiization.   My source has not been optimized in the basic decoder sections*, so there is opportunity for speed improvement.   (All of the hard work/optimization has been done in the anti-IMD, anti-MD processing.)  As a plug-in, an accurate decode might not be much slower than a normalization operation.  (Yes, it is easy to do a simple filter, but a DolbyA decode can not be emulated by a simple filter & this would be a full 3.5band expander, with proper handling of the HF0/HF1 overlap/etc, strange expansion curve, etc.)

 

* There are some design hacks in my code, esp the front end bandpass filters, and there IS opportunity for improvement and competition.

 

Anyway, I am going to make my version of the feralA decoder as easy as possible to use.  All of the detailed filter calculations will be automatic, and I plan for the interface (unfortunately command-line) to be simple, yet complete enough to decode all feralA presented to it.

 

The tricky part is to implement the features and the flexibility, without the user needing to specify every little item in the stack of operations.   A side benefit of the internal feralA filtering is that the infrastructure for the conversion will be all double precision floating point instead of just using single precision floating point interim files.  Also, there will be less need to use pipes on Linux/Windows or temporary files.

 

The quality of the full-bore feralA decoder will be better than anything that I have shown so-far.  Hopefully, it will be feasable for 3rd party developers to create DolbyA HW quality level feralA/DolbyA decodes for the cost of writing a few lines of C++ code.   No more fake DolbyA decoders, and better-than DolbyA HW quality decodes (well, not better in every way, but definitely better sound.)

 

Approx ETA for the feralA decoder:  2-3mos.

 

John

 

Link to comment

John-

 

I've very much enjoyed your commentary and appreciate all the hard work you've put in. That said, most of the examples you've cited-- e.g., The Carpenters, Abba -- are not bands I have much interest in. At the same time, I know you've made the point that the undecoded Dolby A-to-Cd problem exists for a lot of music recorded in the 60s and 70s.

 

So, my question for you is: can you comment on how prevalent this issue is for mainstream bands that I (and I'd suppose a lot of the ok, boomer audiophile cohert that come to this site) are interesting in? I'm talking Beatles, Stones, Zeppelin; that ilk . (Which also raises the question as to whether Dolby A usage was prevalent in England to the same extent as here).

 

i guess i should also extend my question to the stuff recorded in US in the 70s and into the 80s-- Lennon's solo albums after Imagine, Steely Dan, Billy Joel, Talking Heads, Fleetwood Mac.

 

Thank you for any light you can shed on this. Anyway, over to you and your deep expertise. 

Link to comment
19 hours ago, garrardguy60 said:

John-

 

I've very much enjoyed your commentary and appreciate all the hard work you've put in. That said, most of the examples you've cited-- e.g., The Carpenters, Abba -- are not bands I have much interest in. At the same time, I know you've made the point that the undecoded Dolby A-to-Cd problem exists for a lot of music recorded in the 60s and 70s.

 

So, my question for you is: can you comment on how prevalent this issue is for mainstream bands that I (and I'd suppose a lot of the ok, boomer audiophile cohert that come to this site) are interesting in? I'm talking Beatles, Stones, Zeppelin; that ilk . (Which also raises the question as to whether Dolby A usage was prevalent in England to the same extent as here).

 

i guess i should also extend my question to the stuff recorded in US in the 70s and into the 80s-- Lennon's solo albums after Imagine, Steely Dan, Billy Joel, Talking Heads, Fleetwood Mac.

 

Thank you for any light you can shed on this. Anyway, over to you and your deep expertise. 

Of course, DolbyA was used a lot for NR on the analog tape recording scheme.  It started being replaced by DolbySR in the late 1990s.   DBX had sometimes been used, and other NR system called Telcom C4 was sometimes used in Europe.  From a theoretical spec standpoint, all of the mainstream systems other than DolbyA were better, but each one had other sorts of defects that made DolbyA  the generally safest system to use.  I could go through all of the advantages and disadvantages of each system, but that would be a big diversion from your question :-).

 

*  DolbyA nor EQed DolbyA as feralA was never originally intended to reach the consumer.  This is the reason why I coined the term 'feralA' as it has escaped captivity ;-).

 

When doing multi-track recordings, with the tracks being so small, even though they ran the tape fairly quickly, there was a significant amount of noise being added to each track and for every copy made.  Even the small amount of NR from DolbyA (between 10-15dB of very effective NR) really helped.

 

* I DO NOT KNOW WHY feralA was ever created as an alternative to properly decoding the material, but I do have some plausible explanations.

 

From what I have tested & measured, almost all digital distributions of *POP* recordings originally made before the early 1990s seem to be available primarily with the feralA encoding (that is, EQed DolbyA.)  After EQ, the normally shrill, intense DolbyA sound is tamed enough to simply sound like HF compression.  *the challenge of obtainiing the original sound of the recording from the consumer feralA source material is undo-ing the feralA EQ, then decoding the resulting DolbyA signal is simple.

 

Out of the albums that you mentioned, my only Lennon album is a copy of 'Band on the Run' (unlimited version) from HDtracks and is feralA encoded, my copy of 'Rumors' is feralA encoded.   Every 'Dreams' from Rumors example snippet sent to me is also feralA.  I have heard that some people do have non-feralA, properly decoded copies of Rumors, and they have used that as a basis of comparsion for some of my feralA decoding attempts.  I have recently decoded the entire Rumors album, and gotten very plausible results.  (It is tricky to satisfy audiophiles, because the EQ is different from album to album, and I don't know what Rumors was supposed to sound like.)

 

I haven't checked Billy Joel, even though I liked his music when it was contemporary, and in fact one of his CDs was my first one, I don't have any Billy Joel in ripped form, and don't believe that I have his CD anymore.

 

Up until recently, very little vinyl was feralA, and was almost definitely properly decoded material.  However, more and more recent vinyl examples are also feralA.  I suspect the reason is the expectation that the sound be similar to the existing digital copies or possible access to the original, properly handled analog masters.   FeralA decoding is not straightforward without a program that can do calculations, and without a reference standard, it is probably easier nowadays to produce vinyl with feralA encoding.   A lot of digital masters are likely in feralA form.   The encoding is too common to not have been a standardized practice in the industry.

 

Just checked my copy of Springsteen, 1984 Dancing in the Dark, track 11, Dancing in the Dark, is definitely feralA.

 

When checking for feralA, other than listening for the tells like compressed high end, also just doing a decode and failing the decode with the surging sound of bad dynamic range expansion is the biggest veto. 

 

--------------------------------------------

About feralA -- I am in the middle of writing a complete spec...  I have passed around earlier versions, not sure if I am going to distribute the final version.  Will probably do so after it is formally presented somewhere, going to offer it to my project partner for presenting at AES or somewhere else before freely distributing the feralA spec.

 

--------------------------------------------

Attached is a snippet of a feralA decode of the HDtracks unlimited version 'Band on the Run'.   It is a swag, guess because I do not know what the recording sounded like as going into the DolbyA/tape recorder.  There are well defined, settable parameters for decoding -- but I know that this example is within 5 more minutes of being perfect (esp if a reference copy is around.)   The EQ is not approximate, but is in discrete steps, not like using a tone control pot or anything like that!!!  There are 0.5 dB  (sometimes 0.75dB) steps for part of the EQ, and not sure if my results shouldn't have a smidgen (0.5dB) more attenuation per step over the 3k -> 12k range.   I believe that the MF correction IS correct (between 500 to 3kHz) because the effects are pretty obvious when in error.

 

* If you have a feralA copy of Band on the Run, the difference in HF compression can probably be heard.

* Added some Rumors snippets.

 

One more thing -- feralA will sometimes be 'more loud' and sometimes more bright than the equivalent properly decoded material.  FeralA is definitely a component of the 'loudness wars'.

 

John

 

 

 

01-Band on the Run.mp3 Fleetwood Mac - Rumours - 08. You Make Loving Fun.mp3 Fleetwood Mac - Rumours - 03. Never Going Back Again.mp3

Link to comment
  • 3 weeks later...

The PM groupe on the DA/FA decoding might have an interesting message about a unique ABBA example.

If anyone is interested, I can post snippets here for those who aren't participating in the PM group...  (This is for those who might not have looked at the group in a while.)

 

John

 

 

Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now



×
×
  • Create New...