Jump to content
IGNORED

'FeralA' decoder -- free-to-use


Recommended Posts

On 7/4/2021 at 6:13 AM, PeterSt said:

John, pardon me ... I am trying to read texts like these with the required interpretation hence context of your hearing. ... Even if I read this 20 times, there can't be truth in what you are saying. If you can not hear well, you can not hear well, and you would not be able to judge this (see quote) just the same. This also comprises the decoder being perfect for "years" and as many times as days exists in those years.

I hope you don't need these texts for your own motivation because IMHO it really is not useful.

 

Just saying (and derived from your conversation with Jonathan - not really with me because I learned how to deal with it (I think)):

 

Generally, in parallel two streams exist:

1. The user making (IMO) useful comments, like no running water in whatever song to be heard, while there should;

2. You going your own way after (!!) such message, taking out the reference of that message, possibly claiming that no versions will be lost because you save them all (which I believe). Later, you may come back at the issue (like you do here with Lamb) but nobody has a reference any more because you are thee+ versions further.

 

It all reads like a trick (read : sales trick while you sure are not trying to sell anything).

 

If you ask for help, again IMHO, you should react to that before proceeding with your own ways.

The best example could be me myself (and you). The responses are countless, but in the end are useless for myself because in no circumstance that I recognize, you attacked the issue and a new version accompanied with a "so Peter, did this help ?!?" ever is in order. What does happen all the time is a "OMG, again all changed !" from my side. Do you realize this ?

 

Somehow I started this post with telling that you can't know yourself what "is" correct. You should know by now because of the number of days in those years, etc., right ?

Mind you, I work quite similar with my own customers. If day tell me that running water has disappeared, by response would definitely be "Okaay, then we'll get that back for you". But I could also say "and what about that bass then ? is that for the better now ?".

 

 

I hear what you are saying, though first: I don't consider myself a customer rather @John Dyson has postulated a theory that somehow all CDs have been released with FeralA encoding either by incompetence or through a diabolical scheme on the part of the media publishers to give us crappy sound on CDs...

 

So is this true? Many of us first heard CDs with grainy sound back in the old days and that LPs sounded "better" so is the cause Feral A?

 

1) you among others including @Miska have long maintained the benefits of computer based upsampling and these benefits have been **proven** on the basis of your XXHE and @Miska HQPlayer software. Moreover there is objective evidence of the improvement. 

2) John has another theory and presents us his software as a way to test his theory: I have spent a great deal of time testing out his hypothesis and have very specific examples that I have asked about, not merely "I don't like it"

 

In my mind, in some cases I can hear a benefit e.g. less background hiss and less "harshness" but in other cases I hear a distinct lack of low level detail. John tries to tell me that my hearing is broken and this preference for low level detail is simply that I'm used to compression.

 

Err ... sorry I want running water to sound like running water, and twinkles to sound like twinkles.

 

So at this point my conclusion that there is widespread FeralA encoding is not substantiated. There may be examples of it, but when @John Dyson asserts it is there based on his hearing, my conclusion is that his hearing is not sufficient to tell whether someing is "Feral A" encoded ... if there is an objective way to tell then I am, as they say, all ears.

 

I could repose my question very specifically:

1) are Genesis: "Lamb Lies Down on Broadway" and Yes: "Close to the Edge" (studio versions) Feral A (dolby A) encoded?

2) if the answer is YES, then is there a decoder that can remove Feral A while preserving low level details?

Custom room treatments for headphone users.

Link to comment

Answering the comment about wanting to maintain the details while decoding:

 

Yes -- the 'Lamb' recording from CD or download, as you submitted are FA.

No -- the decoder will try to revert the recording to the original, pre-FA version.  You will hear the details similar to what the LP would have provided (but without vinyl noise and the attributes typical of the necessary analog electronics.)

 

Think of the FA decoder as a descrambler, and only does about 10dB, perhaps 15-20dB of downward expansion at most.   The major function is of a descrambler when it can manipulate well over 100dB per second (up and down perhaps +-10dB or more, per second on 1,2,3,4,5,6,7,8 layers.)

 

IMPORTANT:  After decoding, the details are still in the recording.

The details that you expect to hear are reverted to the level in the recording (per the LP version.)   Those details are likely below most peoples ability to easily hear.

 

Good news if you want the best quality (not garbled like FA), but still hear the low level details (magnified from the pre-FA version):   Use a competent dynamic range compressor after decoding.

FA encoding does a real scrambling job on the < -20dB details.

 

 

* IMPORTANT -- I have been thinking about schemes to do the 'decoding' (descrambling) while maintaining some of the compression.   I REALLY have been thinking about it because of the current sensiblities and expectations.

 

 

 

Link to comment

What a relief.   Found another LF problem that had been vexing for so long.

I was originally going to write a long description, but my writing skills suck, and if someone really wants to know the details, I'll explain privately.

 

Result:   Yet another improvement about buried vocals, and post-decoding code being simpler.  (needs to do less.)  Minor bonus, part of the design requirement:  LF is very similar when running at 5layers or 10layers or anywhere in between.

 

Bottom line:  the 2nd order post decoding LF EQ has been moved to be inter-layer EQ.   That requires re-visiting the gain/Q values of the EQ, but the freqs are the same.  Good news about the changeover -- the first engineering estimate was correct.   Basically, two final +9dB @ 50Hz and +9dB at 80Hz/Q=2.0 each.   Those have changed to: +1.5dB each, with Q=1.25 (I think the Q is correct -- 1.19 seems to make more sense, will revisit.)

 

This should  mitigate SOME of the tweakiness and indecision about the LF.   The LF post EQ design seemed to be 'fragile', but instead of doing a 'rethink', I kept beating on the wrong design like a fool.

 

The wake-up came from the LF level & character changing based on the # of layers.   Didn't notice it before, but now noticed it.   When the EQ changes vs. # of layers, that is a strong indicaiton that the EQ should be done on an 'between-layer' basis instead of at the end.

 

 

Link to comment

Following up on the idea of 'decoding' while retaining much of the FA compression.   I do have an idea, and only requires a little 'rewiring' of the decoder -- might do it in a few weeks.

 

Much of the 'scrambled' FA sound comes from the two HF bands not fully working together.   It is not the nature of DolbyA behavior that the signals should be fully  'in sync' upon encoding, because the scrambling is undone upon normal DolbyA decoding.   This 'scramble' doesn't sound terrible with simple DolbyA because it is only one layer of scrambled compression.  The major effect of doing simple DolbyA compression is that the sound is 'tinny',  sometimes with an out-of-character bass boost.

 

In FA, the HF0 and HF1 gains are not the same, very complex variations, and do not work in unision.  If the decoder could gather the HF0 and HF1 gains, the change the HF1 gain into being the same as the HF0 gain (at each layer), then some of the scrambled sound might disappear.   The FA decoder is designed to be somewhat modular, and even though that behavior is beyond the current design, I believe that the architecture can withstand it.  With the admittedly partially descrambled sound, the result might be like a 'hearing aid' version of the full FA decoder.  I might like it!!!

 

The 'trick' is that each layer needs a full FA decode up to that point:  that is, the 2nd layer decode needs the true 1st layer output.  It isn't sufficient to give the 2nd layer a modified (descrambled) 1st layer input.

 

So, the descrambling needs to act on a signal in parallel.  This is architecturally tedious because there are synchronization delays needed thougout the system, and one would  need to be very careful to keep the parallel, descrambled signal in sync with the fully decoded signal.

 

I haven't thought this though entirely yet, but it MIGHT be possible to produce an output that has the compression that everyone is used to, and also the clarity/lack of scrambling that I prefer.   The caveat is that IF I can make this simple 'descrambler' to work, the compressed sound might reveal some of the undesirable aspects of the compression.   That is, the scrambling produces confusion that might hide conceptual defects in the FA system.

 

ANYWAY -- I am already too overloaded doing the simple FA decoder (and it is NOT simple), but once that is working satisfactorily, maybe the 'hearing aid' unscrambled output might even be preferrable to me.   It might even be possible to do partial descrambles, with an improvement, but not completely cleaned up.  (one layer often doesn't sound too awfully bad.)

 

 

Link to comment
16 hours ago, jabbr said:

 

 

I hear what you are saying, though first: I don't consider myself a customer rather @John Dyson has postulated a theory that somehow all CDs have been released with FeralA encoding either by incompetence or through a diabolical scheme on the part of the media publishers to give us crappy sound on CDs...

 

So is this true? Many of us first heard CDs with grainy sound back in the old days and that LPs sounded "better" so is the cause Feral A?

 

1) you among others including @Miska have long maintained the benefits of computer based upsampling and these benefits have been **proven** on the basis of your XXHE and @Miska HQPlayer software. Moreover there is objective evidence of the improvement. 

2) John has another theory and presents us his software as a way to test his theory: I have spent a great deal of time testing out his hypothesis and have very specific examples that I have asked about, not merely "I don't like it"

 

In my mind, in some cases I can hear a benefit e.g. less background hiss and less "harshness" but in other cases I hear a distinct lack of low level detail. John tries to tell me that my hearing is broken and this preference for low level detail is simply that I'm used to compression.

 

Err ... sorry I want running water to sound like running water, and twinkles to sound like twinkles.

 

So at this point my conclusion that there is widespread FeralA encoding is not substantiated. There may be examples of it, but when @John Dyson asserts it is there based on his hearing, my conclusion is that his hearing is not sufficient to tell whether someing is "Feral A" encoded ... if there is an objective way to tell then I am, as they say, all ears.

 

I could repose my question very specifically:

1) are Genesis: "Lamb Lies Down on Broadway" and Yes: "Close to the Edge" (studio versions) Feral A (dolby A) encoded?

2) if the answer is YES, then is there a decoder that can remove Feral A while preserving low level details?

I thought I'd reply directly, answering a few incorrect assumptions, and answering the question about the FA 'Lamb' recording that we have been talking about is explained elsewhere.

 

Also, when criticising the liklihood that something is FA or not - listen for low level details, and most importantly, until you understand the FA distortion -- use headphones.  If you don't care, then don't worry.

 

Until you know what the distortion sounds like, you might need an better/more intimate connection with the recording.   I don't think that all but the very best speakers can show the difference.

 

Remember, as I have written over and over before the difference is at -20dB relative to the signal level (NOT -20dB full scale.)  Also, some experience with raw, non FA recordings might be helpful.  I do suspect that listening to FA recordings all of the time 'programs' ones listening.  

 

If you don't like the clean decoded sound -- then continued to listen to the scrambled FA sound.   Your brain is probably descrambling it anyway. 

 

IMPORTANT:

THERE IS NO CONSPIRACY, there is only an apparent business decision made a long time ago likely to keep the basic intellectual property away from the consumer (protects business interests).  Again, PURE business decision.  It might even be a decision made long ago, ending up a part of the process without really knowing what is going on?  Also, I have heard something about a consumer preference study done back in the 1970s or 1980s.  Of course, nowadays, a good ear should expect  better than FA.  Even with my 'bad' hearing, I expect MUCH better than the FA sound.  Standards are much higher than in the 1970s and 1980s.

 

Also, If anything, I don't do conspiracies -- I would have quit the effort if I didn't clearly see/hear the difference.  I can EASILY hear the difference even with my very strange, unreliable hearing.   It would have been easier on me to have quit, and would even be easier if I quit now.  I don't wimp out and do *easier* like a lot of people do.  I do 'challenging', as it is in my basic nature and part of the success in my career.  Again, the 'conspiracy' that some people conjure up doesn't exist anywhere -- it is simple business.  REAL INNOVATION sometimes happens when there is a lot of pushback.   There are lots of 'overly conventional' people out there, not looking to anything better.   That is okay with me, but I want AND WILL GET (and believe close to getting) something MUCH better.

 

About the decoder -- next release.

 

Good news -- expect profoundly better bass in the output.

I am still debating what level to decode the  demos -- all of them seem to be able to support --f10, but the benefit isn't all that great from just --fa.  The step from --fa to --f9 gives the most improvement.

 

One of the bass fixes actually removes a bug in the highs.   The HF gain curve was pushed into the wrong region by the bass pre/de-emphasis.   The correction vastly improves the 'sweetness' of the decodes.

 

 

Link to comment
22 hours ago, jabbr said:

Err ... sorry I want running water to sound like running water, and twinkles to sound like twinkles.

 

“‘But I don’t want to go among mad people,’ Alice remarked. ‘Oh, you can’t help that,’ said the Cat. ‘We’re all mad here. I’m mad. You’re mad.’ ‘How do you know that I’m mad?’ said Alice. ‘You must be,’ said the Cat, ‘or you wouldn’t have come here.’” - Alice’s Adventures in Wonderland by Lewis Carroll

.

mQa is dead!

Link to comment
2 hours ago, lucretius said:

 

“‘But I don’t want to go among mad people,’ Alice remarked. ‘Oh, you can’t help that,’ said the Cat. ‘We’re all mad here. I’m mad. You’re mad.’ ‘How do you know that I’m mad?’ said Alice. ‘You must be,’ said the Cat, ‘or you wouldn’t have come here.’” - Alice’s Adventures in Wonderland by Lewis Carroll

 

.

 

About the tinkles -- I did not address them.   Consider the decoded version is 3 seconds out of sync with the non-FA LP version of the material.   It *does* appear that the LP is mildly compressed, probably for the reason that @jabbr is concerned (to possibly increase the audible detail at low levels).   However, you should find that the tinkles appear audible close in time sync (modulo the 3 second difference) on the decoded vs LP version.  When doing the comparison, I am definitely considering the vinyl noise can be obscuring at times.   The LP version is more similar to the decoded version than the likely FA version.  (It IS FA, or upon decoding a clean copy of non-FA, the typical dead zone distortion would likely occur.)

 

Compared to the past, the decoder works AMAZINGLY well -- FOR WHAT IT IS SUPPOSED TO DO.  Take a listen to the 'L' demos.   Pretty darned interesting?  Perhaps the highs are still a little hot -- I still don't know if a certain 18kHz, -3dB EQ is needed on the output.   This is likely the primary issue with the highs nowadays.   Also, the LF might be 3dB @ 250Hz too weak, my hearing has trouble making that DISCRETE choice.   There is NO tweaking, just choices. All of the other EQ is nailed down. with engineering decisions made about the EQ.  The previous shrill results came from NOT following the reverse engineering because Instead my hearing told me something very different, and assumed that there was an engineering mistake.   The current decoder needs very few EQ modifications, and only for few recordings.  I should have had more confidence as an engineer.

 

I don't necessarily disagree with ANYONE that the FA version might be more interesting to listen to, but my project is NOT about always choosing the most 'interesting' recording, but instead reverting the recording, as closely as I can figure out, back to the original, pre-FA state.  I have found SOMETIMES (not very often) that I prefer the FA version of a recording.

Both versions in various recordings can co-exist.  There is NO WAY that there is any evangelizing, or trying to replace anyone's recordings,  however I am fairly proud of the project, considering the TOTAL LACK OF SPECS, and the sometimes random or intuitive, and/or reverse engineering walk.  Compared to the past, after a LOT of learning, and a LOT of mistakes, the decoder REALLY works well.

 

Remember - THIS IS START FROM SCRATCH, and nothing about copying someone elses schematics, then tweaking some values, finally calling the results 'genius.'   Most stuff is 'copying schematics' nowadays.  Take a look at the REAL electronics innovators -- look at the research reports and textbooks from the 1950s/1960s. They also had to start from scratch, learning how to apply transistors

 

If my enthusiasm seems like evangelizing, then it might be better to get an idea of what kind of person that I am.  It is wrong to take away someones preferred version, because as I have written before -- THIS IS ABOUT WHAT YOU ENJOY.   The decoder is yet another option for those of us who just haven't adapted to the processing done on consumer recordings.  Some people have adapted, and I am very sure that a brain can unscramble the signal.   My brain doesn't.

 

I am so happy that recently, some people have been kind enough to supply consumer non-FA recordings to me.  I have actually used them to make sure that the direction of the decoder is correct.  With the recent BASS fix, the sound should be very close to correct.  The bass has been the most vexing problem on the decoder -- there are very few tells, just depending on 'taste' till recently.   My hearing doesn't work well, and apparently I have poor taste.  There has been no accurate basis of comparison until very recently.

 

-----

About 'crazy':

 

The world seems generally crazy -- people making crazy decisions all over the world for crazy reasons.  I  have pretty much given up trying to figure out any rhyme or reason how people think.

 

Link to comment
7 hours ago, John Dyson said:

I don't necessarily disagree with ANYONE that the FA version might be more interesting to listen to, but my project is NOT about always choosing the most 'interesting' recording, but instead reverting the recording, as closely as I can figure out, back to the original, pre-FA state.  I have found SOMETIMES (not very often) that I prefer the FA version of a recording.

This is nothing about preference. I want low level details in the recording to be in the decode. It is not acceptable to me to remove information. The decodes remove information.

 

Lets put it this way, from my recordings I can decode using da-avx if I want to remove high frequency signal, but with the decode I cannot restore the stripped out HF detail. The decode removes information from the recording.

 

Take the intros and turn up the volume to the point where you can hear what I am talking about. The LP with high levels of surface noise is not a reference. Can you hear what I am describing when you listen? If your HF hearing is missing then I can see why you don't seem to be understanding what I am saying. This most certainly isn't a compression issue, just turn up the volume on the intro and listen.

Custom room treatments for headphone users.

Link to comment
4 hours ago, jabbr said:

This is nothing about preference. I want low level details in the recording to be in the decode. It is not acceptable to me to remove information. The decodes remove information.

 

Lets put it this way, from my recordings I can decode using da-avx if I want to remove high frequency signal, but with the decode I cannot restore the stripped out HF detail. The decode removes information from the recording.

 

Take the intros and turn up the volume to the point where you can hear what I am talking about. The LP with high levels of surface noise is not a reference. Can you hear what I am describing when you listen? If your HF hearing is missing then I can see why you don't seem to be understanding what I am saying. This most certainly isn't a compression issue, just turn up the volume on the intro and listen.

When decoding with FP in and out, there are almost no details lost, the details ARE in the decoded signal.  The details that you talk about are moved to a level close to the original.  Hopefully we are talking about prefering dynamics compression instead of the dynamics distortion that goes along with FA encoding.   When FA decoding, the decoded  sounds more like the original version.   Along with the FA dynamics compression is also an ugly form of dynamics scrambling.

 

I will repeat -- the decoder is about finding the original version, not producing a version that someone might prefer.   The decoder is totally blind to preference.  It has not feelings, it just 'decodes'.

 

You apparently prefer the 'boosted' details from the FA encoding, and that is okay, but FA encoding  *SCRAMBLES* the dynamics of the signal also.

 

For the best quality do the following procedure:

Scrambled, compressed FA signal -> FA decode -> Compress with high quality compressor -> Clean,  nonscrambled, compressed signal.

The good quality compressor keeps all of the dynamics in the 'correct place' rather than jumbling the levels all over the place.

 

The 'tinkles' are totally in the decoded signal, but brought about 10-15dB down because that is similar to the original.  With 24bit or professional FP encoding, there is little or no REAL audio precision lost.  (I use the professional FP encoding,and in fact double precision FP during the critical calculations.)

 

The decoder is for reproducing the original as close as possiblee from the compressed and scrambled FA signal.  The scrambling is what I don't like,becauses FA SCRAMBLING IS DYNAMICS DISTORTION!!  The compression itself is less irritating, but with FA encoding you get the scrambling also.

 

 

Link to comment

There is a V3.5.1L in the usual places.  Most of the normal demos are available also.

I don't know if I announced it or not as I am having a 'real good time' hearing nuanced details for the very first time.  (I mean REAL details, in the correct place in the dynamics)   The previous versions 'worked' for me, but the full details are now available with the --f9 mode working 100%.  Also, the high quality modes are practical to use.  (on my 10 core, I can real-time decode recordings at --xpp=max.)   On recent 4 core machines, --xpp might be possible in real time.  Just randomly going through my recordings, being very curious.  Some of the ONJ recordings have noticeably more nuanced detail.

 

Most people won't notice a bug that I just unfortunately found -- I made a mistake in a series of EQ, where the anti-distortion mechanism was calculated in a flawed way.  The current version in 'L' works, but suboptimally.   It is far far beyond the quality without the anti disotrotion at all, but just a minor screw up. (The current anti-distortion calculation removes the odd order distortion products, but leaves the even order in the signal.)  The percent of the distortion that it leaves in the signal is small, but possible to hear.

 

If you hear the improvement coming in 'M', let me know.  I can hear the improvement, and wonder if anyone else can also.  I'll leave the M and L demo versions available to compare.

 

'M' will be coming after a more complete sweep of minor bugs like mentioned above.   Maybe a few days, maybe even early in the weekend.

 

 

Link to comment
12 hours ago, John Dyson said:

For the best quality do the following procedure:

Scrambled, compressed FA signal -> FA decode -> Compress with high quality compressor -> Clean,  nonscrambled, compressed signal.

The good quality compressor keeps all of the dynamics in the 'correct place' rather than jumbling the levels all over the place.

 

The 'tinkles' are totally in the decoded signal, but brought about 10-15dB down because that is similar to the original.  With 24bit or professional FP encoding, there is little or no REAL audio precision lost.  (I use the professional FP encoding,and in fact double precision FP during the critical calculations.)

 

The decoder is for reproducing the original as close as possiblee from the compressed and scrambled FA signal.  The scrambling is what I don't like,becauses FA SCRAMBLING IS DYNAMICS DISTORTION!!  The compression itself is less irritating, but with FA encoding you get the scrambling also.

 

 

To be clear, I am looking at the intro in isolation which is very quiet. floating point then normalized. played at 0 dB.

 

The specific chain I am using: for this testing: @Miska HQPlayer upsampled to DSD256 using sinc-M/ASDM7EC modulator to Proj-ect pre box digital S2  DAC to cavalli tube headphone amp to either AKG712 or Senn HD800 headphones. I can also use @PeterSt XXHE/NOS1a DAC with a PCM/R2R chain but trust me same results regarding the FeralA or not question.

 

On the SACD version at 0 dB, the piano comes in first, followed by the tinkles at 7 seconds which move to the right ear and at 17 seconds equally loud  buzzing fly in the right ear. At 24 seconds the volume increases and by 33 seconds when the drum intro starts I turn the volume down to -14 dB to listen. This is not about compression because these sounds are similar volume and the buzzing fly is not destroyed the same way as the tinkles (though you might not be able to tell what it is).

 

It is not simply the volume of the tinkles, rather the timbre, the harmonic structure, similarly the harmonic structure of the flowing water is destroyed by the decode. This isn't a compression issue. You say the same thing over and over yet you aren't addressing the issue.

 

Can you hear it?  What about Close To The Edge intro flowing water? The original is well known as the group is still touring and performing live. I selected both of these recordings because the originals were made in the pre-CD era.

 

The decoder is not reproducing the original, not even close.

Custom room treatments for headphone users.

Link to comment
42 minutes ago, jabbr said:

 

To be clear, I am looking at the intro in isolation which is very quiet. floating point then normalized. played at 0 dB.

 

The specific chain I am using: for this testing: @Miska HQPlayer upsampled to DSD256 using sinc-M/ASDM7EC modulator to Proj-ect pre box digital S2  DAC to cavalli tube headphone amp to either AKG712 or Senn HD800 headphones. I can also use @PeterSt XXHE/NOS1a DAC with a PCM/R2R chain but trust me same results regarding the FeralA or not question.

 

On the SACD version at 0 dB, the piano comes in first, followed by the tinkles at 7 seconds which move to the right ear and at 17 seconds equally loud  buzzing fly in the right ear. At 24 seconds the volume increases and by 33 seconds when the drum intro starts I turn the volume down to -14 dB to listen. This is not about compression because these sounds are similar volume and the buzzing fly is not destroyed the same way as the tinkles (though you might not be able to tell what it is).

 

It is not simply the volume of the tinkles, rather the timbre, the harmonic structure, similarly the harmonic structure of the flowing water is destroyed by the decode. This isn't a compression issue. You say the same thing over and over yet you aren't addressing the issue.

 

Can you hear it?  What about Close To The Edge intro flowing water? The original is well known as the group is still touring and performing live. I selected both of these recordings because the originals were made in the pre-CD era.

 

The decoder is not reproducing the original, not even close.

You aren't listening to the original.   The vinyl is from the original.   The decoder comes close to the original, and is getting closer.

The first part of the tinkles are very low level on the orignal.  Unless you are less than 20yrs old, probably cannot hear them ON THE ORIGINAL.

ON THE ORIGINAL, the tinkles fade up from VERY LOW LEVEL.   The version that you are playing was FA processed from the original.

The data is all on the original, including tinkles,  whatever low level hiss and noise.   ALL of the data is on the decoded version.

 

Do I have to do a gain-up versionof the beginning for you?   Does your software allow for the gain-up?

Do you know what dynamics compressors do?   The example of decoding is the OPPOSITE, the decoder is partially an expander along with descrambler.

 

If you like the currently distributed version, that is up to you.

You are speaking of preference -- I don't go there.

 

Trust me, all of the information is in the decoded version, and is getting closer and closer to the original LP version, sans vinyl effects.

 

 

 

 

Link to comment

Here is an example that the tinkles are not gone on the decoded version.   Remember this:

 

* Anything that the decoder does, the FA encoder can undo.  Vice versa is also true

* The decoder is dynamic, and I have no FA encoder, so my ability to replicate the behavior is weak and flawed.

* The decoder/encoder is at least 8th order to maybe 16th order EQ filters, depending on what is going on.  I just did 4th order in the

region around the tinkles.  

* Decoding was NEVER meant to be -- it is amazing that it almost produces the original vinyl (sans vinyl effects), but admittedly I only decoded to 5 layers,

and that is why the vinyl APPEARS to have even lower levels for the tinkles. (hard to tell though -- the early tinkles are at/below the vinyl noise, as ORIGNALLY

INTENDED TO BE LOW LEVEL.

 

As far as I can tell, on the decoded version, with EQ just as the decoder has *dynamic* and more effective EQ of higher order,

the tinkles clearly start at 13 seconds.   They get louder at about 15seconds and very easy to hear at 20 seconds. (The tinkles might

start earlier than 13 seconds, but  poor hf hearing probably does interfere with my estimate.)

Also, my decoder is very purposefully optimized to gouge out hiss, so there might be some instantaneous decrease in HF levels at very low levels.

On the zeroed-in decoded version, the piano does interfere with the tinkles at 15-16 seconds, but the FA process can remove that confusion by compression.

Most important -- the FA process wasn't meant to encode AND decode, so it should be amazing that the decoder works at all, restoring

something to nearly the original.

 

Anything that you hear, and output by the FA decoder can be brought up to listenable with the FA encoder, just like the version

that is normally distributed.

 

Both the original LP version, with hiss and vinyl effects AND the decoded version are pointed to below.   It is really, really hard to hear the

tinkles under the vinyl noise, therefore unlikely to be hearable at all in the  original state.   The decoded version has the advantage of being a clean copy, properly decoded.

I really, really tried to cleanup the LP version of the tinkles to be listenable, but it WILL hurt your ears when listening, so listen to the decoded one first.

The timiing is approx synchronized -- I can barely hear the tinkles on the LP version at about 16 seconds, while the tinkles are barely audible

on the decoded version at 13 seconds.   This 3 seconds are not representative of the timing difference because the piano starts at the same time.  The

3 seconds is a coincidence So, there was a 3 second difference between the decoded and LP version, and I tried to mostly compensate

for that, so the timing comparisons are similar.

 

An FA encoder can compress the tinkles up to the level of the normally distributed version today --

ANYTHING that the FA encoder does, an FA decoder will undo and vice versa.

 

(These are 16bit files.  Enhanced to zero-in on the tinkles.  Didn't use 8 to 16 order *dynamic* EQ like the encoders/decoders do.)

DECODED:

https://www.dropbox.com/s/dgjnmh2jl9j0nv7/DECLambTinkles.flac?dl=0

ORIG LP:

https://www.dropbox.com/s/ne1562fz82biiey/DECLambTinklesLP.flac?dl=0

 

Link to comment

Here, I want to explain something about the decoder:

 

BEFORE READING THE BELOW, MOST IMPORTANTLY:  I don't care about personal preferences, and much prefer that people who don't like the output of the decoder simply enjoy the music.

 

I beg to find more pure,non-FA materials, and greatly treasure when such are found.   The fact is -- I have just a few examples, and greatly appreciate the kindness of those people who have really helped the project.  If something is FA, I simply cannot use it.  I already have 1000's of FA materials.  I WANT/NEED more non-FA materials, so if I establish that something is FA, it most likely is FA.  I really want non-FA.  The project is benefitted by pure, non-FA materials


The decoder, even when perfect, might might not create what some people always want.   The decoder isn't about preferences at all.  The decoder simply undoes a process that is done to many recordings (seems like all recent recordings, but probably not.) The decoder has not been as perfect as I hope, but as time goes on, I find the bugs.  The decoder is BLIND to personal  preference.   It does SO much processing that even a slight mistake totally screws up the sound.

 

Sometimes bugs might be very different than they seem. (for an idea of such complexity, imagine navigating in 5 dimension space -- the numbers don't work the same as 3 dimension.   This is the kind of complexity of the decoder.)

 

It might be unbelievable about the decoder being  so very complex, but it REALLY ISAlso, for referring to back versions, there are 1356 back versions available immediately, most can be run, and can show the directory list if really needed to believe it  (still not believing it, I can show the file list -- but expect a huge file.)  Previous versions are checked regularly.

 

Freq response balance seems to be most obvious defects that people notice,  and those mistakes should be simple to fix FOR SOMEONE WHO CAN HEAR.   I found that I am unable because of really wierd and variable hearing.  However, the dynamics processing is the hard part for those who can hear or not hear well.  Adjusting the freq response is TRIVIAL for those who can hear.

 

The dynamics processing has been deemed impossible by 'experts', but it has been done.   Even just one portion of the dynamics processing was claimed to be impossible, but it has been done.  Speaking of 'tweaky' -- the intimate semiconductor emulation in the internal DA decoder IS tweaky, but after a lot of patience, works really well.

 

When given the engineering choices, the decoder 'locks in' to correctness, because there are already engineering decisions that limit the choices for correctness.   This is not about 'tweaks', but instead 'choices.'  I apparently have hearing so bad that I cannot make the choices.

 

AGAIN: THERE IS NOTHING WRONG WITH ENJOYING UNDECODED/FA MATERIAL!!!

ALSO:  THERE IS NOTHING WRONG WITH THE GOALS OF THE DECODER, EXCEPT MAYBE TO VEX SOME IP HOLDERS.

 

 

 

 

Link to comment

Doing a major cleanup of the decoder.

I DID find that they had an additional layer of pre-emphasis that couldn't be predicted, and I couldn't hear.  The design that had previously been released had the technically minimal amount of EQ, but couldn't hear or guess at anything additional.   Here are the notes about the changes (so far), so have been VERY active at cleanups and quality improvements.  Running the demos right now, but because of previous commitments, cannot do a release, and need to do some final checks for subtle distortions.  10 layer decoding REALLY pushes the accuracy of the DA decoder, and might never give the performance that I hope/expect.

 

Improvements to V3.5.1R(S or T?), hopefully the last of the V3.5 series, moving to V3.6

 

  • Corrected one case of incorrect distortion cancellation offset.

  • Corrected distortion cancellation in 2nd order EQ. Previous 2nd order distortion cancellation removed only odd order components. The new, corrected version cancels both even and odd order components. (This is actually based on a reverse engineering design artifact and not an innovative cancellation technique.)

     

  • Apparently the designer unexpectedly used a layer of pre-emphasis before encoding (of course, I couldn’t hear it.) Sadly, it sounded better *to me* to not have the EQ. Like most of the other layers is a 18kHz/-6dB & 9kHz, -3dB EQ. This also says that the 18kHz, -3dB wasn’t needed to compensate for the 24kHz portion of the decoding pre-emphasis. I hope that this single layer (which could still comprise another step in the layer) fully undoes the pre-emphasis. The decoder had previously used the minimal, engineering required, amount of EQ.

     

  • Carefully adjusted the rolloff EQ, to help correct the sibilance and cleaner highs.

Added 24kHz, 1st order component to the rolloff EQ. (this is NOT related to the decoding or encoding pre/de emphasis, but only for unintentional recording rolloff.)

 

  • Found a timing offset error in the DA decoder module, vastly improving the stereo imaging, also mitigating some harshness in the sound at the largest numbers of layers (mostly 10 layers), because the gain control wasn’t quite centered on the waveform. 10 layers really pushes the accuracy of the decoder.

     

  • Slow down the DA dynamics a slight amount. This produces weaker sidebands, further cleaning up the sound, and making –f10 recordings sound better.

     

  • Decided to fully support the –xppp mode. For stereo imaging and even LF distortion, a properly configured –xppp mode is sometimes better than the –xpp=max, and uses approx the same CPU. The --xppp mode also upgraded to reach to lower frequencies, canceling more low-side modulation products. The low-side and low frequency modulation products are much uglier sounding than the higher frequency modulation products, so it is a double win.

     

  • Moved some of the LF building blocks around, giving a more flat response below 30-40Hz.

 

Modifications checked for, but not needed

 

  • Gain shift between layers. If not correct, the calibration level becomes sloppier. Strange ‘hitches’ can appear in the sound.

  • Double checked the conceptual engineering correctness in the post decoding EQ.

  • Double checked again the LF pre/de-emphasis that was corrected in V3.5.1L.

 

Link to comment

IMPORTANT:

The vexing post decoding EQ started being insane when finding that adding the additional layer was necessary.

There is now a post-decoding EQ that requires NO tweaking, and is NOT dependent on  anyones hearing. It requires a modification of inter-layer EQ.

 

* BETTER EXPANSION at --fa

* NO NEED FOR MODES OTHER THAN --fa

* CALIBRATION OFFSETS DO NOT CHANGE

 

I kept getting an edgy sound at --f9 on some classical material.  At least the DA decoding module got a very complete re-consideration for distortion.

 

Link to comment

With my bad hearing, I have two choices (just two) which could be correct.

 

There will be two versions: V3.5.1S and V3.5.1T.   ONE of these is correct.   There might be an in-between, but I doubt it.

I won't be able to do the updates until tomorrow night, and there will be two complete versions of the decoder with a two line change between them.

 

This version of the decoder has several important changes.   I was hit by my broadsides with my loss of hearing, and a little more patience will produce something very nice.

The ONLY deficit is in the HF, and the error in the handling of HF percolated down into other sections.

 

Note that the dynamics are VASTLY improved, and my A/B comparisons show  a lot of corrected issues, because of the corrected inter-layer EQ.   THE HF between 9kHz and 20kHz is the only real problem left.  (It is actually between 6kHz and 18kHz, but I think in terms of 9kHz as a turnover point.)

 

I'll be very grateful with the feedback on which one sounds correct:  V3.5.1S and V3.5.1T

 

 

 

Link to comment

Did some verification using raw, pro equipment -- no processing.

 

Was doing some tests with a TLM103, and comparing with vocals on the decoder -- sound very similar, except sibilence is usually handled better on a recording than a raw mic.  If you have any 's' problems (like I do), and get close to a pro-mic, it will give you a very strong 's' sound (along with a cardiod's bass boost.)   Be kind to the decoder when you might find too much HF in these next demos, esp on the Anne Murray examples.   Some recordings are 'just bright'.  The background is very bright, yet if you play Brubeck or something like Clapton, the sound is more mellow (not mellow on  Clapton, just more mellow.)  Material like Bread or Dionne Warwick is down-right subdued.  I really wish they would have recorded Warwicks vocals more bright - again, the decoded Anne Murray recordings might burn your ears.

 

In the demos, we are listening to non-mastered materials.   Expect a very wide variation.  FA encoding is like 'mastering on the cheap.'   Sometimes FA recordings ARE mastered, but the full quality is not available until decoding.  FA encoding homogenizes recordings so that bright and dull recordings have less difference...

 

The new decoder IS more subdued, but still might be bright on some recordings.   Thats what happens on material that isn't truly mastered.

 

So, expect the decoded sound  to sound more like a REAL mic, and not heavily processed mush.  A telephone is not a good example, it really takes a professional mic or near equivalent to compare with on the decoder.   Also, sometimes speaker systems are tweaked for the normal, commonly available recordings.  The decoder output does NOT produce that.

 

I suggest comparing using headphones if really listening to the decoder.   It seems MUCH less bright than consumer recordings, but as I said, sounds very similar to a real mic (modulo proper studio processing -- not the FA garbage.)

 

 

Link to comment
5 hours ago, John Dyson said:

Success!!!

Found that I DO have a copy of non-FA (pure) ABBA!!!

The decoder matches the dynamics 99%!!!

 

Still have some cleanup before the  release.

 

About the ABBA update -- I found out why the Anne Murray demos sounded so intense.   This also fixed some minor problems on the ABBA stuff.  (the match is even closer.)

 

The bug had to do with the HF pre-emphasis/de-emphasis.   The orginal version of the decoder  was using 3kHz/6dB in, 3kHz/-6dB out.   The sound matched when using 3kHz/3dB in and 3kHz/-3dB out.

 

Even though the dynamics were originally very close, the highs on some ABBA selections had the 'Anne Murray' problem to some extent, and the error was too severe to pass it off as 'old LP' vs 'new Digital'.  This pre-emphasis/de-emphasis modifies how intense the highs are in certain bands and  are all about driving the DolbyA units optimally.   I can make engineering estimates and do pretty well, but there will always be errors until the tests/verifications  can really be done.

 

Having the before and after copies (the 'Lamb' copies earlier were also helpful) makes things a lot easier.  I have some other non-FA also, but they were too noisy/too much tick and pop to really decide on similarity/differences.

 

The conversion back to the original will be much closer than before, but I totally promise you'all -- the results will not always be more enjoyable to you (everyone.)

 

As I wrote someone earlier, I really wanted to make a release today, but would be irresponsible at this time.   There are too many things to do, including do comparisons with the 'pure' recordings, making sure that the 'BASIC' EQ is correct.   There will always be issues with the HF until I get some help making simple, discrete choices -- perhaps two or three.  One of the two or three test choices will be correct, or with maybe one very minor nit.

 

One thing that I am somewhat happy about -- during one test, I did an 'extra' EQ of +3dB at 24kHz, and could clearly hear the difference.   Admittedly, a 1st order EQ at 24kHz can easily  have noticeable effect at 12kHz, but I am very sure that I was hearing >>12kHz.   This is telling me that my HF hearing might be attenuated, but isn't gone.

 

 

 

 

 

 

Link to comment

Status report on the pure (non-FA) recordings:

 

  Some of the old vinyl rips have been EQed to hell and back, however there has been some very useful info obtained from the non-FA recordings.   A few are somewhat useful for getting an idea what the EQ should be, but the most important thing to prove are the HF dynamics.  (there is the 6kHz region, the 9kHz region and the 12kHz and above.)   All of the non-FA, even if EQed to death, are helpful on the dynamics.   The dynamics are definitely tricky, because the basic dynamics are mind-numbingly complex, but also each frequency region can be somewhat different.   The 'Anne Murray' problem was with the 3kHz/6kHz regions.

 

Once the dynamics are fully PROVEN*, and I am fully comfortable, then it will be time to do our EQ choice, and I'll need help making the 2 or 3 way choice .  I am close to being fully comfortable with the dynamics, but going to search further in my repo for perhaps more non-FA stuff.   This 'find' in my repository has been a 'godsend', and am so happy to say thank you to whoever sent them to me,  probably over 10yrs ago.  (I already had several versions of each album in my catalog at the time, so I didn't feel guilty.)

 

* Not totally ready to say the dynamics are perfect, but they are indeed darned close to 100% correct relative to the recordings that have.

* This new work will not see the light of day until probably Wednesday.

 

Other VERY useful info:  most of the decoding tests are showing that --fa=4 and --fa=+5 are the most correct (4 or 5 layers) instead of 7 or 8.

 

Something REALLY interesting though...   The decoder can properly decode 'Waterloo' to near the 'pure'  example at 4 or 5 layers, however when using a higher '--fa' 7 or 8 layers, the chorus cleans up and actually sounds more clean.   This is the kind of thing that encouraged me to support more layers.

 

At --fa=+5, the decodes will go almost twice as fast as --fa does now.  (--fa is currently the same as --fa=+8.) 

 

I still like the sound better at --fa or --fa=+7, but my taste might be wrong (like usual.)

 

Link to comment

Planned release for today is delayed a little.   The private test release had also been delayed until this evening, which means that the publically available release won't be available until at LEAST +2 days.   Reliable tests and verification can sometimes take a very long time.

 

I found one really major botch (my mistake), but no-one had complained about its effects.   All in all, correcting the botch will remove a strong sense of LF distortion. A simple EQ problem in a sensitive spot can play terrible havoc.  The frustrating thing is that it was a forgotten experiment (80Hz instead of a correct 75Hz EQ.)

 

Been worried about comments regarding the upper LF/midrange.  I have seriously addressed the matter, and one reason for the delay.   The tests are very tricky to do, but using the FA original  is USELESS for determining correctness .  The sound is JUST different.  I really tried to compare the FA copy and nonFA originals, and it just does not compare well.  There is little equivalence.

 

The other frustrating thing about the comparisons is the strong/narrow 50Hz/80Hz bass boost on FA materials.   I have done experiments on properly reversed (decoded) materials along with never-FA materials, and the result is that the decoded materials can NOT sound like the FA materials.  Without the FA filtering (which is effectively narrow band), any attempt to boost a 80Hz signal creates a wide enough boost to overwhelm with too much true bass.  * A few weeks ago, I did an experiment to undo the FA filtering without decoding the materials -- it sounds decoded, but VERY compressed.  There *IS* something genius about the FA compression/encoding scheme -- but it is JUST WRONG.

 

 

Link to comment

Got it...   Getting stereo like I remember back before the 1980s.

Crazy a** matrix for the EQ from MF down to LF...   (I call it the EQ adapter.)

 

I haven't heard stereo like this in decades (unless Quad decoded.)   NO FANCY QUAD or anything like that...

The decoded version does have better stereo, but this is closer to what I truly remember.

 

Gonna try to do a release soon -- gonna bypass the private reviewers on this...  I wouldn't normally do that, but this is AMAZING.

 

Listen at your own risk...   This is a Carpenters piece that I am amazed at.  The first time that the decoder isn't just putting out a flat image.

 

CarpDemo.flac

Link to comment

This previous demo only has about 1/2 of the stereo image corrected.   Also, a HUGE amount of EQ simplification (hidden internal stuff) has already been done.

 

This is all about the *internal matrix* that I didn't even realize existed.

This matrix does a lot of EQ that built-into the pseudo-stereo from the FA encoding.  (it cancels and adds, kind of doing automatically what I was doing by brute force.)

 

The liklihoold of a release tonight isn't quite 100% because of the need to 'rest my hearing' in between tests.


REMEMBER -- the Carpenters demo has only about 1/3 to 1/2 of the fix!!!

I was surprised again when I took a look at details the 'adapter' between LF and MF which made me realize what was going on.

 

(This reverse engineering game is a real trip -- more and more like Tetris with some whack-a-mole, rabbit chasing and Pac-man all in one :-).

Phew!!!

 

 

Link to comment

This is really crazy stuff.   Whoever designed this thing did an amazing job of obscuring how the signal was encoded.

It almost seems like someone had designed an analog encryption scheme that makes the signal impossible to recover without herculean efforts.   The odd thing is that a DolbyA unit cannot be an element of the decoder very well because of all of the distortion components splattering all over the place.

 

The decoder is NOT ready for a full demo and release, or I'd attach an improved demo.  At this point, there is a bit of instability in the EQ, where the low levels and high levels have a slightly different response curve.   A big part of making the decoder work correctly is to match the crazy response curve, which is NOT fixed like an amplifier or even RIAA preamp.

 

For a single demo purpose only, there COULD be pretty sounding recording produced, but it would be wrong to do so, because I'd estimate about 3dB at 10kHz slide of frequency response depending on signal level.   The defect is very noticeable, and fixing it requires better understanding of how the response vs. gain curves function.

 

In the Carpenters demo, the only purpose was to show that there was a real improvement in how the stereo image sounds.  It is even better now.

Rather than rush things and make a mistake because of perception problems on my part, I will NOT likely do a release tonight, but the upcoming one will be more than worthwhile.  The results are VERY pretty, and show strong, clean dynamics.

 

 

 

 

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...