About the FeralA decoder -- mucho progress.

Topic: About the FeralA decoder -- mucho progress. (Read 4120 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

About the FeralA decoder -- mucho progress.

2020-12-16 21:42:32

Enough progress has been made to claim that the quality improvements are approaching master tape quality instead of just an improvement.*
* For those who don't remember -- this processor takes most consumer digital distributions and undoes the common type of compression used. It is NOT an 'expander' in a conventional sense.

The reason why I am including this in the 'scientific discussion' forum is that the program IS a 'science project' and is NOT a normal expander type thing, as it uses techniques MUCH MUCH more advanced than simple gain * signal stuff.   Consumer recordings are damaged so severely that a normal expander design (based on DolbyA techniques) could not produce the super clean results.   Geesh, a DolbyA can just barely produce clean results when decoding it's native format. (DolbyA decoding has some design bugs -- not obvious, but they are there.)   In months, not years, the source code is planned to be released (but don't expect to be able to just sit down and read it -- it is a very complex piece of massive SIMD DSP code.)

The original decoder software goal was 'master tape', but up until very recently all I could do is make improvements, or make improvements with tradeoffs.   The latest corrections, after getting beat up by a bunch of audiophiles -- I think that it is REALLY CLOSE to success. *The decoder IS NOT a 'DolbyA' compatible decoder when used as an FA decoder, but is a complex mix of technology that uses DolbyA type processing as a basis.

Other than sound quality, the biggest improvement is that the decoder is ALMOST easy to use. It is perhaps 3X more complicated to use than to do a simple DolbyA decode with the software, and the processing speed is normally only 2X realtime on a 4 core CPU. (Thank Goodness I just got a 10 core 'X' processor.)

Here are some snippets -- sorry that they must be snippets.   The software release should have been 15Dec, but will be 17Dec because of some complications on certain recordings...   THESE DO NOT SOUND LIKE CONSUMER RECORDINGS -- instead they sound similar to what comes out of a mixing board.   Don't listen for the grainy compressed sound, or the swishy highs -- they aren't there. The sound character is PROFOUNDLY different than most consumer recordings.    I will announce the software Thu or Fri at the latest.   There a normal distribution point for the regular software users, and will publish it here tomorrow.
*DO NOT DEPEND ON THE DROPBOX PLAYER -- it sucks. Downloads are much much better.

https://www.dropbox.com/sh/mjmdfxu8gdweoc2/AACE7AQA1kZ0AIFNxar_sZoJa?dl=0

John

Re: About the FeralA decoder -- mucho progress.

Reply #1 – 2020-12-17 01:03:43

PS: I need to clarify -- these examples will sound VERY different from what is typically expected.   NORMAL CONSUMER recordings have LOTS of HF compression -- creating a 'sparkle' in the sound that is almost like it is sprinkled.   Likewise the highs are modulated by other middle highs on most consumer recordings.

On my examples, you'll notice a 'suspicous' quiet between syllables, and might not sound correct on a system that was optimized for the normal consumer sound.   I suggest using an ACCURATE set of headphones, trying to unlearn the sound of the consumer stuff.

Think about a microphone and a voice -- just listen to it.   My decoding results might be a little bright, but the sound is MUCH MUCH MUCH more coherent and consistent with a less compressed signal.   The decoder results don't have the random 'gravel' as in the consumer stuff.   If the consumer material doesn't have 'gravel' in the sound, there is often too much HF compression, loosing the sound of things like REAL WORLD cymbals.

I am not claiming that the results are perfect for what it is supposed to do, but it is NEARLY perfect.   I recently had someone be so proud of a recording that they loved -- it was a consumer copy.   Lo and behold -- I had an image of a master tape with the same material (with DolbyA tones and all) -- the consumer recording had the garbled, gravel, compressed sound that I speak about and the FA decoder mostly undoes that mess residing in most consumer material.   The image had NOTHING like the cruddy consumer recording type of sound.

John

Re: About the FeralA decoder -- mucho progress.

Reply #2 – 2020-12-17 01:12:48

One major correction to some of the examples -- I cannot hear frequency response errors very well. At least some of the recordings needed a 9kHz -> 21kHz shelf. I am listening for distortion, and REALLY CANNOT hear frequency balance problems.

The decoder sounds the way that it does based on design -- I just forgot a step. The correction is a single character in the command line.

I'll create another version of the examples. They should sound even MORE distortion free.

John

Re: About the FeralA decoder -- mucho progress.

Reply #3 – 2020-12-17 08:05:17

I listened, but couldn't hear improvements, because there are no references. I'll wait for the software to be released, will apply to some of my music and do ABX test, as usual.

Re: About the FeralA decoder -- mucho progress.

Reply #4 – 2020-12-17 08:32:32

It is nice to read that the project has continued since we spoke the last time. How did the discussion/presentation at AES go?

Re: About the FeralA decoder -- mucho progress.

Reply #5 – 2020-12-17 10:56:20

Quote from: itisljar on 2020-12-17 08:05:17

I listened, but couldn't hear improvements, because there are no references. I'll wait for the software to be released, will apply to some of my music and do ABX test, as usual.

I am all for more A/B testing.   I am fully occupied doing the decoder itself, without ANY spare minutes.   I have done my own experiments and made sure that there isn't any error in them, but of course, doing the measurements in more controlled with statistical confidence is the right thing.

Unlike in a real company, I have no staff to do the work -- the code itself is a major effort and is of significant complexity.   There are algorithms in the code that exist nowhere else, and it has been a challenge to properly implement those.   There is so much layered technology that this effort COULD NOT have been done in the corporate environment without a team of at least 3-5 programmers with strong DSP/EE capabilities.

So, yes -- I have been a little lax in the testing department, but also -- anyone who knows what most normal consumer recordings sound like, the improvement should be obvious. (Sadly, most people have been trained by the horrible quality recordings -- mostly because of the compression that the decoder undoes.)

I am 100% willing to privately supply the recordings along with decoded copies for more complete comparisons.   Believe me -- the 'science project' of simple testing is totally blown away by the 'science project' of the decoder itself. I simply have NO MORE cycles.

There is at least 200-400Hours of work to get the decoder into shape for source release -- there are some real gems in there, including a way of doing compression/expansion with a true reduction in modulation distortion (not like the Orban patents that simply move the distortion to the upper sideband so that it doesn't sound so bad.)

Privately contact me, and I'll provide the basic inforomation to start.

John

Re: About the FeralA decoder -- mucho progress.

Reply #6 – 2020-12-17 11:06:38

Quote from: punkrockdude on 2020-12-17 08:32:32

It is nice to read that the project has continued since we spoke the last time. How did the discussion/presentation at AES go?

I am not sure -- I had to back-off from the official project because of the extreme complexity of the FA development, and there was pressure to cause me to defocus from the ground-breaking FA project that I am talking about today.   I kept getting requests to do other noise reduction systems, but the FA decoder hasn't been starting to be totally accurate until a few weeks ago.

The decoder had to be developed without ANY specification, and all testing for the higher level project (FA) being done by listening. I TRULY DESPISE SUBJECTIVE TESTING!!!   However, there was no other way to do it-- but I have really gotten good at chasing rabbits and playing whack a mole.   (Anyone who has tried to rely on subjective results understands what I am saying.)   I am all for using more scientific methods for comparison, but any comparison needs to consider that most audiophiles hearing has been trained on the nasty, compressed recordings that I am fighting against.   A lot of people don't even hear the compression and extremely dynamic phase distortion of most consumer recordings.   The FA decoder reverses this distortion.

Luckly, the DolbyA compatible decoder is a given, and totally complete.   Since the FA uses DA technology, it has been a life-saver to have the DA portion 100% complete and accurate.   Even the DolbyA compatible mode has been tricky, because the only spec was hardware that relied on selected components.   Since the processing is highly nonlinear and Ray Dolby was a genius designer, some aspects of the DolbyA design are a bit tricky to pick-up on.   There are aspects to the DA portion that make it very clear why people still use the design for enhancers -- it is too easy to misunderstand the schematic -- the schematic is misleading unless read very carefully.

I have gotten interest (and use) from major labels, some doing things that many readers of this forum might be interested in. I cannot publically state who is using the DA decoder, but in some cases -- really elite organizations have copies.

The FA project is separate, but has the same kind of dedication as the DA effort.

John

Re: About the FeralA decoder -- mucho progress.

Reply #7 – 2020-12-17 18:39:02

Quote from: jsdyson on 2020-12-17 10:56:20

Quote from: itisljar on 2020-12-17 08:05:17
I listened, but couldn't hear improvements, because there are no references. I'll wait for the software to be released, will apply to some of my music and do ABX test, as usual.

I am all for more A/B testing. I am fully occupied doing the decoder itself, without ANY spare minutes. I have done my own experiments and made sure that there isn't any error in them, but of course, doing the measurements in more controlled with statistical confidence is the right thing.

[...]

I think that itisljar is just sugesting that including a snippet of the original together with the snippet of the processed file would make more sense. Particularly in this forum.

Re: About the FeralA decoder -- mucho progress.

Reply #8 – 2020-12-17 20:08:33

Quote from: danadam on 2020-12-17 18:39:02

Quote from: jsdyson on 2020-12-17 10:56:20
Quote from: itisljar on 2020-12-17 08:05:17
I listened, but couldn't hear improvements, because there are no references. I'll wait for the software to be released, will apply to some of my music and do ABX test, as usual.

I am all for more A/B testing. I am fully occupied doing the decoder itself, without ANY spare minutes. I have done my own experiments and made sure that there isn't any error in them, but of course, doing the measurements in more controlled with statistical confidence is the right thing.

[...]
I think that itisljar is just sugesting that including a snippet of the original together with the snippet of the processed file would make more sense. Particularly in this forum.

Good idea -- I normally provide both snippets -- gotten lazy lately (it is otherwise called burn-out.)
Just got the release ready, but cannot be released until I put-together a new quick-start for new users. I'll announce by tomorrow.   I produced an AVX512 version (Linux only for this.) this time, but it is actually slower...   I can tweak my 'X' machine so that it is about the same speed as the AVX version - but not worth it.   There are both Linux and Windows versions -- the Linux version works much more nicely because it is naturally a command line program.   Windows works fine though.   (The program loves pipes, but works nicely with specified files also.)   Professional applications most likely would use normal file I/O, but tuning/tweaking for decoding parameters works super smoothly with a SoX I/O scheme.

Some-time tomorrow, I'll also have some A/B material ready.   There were major improvements after the last 'demos', when I got my a** eaten out by some really embarassing botches (both a missing output shelf that I disable for testing and results from my poor jdugement of response balance.)    I could easily scuttle the project by mistake because of my horrible demo skills and atrocious mastering judgement.

I think that the examples/comparisons will be informative, and start to give in INITIAL notion of how to test the monstrosity.   I do have one unachieved important goal -- the decoder is not good at maintaining perfect input/output timing -- it is often off by 10-20samples.   the offset is constant assuming no changes in decoding mode (e.g. multiple tracks will remain in sync), but comparing against undecoded material by simple subtraction isn't very helpful. Originally, out of ignorance, I didn't realize that the file timing was important -- so I retrofitted a scheme, but it probably needs more work to get down to 1 sample accuracy.

Instead of getting perfect timing right away (too difficult given my other current goals) I am going to research a scheme where the delays of UNDECODED material are maintained the same as decoded material -- like a disabled decoding mode.   The propagations would be exactly the same for testing, but wouldn't solve the desire perfect INPUT/OUTPUT file timing. The goal would be to enable uncorrected file-subtraction for decoded/undecoded techncial comparisons.

The timing problem can probably be solved, but the propagation through the program is on the order of seconds.   All I can say: Thank goodness for linear phase filters!!!!   Also, with O(1second) propagation delay, there is no way to use the program in a real-time application :-).

John

Re: About the FeralA decoder -- mucho progress.

Reply #9 – 2020-12-18 08:41:03

What is "master tape quality"? Surely if this compressor "undoes the common type of compression used" then the recordings will be all messed up as compression is applied in an artistic manor in most if not all recordings. EQ and compression are applied in tandem in mastering sessions - you cant remove one without affecting the other.

Re: About the FeralA decoder -- mucho progress.

Reply #10 – 2020-12-18 14:39:55

When i first read about doing a free Dolby A decoder i liked the idea.
It indeed sounded into the right direction on some ABBA songs that were most likely transfered wrongly.
In my small world the scientific method should be to record a suitable recording with a Dolby encoder applied and digitize it.
The Decoder then should be optimized to sound like the source again.
Here it seems this work is done on some recordings that on its way once Dolby A was applied besides other effects no one ever will know exactly.
This approach seems like trying to paint an exact picture of the cow by looking at the hamburger meat it ended in.

Re: About the FeralA decoder -- mucho progress.

Reply #11 – 2020-12-18 16:56:23

Quote from: Wombat on 2020-12-18 14:39:55

When i first read about doing a free Dolby A decoder i liked the idea.
It indeed sounded into the right direction on some ABBA songs that were most likely transfered wrongly.
In my small world the scientific method should be to record a suitable recording with a Dolby encoder applied and digitize it.
The Decoder then should be optimized to sound like the source again.
Here it seems this work is done on some recordings that on its way once Dolby A was applied besides other effects no one ever will know exactly.
This approach seems like trying to paint an exact picture of the cow by looking at the hamburger meat it ended in.

Actually, I do have DolbyA references (true master tape copies, with tones/etc.)   The DHNRDS FA (not DA) decoder is still somewhat experimental, and is intended for consumer applications.

The DHNRDS DA decoder is emulates the original DolbyA HW for decoding, but is better.   The sound is pretty much the same, except transients are more clean, and vocal chorus are more detailed.   True DolbyA units have an intrinsic delay in the feedback loop for decoding, but for encoding it is nearly perfect.
So, the opportunity for the DA mode is to produce more clean results from master tapes.

The DHNRDS can do both DA (master tape stuff), and FA (to undo the compression used on consumer stuff.)   The FA mode is where there might still be errors, but it is getting better.   I recently saw a posting in another group of consumer versions of some material where I do have a master tape copy (actually only a safety tape in that specific case), and the consumer version was compressed, gritty an grainy in comparison with a clean, accurate decode of the DolbyA master tape.

This improvement might not always be true, but a SUPER IMPORTANT improvement in the FA (consumer mode) is that the mode specifiers now build-in all of the needed EQ.   That is, there are only two modes (instead of previously a myriad) and several common submodes each.   The number of mode choices -- everything that is needed - are down to the point where it is almost simple to use, perhaps only a few times more complex than decoding of DolbyA material.

Since the FA mode NEEDS multiple DolbyA units to properly clean up the compression, the FA decoder produces superior results (if the settings are correct) because the DA decoding is so good.   I really doubt that a DolbyA unit is clean enough to decode the material in the way that the FA decoder does.

I am not claiming that the FA decoder is perfect -- but as I get feedback from frustrated users, I try to correct the problems.   There is NO specification for DolbyA or FA -- so it requires careful reverse engineering on DolbyA (I have super high confidence that the DA decoder does all of the tricks in the deceptively simple schemati), but there is no information for FA other than the subjective results.   I could explain the steps needed (lots of whack a mole, and chasing rabbits), but the effort to clean up the compression on consumer recordings started in 2012, and I didn't decide that the processing was DolbyA related until at least 1.5yrs later.

This has been crazy-making, but luckily (in a way), I cannot work at a job because of an odd medical problem - but I can sustain this development effort (it is written in C++, and I use a compiiler that is expert at producing SIMD code (clang++), but also I have some self-written packages that do all of the work to make SIMD simple to use.   Also, I found a REALLY GOOD SIMD transcendental library that is essentially totally free...   I adapted the package through translating routines to fit into my own SIMD infrastructure.)

If a truly C++ savvy person would look at the code, they would see that the it goes through all of the proper processing steps. It is a very complex piece of code -- and is organized so that it acts almost as 'wiring' and 'EQ' between DolbyA processing.

Anyway -- I'll be putting up A/B comparisons soon.   Just ran into a bug where there was drift between different levels of decoding, which should NOT have happend. I make mistakes all of the time and I simply swapped two constants -- one is -9dB (2.818) and 2 * sqrt(2) (2.828).   The use of both constants was intermingled in a very logically consistent way -- but I had a brain fart during transcripton....

I am working hard on this thing, and congent and kind criticism is important to move forward.   I am open to good criticsm, because it helps fix bugs where I simply misunderstand what is needed, an where my hearing-judgement might have been in error.

I will be making the release available in this group soon.   I have some testers somewhere else, and they make sure that I havent done something stupid (again.) It will be available when I am confident of it working as expected.

Thanks!!!
John

Re: About the FeralA decoder -- mucho progress.

Reply #12 – 2020-12-18 17:20:00

Quote from: NateHigs on 2020-12-18 08:41:03

What is "master tape quality"? Surely if this compressor "undoes the common type of compression used" then the recordings will be all messed up as compression is applied in an artistic manor in most if not all recordings. EQ and compression are applied in tandem in mastering sessions - you cant remove one without affecting the other.

From what I can tell -- there is a mechanical step applied to almost every consumer recording, but the criss-crossed compression results in a similar freq response balance.   The major difference is a lower midrange boost, swishy cymbals and a truly distorted bass.   The midrange (other than the lower midrange) generally has very little gain control applied to it.

So, the results provided to consumers is 'kind of close' to the original, but IS NOT the same as the mix.   This goes beyond good mastering.   I quit the HiFi hobby in about 1990 because it didn't seem that CDs would start sounding any better.   Since I used to do recordings myself (stereo, orchestral), it was easy and so obvious that there was a consistent 'distortion' somekind in the consumer recordings.   At the time, I ddn't have the resources or time to correct the problem --f rankly I wasn't interested.

In about 2012, with truly HiFi virgin hearing, I listened to a CD again -- but with all of the experience over the years, I heard some things that I could 'grab on to'.   The damned material was compressed somehow.   After about 2yrs, I found a segmented approx 2:1 compression in the audio.   The speed of the compression appeard close to DolbyA.   After writing an early, approximate DolbyA decoder, I found regions in the audio where it cleaned it up.

Sadly, for about 5yrs, I was trying only one DolbyA decoder, but the good news is that a very nice recording professional started working with me, and I found access to raw DolbyA materials.   Along with very serious research, we have ended up with a super good DolbyA decoder.

OTOH, noone really liked the results that I got with a single DolbyA.   When I remembered that the compression was segmented, I tried multiple DolbyA decoding steps at levels 10dB apart (I tried 6dB -- nogo, 9dB -- nogo), but the only step that sounded really good was 10dB steps.   It took another 2years or so (whack a mole) to find all of the EQ and calibration levels.   (If the settings are wrong, there is audible distortions -- expanders that are expandingmaterial that isn't compressed tend to sound worse than compresors.) The misadjusted decoders with bad EQ sound characteristically bad.

Anyway -- I have to go somewhere.   If you want, I can explain the set-up, and it really does work well for what it is supposed to do.   Recordings REALLY ARE cleaned up -- but my hearing is a problem, so I sometimes make errors.   This time, I would NOT have announced it in THIS specific group if it wasn't really, really good at restoring recordings.

John

Re: About the FeralA decoder -- mucho progress.

Reply #13 – 2020-12-18 22:12:49

Quote from: jsdyson on 2020-12-18 17:20:00

Quote from: NateHigs on 2020-12-18 08:41:03
What is "master tape quality"? Surely if this compressor "undoes the common type of compression used" then the recordings will be all messed up as compression is applied in an artistic manor in most if not all recordings. EQ and compression are applied in tandem in mastering sessions - you cant remove one without affecting the other.
From what I can tell -- there is a mechanical step applied to almost every consumer recording, but the criss-crossed compression results in a similar freq response balance.   The major difference is a lower midrange boost, swishy cymbals and a truly distorted bass.   The midrange (other than the lower midrange) generally has very little gain control applied to it.

So, the results provided to consumers is 'kind of close' to the original, but IS NOT the same as the mix.   This goes beyond good mastering.   I quit the HiFi hobby in about 1990 because it didn't seem that CDs would start sounding any better.   Since I used to do recordings myself (stereo, orchestral), it was easy and so obvious that there was a consistent 'distortion' somekind in the consumer recordings.   At the time, I ddn't have the resources or time to correct the problem --f rankly I wasn't interested.

In about 2012, with truly HiFi virgin hearing, I listened to a CD again -- but with all of the experience over the years, I heard some things that I could 'grab on to'.   The damned material was compressed somehow.   After about 2yrs, I found a segmented approx 2:1 compression in the audio.   The speed of the compression appeard close to DolbyA.   After writing an early, approximate DolbyA decoder, I found regions in the audio where it cleaned it up.

Sadly, for about 5yrs, I was trying only one DolbyA decoder, but the good news is that a very nice recording professional started working with me, and I found access to raw DolbyA materials.   Along with very serious research, we have ended up with a super good DolbyA decoder.

OTOH, noone really liked the results that I got with a single DolbyA.   When I remembered that the compression was segmented, I tried multiple DolbyA decoding steps at levels 10dB apart (I tried 6dB -- nogo, 9dB -- nogo), but the only step that sounded really good was 10dB steps.   It took another 2years or so (whack a mole) to find all of the EQ and calibration levels.   (If the settings are wrong, there is audible distortions -- expanders that are expandingmaterial that isn't compressed tend to sound worse than compresors.) The misadjusted decoders with bad EQ sound characteristically bad.

Anyway -- I have to go somewhere.   If you want, I can explain the set-up, and it really does work well for what it is supposed to do.   Recordings REALLY ARE cleaned up -- but my hearing is a problem, so I sometimes make errors.   This time, I would NOT have announced it in THIS specific group if it wasn't really, really good at restoring recordings.

John

I usually try to meet my goals, but I won't be able to get the first data to compare tonight. If there is any interest as time goes on, I can make new datasets, or teach individuals how to do it.

Being honest, I got jammed up on some recordings where I have some MFSL references -- the results don't always sound the same, for good reason.   The reason is that MFSL does mastering, while I do NOT.   My results can only come from the decoder, and any EQ beyond +-1.5dB outside of the decoder is a NO NO.

However, some of my results were more different than I would hope. In some cases, my results are MUCH better than the reference, but sometimes there is a profound EQ difference in favor of MFSL. I always check the recording source also -- to make sure that there aren't any botches where the sound character is unexpected.   Sometimes this stuff is difficult to do, but has been simplified to two modes (each has 6 common modifiers) and calibration level (just one, even though there could be 10 DA decoders in the chain -- the calibration is calculated for each layer.)   There used to be many many many variables -- there has been MAJOR progress on usability.   Normally I can do a quick decode, pretty good, in about 20-30seconds, but instead on the tests, I am working for precision, and getting the VERY BEST results.   I am NOT cherry picking the recordings, and in fact choosing some recordings that have been real trouble for the decoder in the past.   There is NO cheating in my own demos, testing or testing that I am helping to support (I just make mistakes.)

I am trying for these demos/comparison sources to be very good.   Of course, I am very frustrated by my hearing (which adapts IMMEDIATELY, within a few tries to a frequency response error.)   My ability to detect modulation distortions is probably amazing -- which makes listening to most normal CDs very unpleasant.   Even worse, the new 'loudness wars' stuff is totally intolerable.
My biggest dislike is NOT compression per-se, but it is bad modulation effects.

Think about it like this -- even in a multi-band compressor, there are numerous tones, some loud, some soft.   When the volume of the various components change, they also change levels on associated sounds in the same bands.
There are some psuedo-fixes which might include a Hilbert transform based detector, but I'd suspect that most people misuse them, thinking that they are a panacea, which they are not.   There are some tricks and caveats to get maximum performance using Hilbert detectors.   The modulation problem is not just with instantaneous ripple in the envelope, but one should also consider careful shaping of the measured changes in the envelope -- Hilbert detectors do about 1/2 to 2/3 of the job.

Anyway -- so much for my blather -- Id hope to get the demos ready at +26Hrs from now (approx 19Dec2020, 9:00PM USA EST time.)

John

Re: About the FeralA decoder -- mucho progress.

Reply #14 – 2020-12-20 01:31:43

Here are the limited length comparisons.   Some of the results weren't very good in my opinion, but it would be wrong for me to exclude them.   The choices were made based upon my historical taste along with a couple of 2010's recordings that are also decodeable.
This demo/comparison forced me to clean some things up, but that isn't a matter for discussion here.

There are two kinds of files:   RAW, basically cut from CD, should be same data rate.   'DEC' (decoded), which is the same rate that comes from the decoder.   I believe that I left some of the metadata in, so I believe that you can see the primary decoder settings (most of the command line are for runtime listings, etc -- the actual command is pretty short..

I had to keep the length short for reasons of propriety, but I also produced a 29 second delayed version of 'Roundabout' where there is more interesting music.
As building this group of recordings -- 30 or so of them, I created an infrastucture to produce test recordings en-masse.

Suggestion about listening to the material -- differences that you SHOULD hear, much less hiss on older material, as there should be 20dB to 30dB of noise reduction.   Also, the dynamics are much more strong.
On the other hand, the HF compression on normal consumer recordings can have sparkle, but that 'sparkle' is fully removed while retaining most audible information.
The decoder IS publically available, but there is a new release coming in a few days as I wanted to do the best possible for these demos, I made more improvements.

Important:   I suck t*rds at equalization.   Even though there are few needs for per-recording EQ anymore, I might have made a fundamental EQ error.   SUCH ERRORS CAN BE CORRECTED, and if I still cannot get the correct EQ, I'll instrument a version for someone to try to set-up the correct EQ themselves.   My hearing is good at distortions, but not so good at EQ.

There is ZERO mastering in these tests -- I might have done slight EQ to the tune of +-1.5dB single pole on a few recordings, but that is it. In an hour or so, I'll produce an mpeg of a decoding session, to show the EXTREME amounts of processing -- yet producing reasonably good sound...

Here it is: (use the dropbox decoder only for cursory interest -- it distorts a LOT)
https://www.dropbox.com/sh/v90m7q56g64tfgo/AACao_I34J7x2ZJu91qpKG4wa?dl=0

ADD-ON:   in the interest of full disclosure, the decoder is in development, and there is -1.5dB at 125Hz (single pole, so it isn't severe) in the decoded results. Just recognize that until I can fix them in about 1Hr, that there is a bit of extra rolloff.   I tend to make stupid mistakes on everything that I do -- this is one such silly mistake.

John

Re: About the FeralA decoder -- mucho progress.

Reply #15 – 2020-12-20 11:51:44

I need to re-insert more midrange to bass in the examples, doing the build right now. It will be on the dropbox site within +2Hrs from now. I am doing a quick decoding operation that removes only modulation distortions, and not the associated previous distortion from mixing in the original DolbyA HW detectors used in compression.   So, the highs will still be a little scrambled, but still more clean than the FA version.   I had wanted to update to the super clean version, but decoding at that level, with effectively 8 DA decoders running as a production line, it takes approx realtime on my 10 core X processor using the decoder.   Normal quality can run 8X faster realtime, and the moderate quality decodes can run in about 2X faster realtime.   The super quality decodes run in about 0.5X the speed of realtime, so I don't have time to decode 30+ songs at the highest quality, and then create the snippets, releasing it within 2Hrs.

I hope that the changes have settled down again. Once I decided to do some testing with HA people, I did a re-evaluation, including simplifying decoding even further. I made some major changes/improvements. This happend over the last 48Hrs, and I made a mistake during the minor re-structure. I wanted to 'finish' the last steps -- but got caught with another minor mistake.

Honestly -- my hearing acts like an AGC, and when I work more than a few hours, it loses the ability to choose which EQ to use. (Note that I use the term 'which', not a tweak per se.) There is a sequence of EQ at/below 1250Hz needed, in 250Hz increments.   I didn't know whether to use one or two of them per step.   Normally, I tend to the conservative and used only 1 EQ per step on initial trials. Note that the EQ isn't 3dB per step, but it is two 1.5dB per step. (This 250Hz thing -- the frequency steps must be aligned -- is something that I found out when trying to apply an FA signal to older versions of the decoder -- the odd, little bits of distortion became obvious because of the temporal nonlinearity of decoding DA. Over a few weeks, and many intuitive experiments, I determined that using stepped 250Hz offsets tend to work better.)

* I don't believe that all of the need for frequency steps come from the FA encoding itself.   The need for frequency steps might come from the attack/release nonlinearity on each step of DolbyA encoding. This tends to scramble the decoding effort, and using the steps tends to 'widen' the bandwidth of modulation artifacts so that the DA components of the FA decoding have more bandwidth for quicker responsiveness.      I have found a compatible way to provide the 'steps' while effectively providing the same freq response for many decoding operations (the last few steps do have narrower bandwidth than at the beginning.)

Anyway, on the examples, I had used one EQ per step -- with my hearing compressing the highs, I chose only one. I should have chosen two to compensate the midrange down to the lows.   The direction of EQ is also super important to avoid modulation disotrtions. (There were already two steps each in the midrange -- after many experments while listening for modulaton distortion.)

Being on 'Hydrogen' audio, you might ask: Why use the subjective techniques? Why not use controlled experiments during development?   The answer is obvious: ZERO SPECS.   Every step of NEW development would be extremely time consuming.   A full A/B check would have taken at least a week to do, but this correction will have been done in about 24Hrs or less. There is a huge risk of playing 'whack a mole', but the difference with me is that the goal is not to justify a design, but instead criticise it to be able to make substantive progress.   My bias is NOT status-quo, but instead to make the results better most of the time, with expected changing the design.

Once I have worked through everything, I'll announce later on todya (probably +2Hrs) -- frankly, I had thought that the details had been worked through, but I was wrong (usually true.)
For an example of what I am dealing with, I could, right now, show a section of code, perhaps a page or two -- without specs and the extreme complexity, I had to use intuition. If I had totally accurate DolbyA decoder software back in 2012, it would have only taken a year or so to work through the FA scheme.   As it is, I had a variable accuracy, inaccurate DA decoder -- and any DA decoder that is not perfect will not work at all for decoding FA. (I have already exposed the DolbyA compatible calculation for attack/release, and it is NOTHING at all like a normal compresso/expander design. So, if you want to see the problems of doing design without specs and/or depending on old schematics with selected components, I'll be happy to show a few 100 lines where there are about 10000 lines of SIMD C++ fully dependent on lack of specs!!!)

Do you know the game 'whack a mole'?   That is the game that I have been playing for 8 yrs.   After doing a super high quality decode last night (almost all modulation distortion, even from previous DolbyA operations back at the studio), and I found that the coherent sound quality did further make the lack of lows even more obvious to me (also a few hours sleep.)

+2Hrs, hopefully no more 'stuttering' of releases.   If I had really good, honest, reviewers like from here, and they could describe the defect, I probably would not have made as many mistakes.   No-ones fault except mine.

MEA CULPA
John

Re: About the FeralA decoder -- mucho progress.

Reply #16 – 2020-12-20 15:37:23

I've recovered the code from my last minute mistakes.   The location (the same) of the snippets is the first link below...
The decoder REALLY works, and is almost as easy as doing a DolbyA compatible decode now. (The DolbyA mode is currently licensed, but the FA mode is free touse.)   I'll supply consumers a license file for DolbyA upon request (it will always be free for consumers. Even more free when I open up the source code -- only delay is that it needs internal clean-up.)

1) The bass is now correct.   (really frustrated that I made the mistake)
2) I ran the decoder at a higher-than-usual quality mode: '--fz=opt', which is a good tradeoff between improved qualiity (mitigation of modulation sidebands from the decoder, and some mitigation of any previous DA encode/decode cycles that was done during production.   --fz=max takes more time as --fz=opt is approx realtime, --fz=max is approx 2X slower.   There are almost 'forensic' modes that are incredibly time consuming, but push the sidebands (distortion) down to almost audible insignificance.
3) For those who cannot or not all that interesed in the decoder, I produced a movie from the screen of the decoder running.   The message about the 'movie' is at the end of this posting.
4) I'll be doing the *corrected* decoder release on the site where I regularly work with the testers/those who participate.   When it passes the tests, I'll definitely make it available here.

First -- the new decodes have replaced the old ones, and here the snippets are (you should notice more correct bass, but a surpisingly clean sound on most recordings):
DECODES:
https://www.dropbox.com/sh/v90m7q56g64tfgo/AACao_I34J7x2ZJu91qpKG4wa?dl=0

================================================================
Here is info that I wrote for the movie (the sound is that of an earlier decoder.
The information is 'dense', but when you look at the video:

The layer is specified on the left hand side.   Each decoding operation is comprised of multiple DA decoding steps to produce an FA decoded result. The gainsfor each band are in each of 4 columns -- the first is LF, then MF, then HF0, then HF1. The most active are HF0/HF1. Each gain for each layer has a MID/SIDE component. (On DolbyA it is Left/Right.)

The number to the far right, for each layer, is the output level for that layer (DA output before EQ):

When you look at each of the gain values, there are three numbers: lowest, average, highest.   On the HF1 band, you might see something like: -14.50dB, -14.50dB, -14.50dB for when there is no audio.   When there is audio, you might see something like: -14.50dB, -6.00dB, 0.00dB.   This means that the gain actually sweeped between -14.50dB and 0dB, probably between one and ten times for each layer (yes, it could be 10 times!!!), multiple times during the sample each second. (a lot of gain change.)

Again, when you look at each gain each band (LF through HF1), the effective gain is all added up. So, when you see that there is 8 layers, and the gain can be sweeping all the way from -14.50dB to 0dB or -10dB to 0dB on 3-4 of the layers out of 8, that is one hell of a lot of gain control to do without distortion!!!

The audio output that you hear is DECODED from a CD.

Here is the movie: (you'll have to download it first.)
https://www.dropbox.com/s/xy5oz7c1tywuacg/example4-2020-12-20_01.04.02.mkv?dl=0

John

Re: About the FeralA decoder -- mucho progress.

Reply #17 – 2020-12-20 21:12:08

The video is viewable with Firefox 84.0 so no need to download for me at least..

Re: About the FeralA decoder -- mucho progress.

Reply #18 – 2020-12-21 09:16:55

I'm so confused after listening to the samples posted... Are they named in correctly?

All I hear is what equates to an extremely warm EQ (and other artefacts). The presence and fidelity of the recordings is gone. Am I wrong? There's no way a master recording would sound this muffled. It's made them sound like b roll amateur demos.

@jsdyson, by no means do I want to diminish your work, clearly a lot of effort and knowledge has gone into this. I've not been able to make it through all of your posts yet, as they can be quite verbose, however - are you just trying to make music sound like how you think it should? Maybe you don't agree with these mastering choices, but a lot of music is purposely giving this shining, bright sound. I also don't understand the Dolby-A stuff, as some of the music you posted would have been recorded digitally (I'm talking Nikita here, not so familiar with ABBA) and would never be subjected to and Dolby processing.

Are you sure your listening setup is correct?

Re: About the FeralA decoder -- mucho progress.

Reply #19 – 2020-12-22 10:15:56

Quote from: NateHigs on 2020-12-21 09:16:55

I'm so confused after listening to the samples posted... Are they named in correctly?

All I hear is what equates to an extremely warm EQ (and other artefacts). The presence and fidelity of the recordings is gone. Am I wrong? There's no way a master recording would sound this muffled. It's made them sound like b roll amateur demos.

@jsdyson, by no means do I want to diminish your work, clearly a lot of effort and knowledge has gone into this. I've not been able to make it through all of your posts yet, as they can be quite verbose, however - are you just trying to make music sound like how you think it should? Maybe you don't agree with these mastering choices, but a lot of music is purposely giving this shining, bright sound. I also don't understand the Dolby-A stuff, as some of the music you posted would have been recorded digitally (I'm talking Nikita here, not so familiar with ABBA) and would never be subjected to and Dolby processing.

Are you sure your listening setup is correct?

First, my announcement was a little premature.   The results are now ready.   It is important to do an immediate A/B, because when you are used to listening to the RAW sound from CDs, you won't hear the horrible hiss on them. (I have the problem if I listen to them too much.)   Also, with continual listening to compressed materials like CDs, then the brain somehow 'expands' it.
Even though I dropped out of the HiFi hobby back in about 1990 because CDs sounded so bad, it wasn't until a casual listen in 2012 with high quality equipment that I heard some 'tells' in the CD recordings.   It took me over a year (perhaps two) to figure it was SOMEHOW DolbyA related.
Initially, I have two suggestions:
1) Be patient with relearning what a recording really sounds like (really -- go to a music store and talk into a TLM103 or somesuch -- sounds NOTHING like vocals in a CD.)
2) Be patient with me.   I make lots of mistakes, like my pre-mature announcement. It was getting closer than ever, but then some of my reviewers 'corrected' me.
3) This is an exquisitely complicated program with math that would knock your socks off. Some of the algorithms do not exist in public (esp regardning removing the modulation components from gain controlled material.) BTW, the math 'knocks my socks off' also -- it isn't easy stuff.

As of now, the comparisons made at 3:30 (or something 30) wherever you are in the world, today on 22 Dec -- those are the only valid 'DEC' versions.   The next posting from me will be a prelim release of the decoder...   The current command line is designed for DolbyA decoding and most importantly here -- FA decoding, but with numerous prospective modes.   In the last week, the reverse engineering became complete, and the command line will be simplified.   It was already very simple for DolbyA, but decoding of pure DolbyA requires a license file, which I'll happily provide to consumers (or most commercial) people gratis.

I'll explain the simplest use of the command line in the next posting, and a pointer to the program.   It works best in Linux, but does a so-so command line job on Windows.   (I like Linux because using pipelines is normal instead of as an exception, and Linux is more responsive.)   However, the code is tested on Windows also.

https://www.dropbox.com/sh/v90m7q56g64tfgo/AACao_I34J7x2ZJu91qpKG4wa?dl=0

John

Re: About the FeralA decoder -- mucho progress.

Reply #20 – 2020-12-22 10:46:45

Here is the location of the DHNRDS FA decoder.   It can decode materials that were recorded with DolbyA, but requires a license file that I will supply gratis.   But, we really aren't talking about decoding DolbyA here.

There is a problem with most digital recordings - they are heinously DRC compressed*, and our hearing has become adapted to it.   It takes a while to unlearn the 'sound'.   Sometimes, in fact, that 'compression' does sound better, but most of the time it does not. Do a direct A/B comparison (blind comparison isn't needed -- the improvement from RAW to DEC is mostly profound.)
* When I say 'heinously compressed', I don't mean extreme 'loudness wars' compression, but a different sort of DRC.

Using the decoder is relatively simple, and you can use pipelines instead of --input and --output.   It takes wav files, and for a CD .wav file at 44.1k, it will output 88.2k, only because the internal processing cannot work at 44.1kHz sample rate.   At rates above 48kHz, the in sample rate will be the same as out (e.g. 384k in/ 384k out.)   There is no benefit to using sample rates above 96k, becauase the freq response limit is a flat 40+k (at higher sample rates), or a few kHz below Nyquist at low sample rates.   DolbyA units are NOT flat to 40+k, but instead about 35k -- the DA/FA decoder IS flat to 40kHz when it can be.

Here is the simplest command line:

da-avx --input=in.wav --output=out.wav --info=1 --tone=-42.675 --fcs="6,auto,fgG" --wof=1.19 --outgain=-10
ADD-ON: use the '--overwrite' switch so that old '--output' files can be overwritten.   This feature was intended for situations where high value recordings are nearby.

(The --fcs argument will always be similar, maybe changing the first number -- the last two switches are about level and unwarping the stereo image, mentioned in the Usage guide along with the --info switch. You need 'outgain' because the output can sometimes be greater than the input peaks at 0dB.)   Everything is in units of dB.   The --wof is sqrt(sqrt(2)), and spreads the stereo image to close to correct.   The decoder LIKES floating point wav files, but works fine with 16bit, 24bit and FP files input and 24bit or FP files on output.   To force floating out, use --floatout , for signed integer 24bit, use --intout.   Floating is preferred for the program because of the focus on pro use, like the creation/support of BEXT and support for RF64.

The --tone value is roughly equivalent to the typical DolbyA tape tone level of -12.80, except the program does have an 'undesired' offset, and with four 10dB layers in the first sequence (sorry about the terminology), it matches -12.675.   The decoder isnt' too far off, but in DolbyA mode, then the reference tone measurement directly tells you what the calibration (--tone=) should be.   Again, this is NOT about DolbyA, but the two schemes are inter-related.

There is a Windows version (no installer, but instructions are in the big manual.)   Just kerplop it (and .dlls and aoplay.exe) into a 'bin' directory, and you should be able to access it by a properly specified PATH variable.   Also, it can work in the current directory.
The Linux version is super trivial, probably obvious for most Linux advanced users or programmer.

Because of all of the dense inline SIMD, and splitting the code into CPU types would be a mess and probably affect the speed -- there are 3 versions on Linux (avx, sse3 and avx512) and 3 versions on Windows (core2, atom, avx).   Core2 and atom are essentially the same.   Except for avx512, choose the most complete architecture for your machine.   AVX512 can be coaxed into being faster than AVX, but requires careful set-up, so I suggest just using AVX version when you can.   The code runs over 95% in SIMD, and is carefully crafted for efficiency, but is still a bit slow.... The program does NOT like hyperthreading, because it will take every CPU resource availalble -- there is an affinity switch for Linux to restrict to single hyperthreads, but not yet on Windows. The program will work on hyperthread machines, but machines with 4cores, there will be a performance impact if hyperthreading is enabled.

The program is very slow when you enable the anti-MD (modulation distortion avoidance) modes.   Most people don't really hear modulation distortion, or no-one could listen to almost any recording being sold today.   Orban has played some games to move it around to be less ugly onto upper sidebands, but the DHNRDS scheme goes much further.

Enabling the anti md:   The simplest mode is --fx, and removes the thin/fine amounts of distortion. --fz removes more of it, and there are really aggressive modes that strongly clean up the recordings, but no need to deal with that here. --fx should go before the --fcs switch.

The program started as a research project, and has NOT been formally productized, etc.   No installer, but it does work.   Newer versions are going to get rid of the '--fcs' switch because changing modes was found to be unneeded.

There will be a source release in months (perhaps >1yr), not years (not greater than 2yrs), because there is planned a lot of clean up (e.g. removing a lot of the code supporting the previous experiments and reverse engineering.)   This hasn't been a weekend project, and 'DolbyA' wasn't originally the target -- it took two years to figure that out. This thing REALLY DOES work.

HERE IT IS:
https://www.dropbox.com/sh/1srzzih0qoi1k4l/AAAMNIQ47AzBe1TubxJutJADa?dl=0

Enjoy!!!
John

Re: About the FeralA decoder -- mucho progress.

Reply #21 – 2020-12-22 12:48:43

Quote from: jsdyson on 2020-12-22 10:15:56

Initially, I have two suggestions:
1) Be patient with relearning what a recording really sounds like (really -- go to a music store and talk into a TLM103 or somesuch -- sounds NOTHING like vocals in a CD.)
2) Be patient with me. I make lots of mistakes, like my pre-mature announcement. It was getting closer than ever, but then some of my reviewers 'corrected' me.
3) This is an exquisitely complicated program with math that would knock your socks off. Some of the algorithms do not exist in public (esp regardning removing the modulation components from gain controlled material.) BTW, the math 'knocks my socks off' also -- it isn't easy stuff.

John,

a quick retort.

1) I understand that a direct feed doesn't sound like the finished material. This is why teams of people are employed to (essentially) post process the material to make it sound good (or different at least). This is why we invented EQs, compressors, etc. There's a whole industry devoted to this. For some reason your tastes seem to be stuck with records made in the 60s, with this very warm, dense sound (which, I do like too. This is why the original Teo Marcel masters of the Davis stuff are my fave). These changes are very much the intent of the artist. Why do you want your recordings to sound like unfinished, unmixed, unmastered demos?

2) Of course, it's clearly a very technical and complex device you're making. Well beyond anything I understand.

3) I understand, and fully commend your efforts. However, I guess my criticism is that what you're doing is selling this as a device which "fixes these recordings". When in reality what you're making is a device which drastically changes the recordings to match your own personal preference.

Re: About the FeralA decoder -- mucho progress.

Reply #22 – 2020-12-22 16:45:52

Quote from: NateHigs on 2020-12-22 12:48:43

Quote from: jsdyson on 2020-12-22 10:15:56

Initially, I have two suggestions:
1) Be patient with relearning what a recording really sounds like (really -- go to a music store and talk into a TLM103 or somesuch -- sounds NOTHING like vocals in a CD.)
2) Be patient with me. I make lots of mistakes, like my pre-mature announcement. It was getting closer than ever, but then some of my reviewers 'corrected' me.
3) This is an exquisitely complicated program with math that would knock your socks off. Some of the algorithms do not exist in public (esp regardning removing the modulation components from gain controlled material.) BTW, the math 'knocks my socks off' also -- it isn't easy stuff.

John,

a quick retort.

1) I understand that a direct feed doesn't sound like the finished material. This is why teams of people are employed to (essentially) post process the material to make it sound good (or different at least). This is why we invented EQs, compressors, etc. There's a whole industry devoted to this. For some reason your tastes seem to be stuck with records made in the 60s, with this very warm, dense sound (which, I do like too. This is why the original Teo Marcel masters of the Davis stuff are my fave). These changes are very much the intent of the artist. Why do you want your recordings to sound like unfinished, unmixed, unmastered demos?

2) Of course, it's clearly a very technical and complex device you're making. Well beyond anything I understand.

3) I understand, and fully commend your efforts. However, I guess my criticism is that what you're doing is selling this as a device which "fixes these recordings". When in reality what you're making is a device which drastically changes the recordings to match your own personal preference.

Actually, I do have a master tape under NDA where I took an FA version of it, and decoded it.   It certainly sounds similar after deocoding.   Otherwise, I don't have as many examples as for DolbyA.   The reason for not having as may comparisons as for DolbyA is that I havent encountered FA versions of them.

The resulting recordings sometimes need 'mastering', which I do not do. Feel free to do a treble cut on some of them -- I am not about mastering, my project does the part of correcting dynamics.   Mastering IS a matter of taste, the FA decoding process is very mechanical.

Also, if the *decoder* didn't exactly match the compression characteristics, then all kinds of sins (currently non audible because of correct behavior) start happening. I can make a version that deviates from DolbyA processing -- you wanna hear the results?   The answer: A LOT WORSE.

On the other hand, just restoration of the dynamics and getting rid of hiss is a big advantage.   You can think of it as an improvement (as soon as you can adjust away from the aural adaptations to the compressed sound), instead of deocoding.   I'll bet you that a lot of even young people who have fully adjusted don't even hear the hiss!!!   I am 64yrs old, with tinnitus, and can STILL hear the hiss.   The dynamics are so badly squished on the consumer recordings that there MUST be some kind of mental process dynamic range expansion to even tolerate it.   If you start with virgin hearing (like I did in 2012), you might really be surprised about how bad the recordings sound directly off the CD or download.   I have found some music videos that are more conventionally compressed though -- they DO sound better.

There is NO linear device which can correct the nasty digital sound. Trying any such improvement without dynamics processing is futile.   Any linear high tech, grounding improvement, jitter reduction or tube amp cannot correct it.   I would expect if the well known TBC technique uses masters without allowing the nasty compression, they would also sound better, even though the TBC tecchnique should certainly help a little also.

ADD-ON: why wasn't this done before:   ANSWER, it is difficult to do, and no accurate SW decoders capable of decoding DolbyA materials until mine.   (There are both approximate and fake decoders, but the DHNRDS does everything that a DolbyA does for decoding, but much better.) (It might have some errors, but so do DolbyA units from version to version.)   Where there are differences, the DHNRDS tends to provide better clarity because of more agile processing and understanding of modulation distortions causing a fog.

John

Re: About the FeralA decoder -- mucho progress.

Reply #23 – 2020-12-24 01:02:43

There will be a substantial quality update for the decoder coming.   Because I have such unreliable hearing, I am dependent on either my hearing working correctly for an instant, or someone explaining what is wrong.   My hearing let me down on detecting the defect.

Anyway -- there was a bit of a 'grind' in the sound, which I know exactly the problem, but couldn't perceive it.   Anyway, some of the front-end MD avoidance was too heavy-handed, slowing down the gain slew too much. (cannot do that in a feedback loop very easily.) The effect of the nonlinear anti-MD being heavy handed is counter-intuitive, causing a grain in the sound...   I just didnt' hear it, and know exactly the steps that need to be taken.   The tradeoff is making sure that the negative effects are below the ability to perceive, but still have some anti-fog effect.   Honestly, I did use a heavy handed setting.

There will be a new decoder release in about 2days, with no changes except a slight tweak in the front end anti-IMD code, and the correction in the brute force anti-MD.   (The anti-MD that I am speaking about is NOT the high-tech portion.)

The comparison demos will be updated tomorrow, but I have a large amount of testing to make sure that the improvement doesn't cause regressions elsewhere.   Happily, both changes are in semi-optional pieces of the code -- so the risk is very low. The changes can only create an improvement.

This fix has an effect on both decoding DolbyA materials and FA.
Just wanted to make sure that if you heard the defect, that such defects are not tolerated, it is just that my hearing isn't the same as it was 40yrs ago.   And.. most importantly, there is NO SPEC. Whack a mole is the game that I play all of the time. (about bugs that are just frustrating to perceive and fix.)

John

Re: About the FeralA decoder -- mucho progress.

Reply #24 – 2021-01-01 08:29:00

Quote from: jsdyson on 2020-12-24 01:02:43

There will be a substantial quality update for the decoder coming.   Because I have such unreliable hearing, I am dependent on either my hearing working correctly for an instant, or someone explaining what is wrong.   My hearing let me down on detecting the defect.

Anyway -- there was a bit of a 'grind' in the sound, which I know exactly the problem, but couldn't perceive it.   Anyway, some of the front-end MD avoidance was too heavy-handed, slowing down the gain slew too much. (cannot do that in a feedback loop very easily.) The effect of the nonlinear anti-MD being heavy handed is counter-intuitive, causing a grain in the sound...   I just didnt' hear it, and know exactly the steps that need to be taken.   The tradeoff is making sure that the negative effects are below the ability to perceive, but still have some anti-fog effect.   Honestly, I did use a heavy handed setting.

There will be a new decoder release in about 2days, with no changes except a slight tweak in the front end anti-IMD code, and the correction in the brute force anti-MD.   (The anti-MD that I am speaking about is NOT the high-tech portion.)

The comparison demos will be updated tomorrow, but I have a large amount of testing to make sure that the improvement doesn't cause regressions elsewhere.   Happily, both changes are in semi-optional pieces of the code -- so the risk is very low. The changes can only create an improvement.

This fix has an effect on both decoding DolbyA materials and FA.
Just wanted to make sure that if you heard the defect, that such defects are not tolerated, it is just that my hearing isn't the same as it was 40yrs ago.   And.. most importantly, there is NO SPEC. Whack a mole is the game that I play all of the time. (about bugs that are just frustrating to perceive and fix.)

John

IMPORTANT: I had been misusing my own decoder.   Basically, I was trying to remove traditional 'compression' by mistake. I kept getting feedback about that, so I re-adjusted the usage (along with the normal progress/improvements from numerous aspects.)

Frankly, I found 2 bugs in the DA mode and also 2 improvements in the DA mode also.   This is a bit of an embarassment, but there were some (usually minor) tracking errors at very low levels.   The attack/release calculations needed more feedback than the 'gain of 1' that I was using.   So, after a lot of testing, and intuit the design of the true DolbyA, I increased the gain to 3.16 (equiv of loop gain to 2.16 for the low level attack/release speed calculations) with much better very low level behavior.

On the FA side, not many changes other than making the EQ more consistent so that settings drift much less often, and usage is much easier.   I wont' be doing an official release until later today, but there is a Linux version available on the distribution site since it is automatically created.

The comparisons reside as below (same place), and the sound qualities are generally more consistent with the original, sans hiss and sans scrambled sound.   I regularly make the wrong choice in EQ, and there are a very limited set of EQ choices, but adequate for all recordings that I have encountered.   When doing mass efforts, I do make mistakes, but generally ANY errors can be corrected with minor EQ.   The BIG difference is that there is now NO attempt to undo normal compression -- I simply didn't realize that the decoder settings were causing that to happen. (It is all about the calibration level -- no other changes.)
https://www.dropbox.com/sh/v90m7q56g64tfgo/AACao_I34J7x2ZJu91qpKG4wa?dl=0

Notice