The DHNRDS-DA Decoder (FA Mode) Best Settings List

[John Dyson]

I intend to start adding some entries to the list in a day or so...  Today, I am nailing-down the settings for a few recordings being used for testing & review.  So -- I'll definitely be contributing to this list and 150% in favor of doing this now.  I am so glad that @lucretius is getting this started.

 

My commitment:  The decoder is 'stable' in the sense that settings that work now will also work the same in the future.   Any changes will be add-ons, currently available standard settings will not be removed.  Some features might be added, but won't interfere with currently expected behavior.  That is:  documented decoding behavior will be consistent in the future.

 

@Lucretius timing is perfect.

 

Add-on:  when I do a contribution along with everyone else, I want EVERYONE to know that it should okay to create alternative settings suggestions.   Some of the variability might result from subtle differences in the recordings, others might simply be matters of 'taste' or 'hearing'. 

 

The group/moderator/etc might either choose to negotiate the variants or maybe just document multiple decoding choices?   As anyone knows who uses the decoder, sometimes two settings sound similar -- there are SOMETIMES multiple 'best' choices.   This quandry is especially real for my results - sometimes choices might eccentric because of extreme variability in hearing ability.

 

I would rather see multiple versions of 'reasonable' settings instead of someone suppressing a strongly held opinion.  We should not dismiss a differing opinon -- but try to understand the basis for the differences.

 

John

 

[John Dyson]

When I looked at the postEQ field -- I realized something important that might help people when they do their own decoding & determining the EQ.  Since this is about using the decoding for a given recording -- I have a few comments about that process.   I do suggest AUTOMATING your process as much as feasible.

 

Here is a suggestion about keeping the 'before EQ' version:

When adding post EQ,  consider keeping your original  'before EQ'/direct output from decoder version around also.  

 

My own process is usually to either create an 88.2k  'floating point' .wav file or a '24 bit signed integer' .flac file before EQ.   I do the EQ  and conversion to the final/listening format in a 2nd pass after very careful review and listening.   Your final output format might depend on the quality of the recording, and for me, is usually 88.2k 24bit signed integer or 48k 24bit signed integer.   If you need to make a smaller file for your brother/sister or your portable listening, then when you start with one of these two -- you are in a perfect position to create almost anything.   Since I almost always keep the original 88.2k .flac or .wav file produced directly by the decoding operation, then I can always produce that 'perfect' 88.2k/24bit version if I really want -- even if I might normally produce only a 48k version for certain material.   (I often pipe directly to sox, and let sox do the conversion to 'flac'.  Warning -- that SOX .flac conversion loses the in-file metadata!!!)*

 

* Because of the loss of metadata when using sox for conversion, I am seriously considering an optional shared library link to a .flac library, but I really need to be careful about licensing because the decoder is not free source-code software yet.  It is free to use, but I do not want to disclose source until I can mathematically defend the tricky parts.

 

What I am trying to say is keep your direct decoder output version if you can.  Stash the 'unmolested' away in your archives or wherever you put things away.  The reason is that a good decode can take many minutes, but another EQ try only takes seconds.   Don't lose the original decoder output, just in case the EQ needs to be retried.  Re-decoding something is something that can take very significant amounts of, unless you want to try different settings for decoding.

 

EQ is trickier for me, because when decoding, the recording often gives stronger hints about the best settings.  EQ is more based on taste, but SOMETIMES recordings give hints for EQ also.  You'll see that my upcoming Breakfast in America decode does use a little more EQ than my personal guidelines.  There were serious and important reasons for my choice -- and I'll note them.   My own rule for EQ -- have real reasons, based on perception, why I did post decoding EQ.

* it is perfectly okay to do post-decoding EQ for taste reasons.  My reason for avoiding doing much EQ is that my own taste is not the same as everyone else :-).

 

John

 

[John Dyson]

This might be a little premature, but I am adding the first 3 Supertramp decodes.  However, I just realized that the intensity of the recordings might be coming from the IEC vs. NAB tape EQ.   The basic decoding parameters are pretty close to correct, so I am adding these entries with the caveat.  Most recordings won't have this problem...

 

At least, we are getting this started.  If it wasn't for the IEC issues, these would be pretty good...

 

Artist:                         Supertramp
Title(Format):                  Crime of the Century (CD)
Original Year:                  1974
Label:                          A&M Records
Release Country (Year):         UK
Catalog Number:                 393 647-2
Number of Tracks:               8
External Links:                 https://www.discogs.com/Supertramp-Crime-Of-The-Century-/release/8987246
Submitted by:                   @John Dyson
DA Commands:                    --fb=a --tone=-14.47 --wof=1.19
Pre/Post EQ (Sox):              gain -0.40 treble -0.375 9k 0.50q treble -0.375 9k 0.8409q
Comments:                       Maybe too intense because of IEC tape EQ? Needs IEC conversion on output?

 

 

Artist:                         Supertramp
Title(Format):                  Breakfast In America (CD)
Original Year:                  1979
Label:                          A&M Records
Release Country (Year):         ?
Catalog Number:                 393 708-2
Number of Tracks:               10
External Links:                 https://www.discogs.com/Supertramp-Breakfast-In-America/release/12834576
Submitted by:                   @John Dyson
DA Commands:                    --fb=aX --tone=-13.90 --wof=1.09
Pre/Post EQ (Sox):              gain -0.50 treble -0.375 9k 0.50q treble -0.375 9k 0.8409q treble -0.375 12k 0.50q treble -0.375 12k 0.8409q treble -0.375 15k 0.50q treble -0.375 15k 0.8409q
Comments:                       Maybe too intense because of IEC tape EQ? Needs IEC conversion on output?

 

 

Artist:                         Supertramp
Title(Format):                  Even In The Quietest Moments (CD)
Original Year:                  1977
Label:                          A&M Records
Release Country (Year):         Europe
Catalog Number:                 394 634-2
Number of Tracks:               7
External Links:                 https://www.discogs.com/Supertramp-Even-In-The-Quietest-Moments/release/3102806
Submitted by:                   @John Dyson
DA Commands:                    --fb=a --tone=-14.47 --wof=1.19
Pre/Post EQ (Sox):              gain 0.90 treble -0.375 9k 0.50q treble -0.375 9k 0.8409q
Comments:                       Maybe too intense because of IEC tape EQ? Needs IEC conversion on output?

 

 

[John Dyson]

7 hours ago, lucretius said:

 

Hi John,

 

Are those SoX effects pre or post DA decoding?

 

 

 

Any EQ is post decoding, and my previous comments about output IEC were wrong.  My posted decoding values stand as 'best' so far.   All pre-decoding EQ is done inside of the decoder.   I try to keep the post decoding EQ down below 1.5dB, even though 'Breakfast' was a total of -2.25dB, which is on the strong side.   I kept getting complaints about the HF being too strong.  If you see me doing THAT much EQ, it doesn't result from my own opinion.  My habit and listening tests see some recordings needing between -0.75 and -1.5dB -- not sure what the reason is -- I don't do that unless I hear a defect.

 

Explanation about the IEC: initially, the results sounded good, tamed the sound.  This morning -- with better "virgin' morning hearing, the sound was deadened, and the excessive result was found on the spectogram.

 

When doing comparisons against my reference recordings, where the DHNRDS might be more bright -- the other reference recordings sound like the DolbyA HW 'fog' more than relative suppression of highs.   The DHNRDS retains dynamics, so might appear to be a bit more intense than reference recordings.

 

John

 

[John Dyson]

5 hours ago, AnotherSpin said:

Would it be possible to use this on mac OS? Thank you.

We found that the Linux or Windows version might work under an emulator on MAC OS.   Honestly, I don't have a MAC, and there is zero funding for the project (and I don't want funding -- this is a contribution not a money making effort!!!)   Someone got early versions of the decoder in DolbyA mode  to work on a MAC under an emulator.

 

I don't know anything about emulators on MACs, but it is important that they do full multi-CPU/multi-threading emulation.

 

You know the old comedy Q&A from the movie 'Vacation', paraphrasing the question:   How much CPU do you need?  Answer: How much do you have?

 

My next machine will probably be an AVX512 in a few months.  AVX512 plus 8 or more cores will make the decoder run much more quickly.

 

You might ask -- is 'John an inefficient programmer?'  Why does DolbyA/FA decoding take so long? 🙂  Well I might be an inefficient programmer 🙂, but that isn't the reason why the decoder is slow.   It is slow because it doesn't emulate a DolbyA, but does lots more to avoid distortion.  It is intended for archives, less-so day to day listening.

 

However, if you want to try to set up an emulator on your Mac, I'll do what is necessary to tweak the decoder to help it work better for you!!!

 

John

 

[John Dyson]

Here are Carpenters 1970/1971 settings

These are the CDs that I have used for years. (some metadata on stashed away CDs -- sorry)

More to come.

 

Quote

Artist: Carpenters
Title(Format):  Close To You
Original Year:  1970
Label: A&M Records
Release Country (Year): US (early, from the 80s)
Number of Tracks: 12
External Links: https://www.discogs.com/Carpenters-Close-To-You/master/84990
Submitted by: @john dyson
DA Commands: --fb=a4 --tone=-13.80
Pre/Post EQ (Sox):

 

Artist: Carpenters
Title(Format):  Carpenters
Original Year:  1971
Label: A&M Records
Release Country (Year): US (early, from the 80s)
Number of Tracks: 10
External Links: https://www.discogs.com/Carpenters-Carpenters/release/1700734
Submitted by: @john dyson
DA Commands: --fb=a4 --tone=-13.475
Pre/Post EQ (Sox):

 

 

 

[John Dyson]

The first ABBA decode.   This is as good/clean as my results have ever been.  The decoder settings are based on distortion, not EQ - so you might wish a different EQ choice.  WIth the settings below, you can use whatever EQ you want, and the sound will remain very clean.

This is an early ABBA CD -- most of the early ABBA CDs require very similar settings, and some require the exact same settings.  More will be coming.

 

Quote

Artist: ABBA
Title(Format):  Ring Ring
Original Year:  1973
Label: A&M Records
Release Country (Year): Europe (1990)
Catalog Number:  Polydor 843 642-2
Barcode: 4228436422
Number of Tracks: 12
External Links: https://www.discogs.com/ABBA-Ring-Ring/release/8531127
Submitted by: @john dyson
DA Commands: --fb=ay --tone=-13.4775 --wof=0.8409
Pre/Post EQ (Sox): treble -0.375 9k 0.8409q treble -0.375 9k 0.50q treble -0.375 8.75k 0.8409q treble -0.375 8.75k 0.50q

 

 

[John Dyson]

Next ABBA decode -- The 'ABBA' album is one of the group requiring same parameters as 'Ring Ring'.  (SuperTrouper and Waterloo, maybe a few others are a little different.)

 

Quote

Artist: ABBA
Title(Format):  ABBA
Original Year:  1975
Label: A&M Records
Release Country (Year): Japan (1989)
Catalog Number:  PolydorKK P33P 20056
Number of Tracks: 11
External Links: https://www.discogs.com/ABBA-ABBA/release/10522045
Submitted by: @john dyson
DA Commands: --fb=ay --tone=-13.4775 --wof=0.8409
Pre/Post EQ (Sox): treble -0.375 9k 0.8409q treble -0.375 9k 0.50q treble -0.375 8.75k 0.8409q treble -0.375 8.75k 0.50q

 

[John Dyson]

Wow -- I was just getting ready to publish the 'SuperTrouper' numbers -- and was looking up/comparing my normal/quick access .flac files for the album, searching for the originals -- so I can then pull together the metadata external to the files.

Lo and behold, during my comparisons, I found an AMAZINGLY, EVEN BETTER copy of SuperTrouper.  I cannot believe that I missed it.  You gotta realize, that back in the day I bought lots and lots of material -- for some reason I collected it, knowing that it sounded like cr*p, but also somehow looked into the future and realized that I could eventually clean the recordings up...

The version that I have been 'proud' of up until now, has a pretty darned good peak-RMS of 20,67, but this just found version sitting, hidden in my archives has a peak-RMS of 21.10.  The difference doesn't seem like a lot, but it makes a substantial difference in quality, especially when decoding FeralA.  (A lot of 'normal' copies of SuperTrouper have 15-18dB peak-RMS -- very significantly compressed.  The 'Complete' Studio recordings have peak-RMS of 13dB!!!)

Almost every time I go back through my collection, and review the recordings by explicit measurment, it seems like I find something very significantly better!!!  WOW!!!

 

It appears that the 'super-duper' version benefits from slightly different parameters, so I'll probably publish the parameters for both versions.  The SuperDuper version is PolyGram/PolyDor OZ, 800 023-2.

 

John

 

[John Dyson]

5 hours ago, lucretius said:

 

Do you mean this one:  https://www.discogs.com/ABBA-Super-Trouper/release/5409065 ?

Yes, I think so...   It is *really* clean and has more clean, slightly less compressed dynamics.   It is possible to de-fuzz it quite a bit more than the normal 'SuperTrouper'.   (The SuperTrouper title song is one of the windmills that I tilt.)  I was elated to find this because it IS an important test case for me.

 

John

 

[John Dyson]

6 hours ago, jabbr said:

Are there settings that go along with this?

Yes, more, new, easy-to-use decoding parameters are coming along with the new decoder update.  You CAN use the V1.4.1F decoder to decode it nicely, but the V1.4.2A (or B) decoder has two new subfeatures that allow digging deeper into the signal.   This is why I stopped publishing new parameters for now.   Richard asked for some more features (e.g. an .ini file), so I decided to delay until Sunday night so that I don't keep stuttering releases.

 

With the addition of the new settings, I intend that V1.4.2A contains the last new mode until a significant improvement is needed.  I believe that it is time to freeze, because the quality has been very good, and an expanding feature set can contribute to more damage/confusion than help.  On very extreme cases,  I can explain the internal commands on a on-on-one basis.  It is not simple to map the 'sound' to an internal setting like it is when using the '--fa' series of commands.  This is why it is important that we all nurture and protect the '--fa' series of commands (like '--fa, --fb, --fc, and --fD.)

 

ALL of the previous published parameters are good, very good.  However, I decided to expose another previously hidden capability which benefits some of ABBA, Carly Simon and MAYBE EVEN Supertramp.  It might be easier to resolve a problem while decoding Supertramp with the new user accessible feature.  So, the new modes enable a simple tweak, similar to the 'a' and 'b', but do something previously unavailable in the '--fa' command series.  (Note that I eat my own 'dog food', andI restrict my normal usage to the suggested command set -- I  use the internal controls only to consider upgrades to the standard command set.)

 

(There is incredible flexibility in the general command line interface, but I am keeping the --fa, --fb, --fc and --fD commands at a sane level of complexity -- only adding capabilities when there is noticeable benefit.)   Manually controlling the decoder needs synchronization of various frequencies/Q values/gains and EQ/filter types.   Anyone with deep audio experience could directly control it, but it quickly becomes complex and UNFUN.  I never use the low level commands nowadays unless I am testing for upgrading the normal '--fa' based commands.

 

John

 

[John Dyson]

6 hours ago, jabbr said:

Those settings with --wof= ? .8409

I left --wof=1.0 (or the default) because I couldn't justify a change.  This is one of those 'fine edged' things that I am not 100% comfortable adjusting unless I have proof.  Previously, I had modified the stereo image beyond the default -- but recently re-visited the matter.

 

Background on the stereo image -- it is a very subtle thing that spent a lot of my time. When developing the decoder, I knew that the stereo image was messed up, but not sure what to do about it.  At one time, I tried modifying it on 'before decoding', and it sometimes worked okay.  Then, I tried 'after decoding', and it worked okay on other recordings.  I couldn't get 'most' recordings to do the right thing, until about a few months ago, when I had a Eureka that 'MAYBE' both need to be modified all of the time, by about sqrt(2) -- I was right.  Everything always came out plausibly correct (no severe shifths between foreground and background) when doing both.  However, sometimes the image was still messed  up.   I got into the habit (a bad one) of modifying the image by default, because I was used to doing it.  Sometimes, the default is a good value.

 

A good example of a case where the image MUST be modified -- the Led Zeppelin album that I recently did -- it sucks unless the setting is wof=1.19...   It isn't happy with 1.10 and isn't happy with 1.41 -- it really likes 1.19.  This, along with other things, validated my fractional power of 2 approach.

 

John

 

[John Dyson]

Take a look at the FeralA decoder announcements -- new decoder, well worth it.  All known 'compromises' are now fixed but previous decoder settings are still 100% valid.   Mostly, we just have more control.  Personal note:  family member getting operation, might be in and out, so a LITTLE less responsive for the next few days.  Believe me -- this new version V1.4.2A is really worth it.  If there will be stuttered releases (something in the next few days),  only because of uncaught bugs.  This code is REALLY NEW!!!

 

John

 

[John Dyson]

The new V1.4.5F release of the DHNRDS has had some serious polishing and some updates.  This includes support for the settings method and a newer, easier kind of settings.  I'll be updating my setts for @lucretius's list soon (in a day or so -- busy today.)   I got feedback from my project partner about some issues with the DA processing, so the new DA decoder works even impossibly better.   Also, there is a simplification of the FA decoding (even though the man page might look longer, the actual new commands are simpler/easier to create.  I'll be updating the man page with the old commands further diminished in the doc, but I wanted to create a complete doc on this release.

The big difference is that there is a finer grain of settings, but the settings are simpler -- the best of both worlds.  Many decodings can be done now with the EQ being as simple as '--fcm=--' or '--fcm=--CD' as the typically most complex normal decode.  There is this new 'k' thing also, but that is a secondary level of precsion, where the exact shape of the eq can be changed, and it would be no extra 'k' characters, one or two.  Three 'k' characters brings the decoder back to the original/legacy mode, even then, most of the time, no 'k' characters are really needed.  These are all about 'fine grained' perfection.

The decoder is better yet, and handles more cases, more perfectly.

* Even the start-up message has been polished :-).

 

John

 

[John Dyson]

This is intended as an honest initial encouragment about the new releases V1.4.5F and V1.4.5J.  Both versions act similar, but eventually you'll want V1.4.5J.  I can make demos of any one of these within minutes, but I don't think that it is needed.  The decoder is *REALLY* easy to use now.   This list is simply cut and paste from a manually maintained test table -- these are NOT my official entries yet, but trying to encourage that the complicated manual doesn't mean that decoding is tricky.  (Still can be tricky to make choices only because of hearing issues, but not too bad now.)

 

Quote

Initial/tested/verified decoding results for the V1.4.5F and todays V1.4.5J,
both decode the same, V1.4.5J will produce prettier results.  Note the very good
consistency and the small number of modifiers in the FA initiator (like --fbm=--)

I'll produce formal results for the OFFICAL table later today or tomorrow.

Lots of stuff to do today, including the 'J' release, but it is very safe to use the 'F' release

until I do all of the logistics for the official Windows 'J' version.

 

Each one of the below is as clean/more clean than any thing that I have demoed
before ...

Anne Murray: --fbm=--, --tone=-13.50
Petula Clark: --fba=--BCD --tone=-13.50  (--fba=--BCD is half strength of --fbm=--)
Carpenters, 1970: --fbm=--, --tone=-13.50
Carpenters, 1971: --fbm=--, --tone=-13.50
Carpenters, 1972: --fcm=--X, --tone=-13.50 (A little too much sibilance on Top Of the World.)
Carpenters, 1973: --fcm=--, --tone=-13.50
Carpenters, 1975: --fcm=--, --tone=-13.50
Carpenters, 1976: --fcm=--, --tone=-13.50
Styx, 1977, --fbm=--, --tone=-13.50
Styx, 1980, --fbm==--X, --tone=-13.475
Supertramp, Crisis: --fbm=--, --tone=-13.50
Supertramp, Crime: --fbm=--, --tone=-13.50
Supertramp, Breakfast: --fbm=--, --tone=-13.50
Supertramp, Quiet: --fbm=--, --tone=-13.50
Carly Simon, Greatest Hits: --fbm=--, --tone=-13.50
Linda Ronstadt, 1977: --fam=--CDE, --tone=-13.50 (Fairly light EQ needed, X doesn't work in '--fam')
Linda Ronstadt, Greatest Hits: --fam=--CDE, --tone=-13.50 (light EQ, X doesn't work in '--fam')
The Cars, 2002 Complete Gtst Hits: --fcm=--, --tone=-13.50
London Philharmonic, 2011 50 Gtst Pieces Classical: --fam=--classical, --tone=-13.50
Judy Collins, Fifth Album: --fcm=--, --tone=-13.50
ABBA (All orig albums): --fcm=--, --tone=-13.50, --wof=0.91
Phil Collins 1985: --fcm=--, --tone=-13.50
Dionne Warwick, Walk On By 1968: --fbm=--, --tone=-13.00
Carly Rae Jepsen, Kiss (2012 Japanese Deluxe), --fbm=--X, --tone=-13.50

 

 

 

[John Dyson]

We are going to restart & add-to the decode list.  The decoding choices are now VERY simple, and the list might be as much an indicator of the very large number of FA encoded materials as providng the actual decoding parameters.   We are down to two prominent modes, a choice of the classical encoding or not, and a choice of the stereo image manipulation.  I often add tweaks that provide minor improvements or zeroing in, but superior results can be achieved after choosing from a very small set of choices.   Imagine that it used to require perhaps 20-30 parameters AT LEAST, each one with several possible values!!!   The amazing thing is that these fewer parameters result in decodes MORE ACCURATE than when using dozens of parameters...

There HAS been a delay, but over the next weeks I'll be contributing some much better examples.   The Beatles results are showing the ability to reach back across multiple generations of damage (it is an artifact of the way that DA works.)  The damage mostly happens once, then subsequent cycles don't make it much worse.  It is so easy to hear the little details in the vocals, and the crickets in one recording sound real.

 

John

 

[John Dyson]

This is a note for people who might be trying their own decoding.   I have noticed either a change in my perception or a change in the behavior of the decoder.   You MIGHT want to try adding the 'm' submode, like --fbm=--mk or --fcm=--m.   I have noticed that the special LF EQ isn't always correct.  On some recordings, it smooths out the lower MF distortion and sometimes not.  The difference isn't major, but if you are being very precise, there can be a difference.

 

During this several days of 'playing with decoding', I have been testing a LOT, and had noticed this 'm' thing a few days ago, and finally decided to mention it.  (like all things when listening, one needs to be careful -- hearing does change!!!)

 

John

 

[John Dyson]

The decoding list really needs to be brought up to date, and I will be focusing on it about 1/2 or more of my time for at least several days.  One thing that worries me, especially after this two layer decode that I tried to do -- my judgement is NOT always the best.  When I add to the list, there will be a 'quality' entry of sorts.  This entry will necessarily be MY opinion, and based upon the improvement over the original.

 

Normally, I don't listen to the 'original' CDs, because they usually sound bad to me.  However, I have to be more critical on the decodes for the 'quality' entry to be meaningful.   This means that the nice, useful --skip command on the decoder will help doing A/B during realtime listening and decoding.  If the decode result is not as good as the original, then the quality will be like a '0' or '1'.  If the result is better than the original, then maybe a '3'.  If it is a lot better than the original, then the quality level will be '5'.

 

There is another kind of quality that might need to be added later, where other people who try might give feedback?  The other thing about when I do decodes -- I need to log every one, and usually that requires paper (YUCK!!).  I'll be updating my real-time play facility to log my runs, and then optionally create a log into a master text file.  Keeping all of this information together is going to be a bit of a challenge!!!

 

John

 

[John Dyson]

I don't want interest to disappear -- I have been very busy, and have been working on the list - but being interrupted.  I am definitely a major slow down on this matter, but keep needing to do weird little upgrades (last one was 352.8/384k.)  Every darned time I open up the code, then something else calls for help.   I am hoping to start again on my additions to the list very soon!!!

 

John