The No.1 Website for Pro Audio
 All  This Thread  Reviews  Gear Database  Gear for sale     Latest  Trending
About DolbyA, and why it seems to lurk in the consumer realm
Old 26th May 2019
  #31
Gear Head
 

Quote:
Originally Posted by Thomas W. Bethe View Post
I know in the "good old days" certain Dolby A decoding units sounded "different" I guess changes in manufacturing, aging components OR Dolby A encoders that were not properly set up when the original encoding was done. I personally like the way Satin handles Dolby A decoding and also how it handles DBX decoding.

We sold two of our Dolby 361 units to a recording studio, they were going to use the as vocal processors. I guess they liked the Dolby A encoded signal.

FWIW

If another DolbyA decoder is sufficent -- that is okay, but the DHNRDS actually does work and sound like a DolbyA for decoding. If you are curious how good the DHNRDS sounds -- listen to the demo at the bottom of this message.

On any alleged DolbyA decoder, I'd suggest running level vs frequency curves, and check to see if there is really 4 different bands of operation (really, kind of 3.5), and verify that there is really 10dB of gain control/NR up to about 9kHz and 15dB of gain control perhaps at the top of the spectrum. Precisely matching the attack/release is also critical -- especially when controlling 4 (or 3.5) bands. They have to match reasonably well, or all kinds of grinding distortion can appear.

When the DHNRDS sounds noticeably different from a real DolbyA, the diifference is less modulation distortion (velvet sound -- more clean, more pristine sound), and has a bit more bass. Also, the DHNRDS handles transients much more quickly (none of the softening of a DolbyA.)
* A side benefit for those working with decoding material onto digital -- it seems like the DHNRDS tends to produce fewer 'extra', 'errant' peaks in the output. For exactly the same characteristics/sound otherwise, there seem to be fewer of the 'surprise peaks' in the output. It is probably due to the constant delay filters instead of using IIR/pseudo minimum phase filters. I have sometimes found about 0.5 to 1.0dB extra loudness with the same dynamics -- but not all of the time.
For the typical difference in sound between the DHNRDS vs DolbyA -- listen to the 'Im Not the One' demo. It is VERY difficult to do a decoder that matches all of the attack/release curves for each band and still not produce as much modulation distortion. I thiink that there is a little bit of left over hiss in both the DolbyA and DHNRDS examples, but not sure -- just had to use some aspirin -- I am hearing hiss all of the timeanyway :-).

The decoding results of the DHNRDS are pretty darned good esp when a full, somewhat accurate decoder hasn't been done before, and addtionally mitigates the IMD. (The benefit of extra bass on the DHNRDS IS indeed questionable, but I have also heard other complaints about the DolbyA low end.) So, I gave the DHNRDS about 1dB more, but also rolled off below 20Hz a bit more aggressively (I carefully maitained the LF transient response nonetheless.)

The DolbyA version came from a professionally and correctly mastered 192k/24bit type release, and the DHNRDS versioni was decoded from a leaked DolbyA CD. I was happy to hear the correctly done audiophile release (the DolbyA version), but the DHNRDS gives it a run for its money.

DolbyA version:
https://www.dropbox.com/s/4kd1vfi9rk...lbyA.flac?dl=0

Newer DHNRDS version (no EQ, better HF1 curve than previous example which was deleted 9.0G):
https://www.dropbox.com/s/wko7h3jm67...9.0H.flac?dl=0

(Getting ready for a release -- I found that the 9-20kHz band was about 0.5dB too hot, this decode has no EQ, not needed.)

(IMO, the biggest difference is a bit more bass on the DHNRDS, and the relative lack of fuzz associated with the synth on the DHNRDS version.) Also, (to maintain integrity in the demo) I had to drop the 9kHz on up by 1.5dB, because the DHNRDS has much faster transient behavior and passes HF details (transients) a bit more intensily than a real DolbyA. Tests show that DHNRDS generlaly sounds MORE similar to the original material than a DolbyA. I made absolutely NO other modifications to the DHNRDS decoded version. The mastered material might have had a bit of EQ, or my source CD might have been modified also -- so this is NOT a controlled comparison, but does show a kind of near equivalence.


John

Last edited by John Dyson; 27th May 2019 at 07:42 AM.. Reason: Example was too long -- corrected
Old 27th May 2019
  #32
Lives for gear
 

Verified Member
One problem with using commercially available material is that you don't know what other processes have also been used in the mastering. You also don't know whether they used the original mixdown or whether they've used some kind of processed production master tape - there can be many different versions of the same recording in a label's archive. Do you have any comparisons between your software and properly working hardware that are guaranteed to use the same source audio?

I realise that Richard Hess is involved so I assume that he has access to such material. I must admit that I'd also be interested in trying your software on some of the recent transfers that I've done.
Old 6th June 2019
  #33
Gear Head
 

Quote:
Originally Posted by jamesp View Post
One problem with using commercially available material is that you don't know what other processes have also been used in the mastering. You also don't know whether they used the original mixdown or whether they've used some kind of processed production master tape - there can be many different versions of the same recording in a label's archive. Do you have any comparisons between your software and properly working hardware that are guaranteed to use the same source audio?

I realise that Richard Hess is involved so I assume that he has access to such material. I must admit that I'd also be interested in trying your software on some of the recent transfers that I've done.
I one zillion percent agree with you about the troubles trying to guess how to decode the undecoded 'released' material (also without 'tones'.) There has been success 'improving' material that had been processed but not decoded first. Decoding modified material isn't formally a 'decoding' operation. I have seen both EQ and compression in 'leaked' material, however there are times where 'pretty good' or 'outstanding' results can be retrieved. The EQ is usually [email protected]/Q=0.707 or [email protected]/Q=0.707. Sometimes, there is also a 9k EQ -- not very often. Very interestingly, sometimes compression doesn't seem to destroy the DHNRDS ability to 'improve' (not formally 'decode') 'leaked' materal.

Normally, I don't think that a professional would trust decoding without tones or be a well maintained source recording. I have a dual interest -- trying to find 'diamonds in the rough' for my personal interest purposes, and also help the recording community retrieve the older (but still nice) material. If I run into die-hard hobbiests, I am very happy to make a timeout verson available to them, and give them a helping hand to get started. The DHNRDS really does help 'diamonds in the rough' from time to time.

The DHNRDS is not really meant for consumer use -- it is not consumer-friendly -- mostly intended for a single decoding operation from DolbyA encoded digital to digital or perhaps mostly superior DolbyA decoding from original analog material. This is one reason why I am happy to communicate with recording professionals from time to time -- I want to *learn* what they (the professionals) are interested in.

Richard is my project partner, and we have another individual participating for the Telcom C4 effort -- but I am doing the programming. Apparently, considering the feedback from users, the Satin DBX decoders are pretty good -- I might not bother doing DHNRDS DBX versions. The DBX decoding is relatively simple, the key for DBX being the correct detection algoirthm - even though they use the term 'RMS', the DBX version of 'RMS' isn't generally what a typical EE would think about. Instead of doing a time average of the square of the audio signal, the DBX detection technique does a time average of the log (similar to the level in dB) of the signal. Not realizing that difference in the definition of 'RMS" could cause some serious time-wasting trying to make a SW DBX decoder work/sound correctly.

The DHNRDS DA is going to be available soon again -- there is a new version coming (I corrected a few serious bugs in the code, also making some rather astouding sound quality improvements). The big impediment is right now, here are a few distractions.

My biggest development problem is testing -- I do have some EXCELLENT test material, but as you know as a recording professional -- audio can be very different from recording to recording, and there is no 'well defined' DolbyA specification to work from. SO, to keep from making silly mistakes, I have to do lots of testing. Sometimes, I make test recordings available, and I REALLY DO listen to criticism. Sometimes, it knock my ego down a bit, but I realize that I normally make mistakes -- no since trying to hide that fact :-).
*
Not to worry -- the design isn't being done blindly, lots of resources have been used to be functionally equivalent, except some of the translations are very non-trivial. For example, dynamic attack/release times and also concurrent conversion from feedback to feedforward, while dealing with uncomfortable conversions from analog to digital filtering.
I don't know if Richard is waiting for the C4 code or will be making DA versions available before C4. I do know that demo versions are sometimes available (I wrote the licensing software also), and we can do timeout demo versions or fully licensed versions. The price will NOT be expensive. But, when the new release is actually ready, I am sure that Richard would make demo versions available to seriously interested parties. (In an emergency, I can make a timeout version available, but Richard is the project member who knows the audio world -- I dont know anything about the audio world.)

Personally -- I only expect that the product will have only 10-100 units distributed, but those who need the software, will really like the results and use it a lot. The customer will NOT usually like the command line interface (Richard is working on some mitigation of that problem.) I use the decoder for realtime decodes on command lines all of the time -- so it IS usable.

John
Old 27th June 2019
  #34
Gear Addict
 

Hello John,

The only way for a restoration engineer to properly audition your SW Dolby A decoder is for you to gather some high quality recordings, encode them with a well calibrated HW Dolby A unit, make those recordings, original and encoded, available as test material so the potential users of your SW decoder can compare the sound between HW and SW decoders. By the way, did you ever run some Dolby A encoded pink noise, at various levels, through your SW decoder? Did the decoded results stay in pink? This is a rather difficult test for any compander.

SW Dolby decoder is an intriguing idea. But most restoration engineers I know have HW Dolby A units, as well as other noise reduction systems, such as DBX I and II units. They can’t be in this business without them. I have all of those myself as I occasionally do some transfer work for my clients.

HW Dolby A encoder is a real-time apparatus. But since playing back tape is also a real-time event so one will not save time using a SW decoder. Actually, it will take more time to use SW decoder because you can’t do the decoding during ingestion.

If you can provide some materials as I suggested above, you have my ears.




All the best,

Da-Hong Seetoo
Old 28th June 2019
  #35
Gear Addict
 
jaddie's Avatar
 

Quote:
Originally Posted by dseetoo View Post
Hello John,

The only way for a restoration engineer to properly audition your SW Dolby A decoder is for you to gather some high quality recordings, encode them with a well calibrated HW Dolby A unit, make those recordings, original and encoded, available as test material so the potential users of your SW decoder can compare the sound between HW and SW decoders. By the way, did you ever run some Dolby A encoded pink noise, at various levels, through your SW decoder? Did the decoded results stay in pink? This is a rather difficult test for any compander.

SW Dolby decoder is an intriguing idea. But most restoration engineers I know have HW Dolby A units, as well as other noise reduction systems, such as DBX I and II units. They can’t be in this business without them. I have all of those myself as I occasionally do some transfer work for my clients.

HW Dolby A encoder is a real-time apparatus. But since playing back tape is also a real-time event so one will not save time using a SW decoder. Actually, it will take more time to use SW decoder because you can’t do the decoding during ingestion.

If you can provide some materials as I suggested above, you have my ears.




All the best,

Da-Hong Seetoo
This is exactly what I was thinking too. What we read is mostly about developing the SW decoder based on published data about Dolby A and auditioning unverified material, and highly sighted auditioning at that (non-blind). That's a great approach, but there is plenty of actual Dolby hardware available out there, even some right here that may volunteer to provide verified encoded material, even with tones, sweeps, pink noise, tone bursts, etc. There also exists the opportunity to verify things like the supposed Dolby-induced distortion.
Old 29th June 2019
  #36
Gear Head
 

Quote:
Originally Posted by jaddie View Post
This is exactly what I was thinking too. What we read is mostly about developing the SW decoder based on published data about Dolby A and auditioning unverified material, and highly sighted auditioning at that (non-blind). That's a great approach, but there is plenty of actual Dolby hardware available out there, even some right here that may volunteer to provide verified encoded material, even with tones, sweeps, pink noise, tone bursts, etc. There also exists the opportunity to verify things like the supposed Dolby-induced distortion.
There has been a LOT of ongoing testing with not only 'easy' material to decode, but some of the most difficult material available.

Without doing a careful A/B, it is a bit of a challenge to hear the difference now between a DHNRDS and a true DolbyA. When there is a difference, the DHNRDS is usually more clean sounding (on difficult material.) Transients are more clealy maintained (relative to the original material.) There *are* defects -- as there are defects in both the HW and SW.

I am not claiming perfection -- but the DolbyA HW isn't perfect either (at least, in the decoding phase.) Theoretically, the DolbyA HW encoding is pretty good - it basically defines what is needed for decoding. The decoding is where the troubles exist -- the feedback scheme does seem a good thing -- but there is a delay (and other problems) which manifest like all feedback schemes, causing the 'unencoding' to be not quite as perfect as intuition might inform. (The encoding mostly has modulation distortion -- which can be tolerated at the level that occurs -- but the HW decoding scheme just cannot undo the encoding process completely.)
So, the DHNRDS is an alternative, computer/digital domain, often superior (probably sometimes not quite perfect) alternative. It gives the convenience of staying in the computer world for digitized archives.
I have one recording 'Dreamworld' from ABBA which is one of my bemchnark tests -- it is encoded at least twice in sequence without decoding -- and a real challenge. The DHNRDS does a super-plausible job on it... If it didn't work pretty darned closely to an actual DolbyA -- it would definitely sound worse, pump, surge, etc -- but it doesnt.
If the testing/proof was done ONLY on generally 'easy' material, it would porobably suck on most material -- but the testing is being done on the WORST material, and then verified on very high quality jazz and classical material.
Test tones -- does pretty well on those also -- but the dynamics are the hard part, the static frequency response is difficult to achieve, but not the hard part on sounding plausible. I am NOT claiming perfection in the static behavior either -- but no decoder is perfect. I do fret over 0.25 dB errors... And, there are some places where the errors are greater (esp at LF, where it is about 0.5dB or so too hot in general.)
Again -- the static frequency response issues are NOT the hard part. Matching the dynamics is mind-numbingly detailed and DOES require careful nonlinear dynamics (not fixed R/C) on top of the normal exponential attack/release. Tricky thing to accurately emulate -- DBX would be much easier.


John
Old 29th June 2019
  #37
Gear Head
 

Quote:
Originally Posted by dseetoo View Post
Hello John,

The only way for a restoration engineer to properly audition your SW Dolby A decoder is for you to gather some high quality recordings, encode them with a well calibrated HW Dolby A unit, make those recordings, original and encoded, available as test material so the potential users of your SW decoder can compare the sound between HW and SW decoders. By the way, did you ever run some Dolby A encoded pink noise, at various levels, through your SW decoder? Did the decoded results stay in pink? This is a rather difficult test for any compander.

SW Dolby decoder is an intriguing idea. But most restoration engineers I know have HW Dolby A units, as well as other noise reduction systems, such as DBX I and II units. They can’t be in this business without them. I have all of those myself as I occasionally do some transfer work for my clients.

HW Dolby A encoder is a real-time apparatus. But since playing back tape is also a real-time event so one will not save time using a SW decoder. Actually, it will take more time to use SW decoder because you can’t do the decoding during ingestion.

If you can provide some materials as I suggested above, you have my ears.




All the best,

Da-Hong Seetoo
Regarding the 'noise' test -- one of my tests is stepped level noise (-6dB to approx -45dB) -- where the decoding (spectrum) results are essentially the same as a DolbyA (there are differences -- on the generally low sub-dB range.) One obvious difference is the rolloff towards 40kHz on the HW, and the DHNRDS is brickwall at about 41kHz. On a tone based test (fundamentally different than a noise test), there are approx +-0.25dB errors, except at the lower frequencies where it is about +0.4dB at 20-40Hz.) (fixed level -- about -15dB.)

For 'easy' material -- the decoding results are similar -- one might prefer the HW or the SW -- depending on taste/decoding characteristics, etc. For really 'tough' material -- which admitedly appears mostly on pop recordings from what I have seen -- the DHNRDS seems to provide some pretty impressive results (remember -- I am biased because of authorship -- but, please don't take my opinion as totally untrue.)

The noise tests visually look very similar -- I haven't recently done an audible difference for pink noise, but I am sure that there would be some difference, even though the noise source stepped frequency response comes very close to a cat22 style DolbyA..

About playing back a tape -- one situation where the DHNRDS is obviously superior: digitized online archives -- no realtime limitation. Looking at the LOC standards for tape archiving onto digital, they demure on the matter of NR decoding. In the relentlessly high quality mode (extreme avoidance of the LF IMD products from the HF bands), the decoder does slow to realtime. That kind of decoding performance is not normally needed. (for example, ABBA SuperTrouper -- when it is released in truely decoded form tends to have a lot of IMD. This is probably one reason why it is sometimes not DA decoded, like on the high compressed 'TCSR' release.) The IMD is reduced to 'not obvious' levels using the DHNRDS in the highest quality modes... How often are such extreme measures needed? Probably not often.

There are cases where the DHNRDS can actually decode material that DolbyA units seem to have troubles with. I will admit that there are probably some cases where the DHNRDS might have troubles -- but except for botches in experimental versions, I haven't recently run into serious problems.

If someone is seriously professionally interested, I am sure that time-out versions of the decoder can be made available. The license scheme supports both an individual user, and/or a timeout. There is also a weak security local user count limit scheme also. However, the decoder is NOT going to be expensive, and demo time-out versions will almost certainly be available. This project series is NOT intended on being some way to sell junk software. The program is NOT perfect, but is pretty darned good (in my not so humble opinion.) Professionals really interested -- check out the WWW site (very primitive), and I promise you that the price (for a non-timeout version) is not going to break anyone's bank. Frankly, the price is a nuisance fee..

Those who are curious and know what a DolbyA does, and where it needs to be used -- just *might* be pleasantly surprised where old recordings might just sound noticeably better (and actually be noise reduced, not just EQed or approximately re-expanded.)

John
Old 30th June 2019
  #38
Gear Addict
 
jaddie's Avatar
 

Quote:
Originally Posted by John Dyson View Post
There has been a LOT of ongoing testing with not only 'easy' material to decode, but some of the most difficult material available.

Without doing a careful A/B, it is a bit of a challenge to hear the difference now between a DHNRDS and a true DolbyA. When there is a difference, the DHNRDS is usually more clean sounding (on difficult material.) Transients are more clealy maintained (relative to the original material.) There *are* defects -- as there are defects in both the HW and SW.
Sounds like quite a project.

Something to consider, though. It sounds like you are developing your decoder based on subjective testing using material presumed to be encoded, but without an known reference levels. That's a LOT of assumptions!

Would it not make more sense to invest in a bunch of actual Dolby A cards and a card frame? Perhaps a Cat. 35 card tester? Sure, all of this stuff is old now, and much of it may be in need of repair, but wouldn't you want to start with a couple of "reference" encoders to characterize?

Then, once you have several known good Dolby A cards, encode some fresh material that you have total control of, including reference Dolby Tone level, mix in a bit of tape noise, and create real, known, controlled source material for you to audition?

Would it also not make sense to record some step-tones, or tone bursts of various duration and frequency to test the precision of your decoder?

Since all Dolby cards encode and decode, you'll also have hardware reference decoding to compare to.

Otherwise, you're tweaking like crazy based on subjective supposition of if or how the material was encoded. Like that double-encoded stuff...can't imagine any decoding tracking that well without being very, very custom.

Dolby hardware is really cheap now. A couple of 361 frames, or a 363 frame, including a pair of Cat 22 cards would be a very minimal investment in the development process. Even if you had to buy half a dozen cards to get a good pair, that still wouldn't break any bank.

Hate to see this much programming talent be misled by working with unknown material, or even material that is judged to be Dolby A encoded when it might not be. Just seems to be flailing.
Quote:
Originally Posted by John Dyson View Post
I am not claiming perfection -- but the DolbyA HW isn't perfect either (at least, in the decoding phase.) Theoretically, the DolbyA HW encoding is pretty good - it basically defines what is needed for decoding. The decoding is where the troubles exist -- the feedback scheme does seem a good thing -- but there is a delay (and other problems) which manifest like all feedback schemes, causing the 'unencoding' to be not quite as perfect as intuition might inform. (The encoding mostly has modulation distortion -- which can be tolerated at the level that occurs -- but the HW decoding scheme just cannot undo the encoding process completely.)
The funny thing about this is that while Dolby A isn't perfect in hardware, and results always did vary a bit, that wasn't really a big deal. Again, it wouldn't be hard to come up with a hardware mean with which to work. But you'd need real hardware first.

As to modulation distortion...that's never really been a problem with Dolby A in practice. But I'm not sure what you mean by modulation distortion either.
Assuming you mean tracking issues, or noise modulation, which isn't a big deal at all. Actual intermodulation distortion is not an issue with Dolby A either, at least, not with respect to analog tape. You might see somewhere around 0.1% encoded>decode, but tape is far, far higher than that. That's assuming conventional IMD testing, or custom dual-tone testing with both tones within one compressor band. Just not a problem. So I have to assume you must be talking about something else.
Quote:
Originally Posted by John Dyson View Post

I have one recording 'Dreamworld' from ABBA which is one of my bemchnark tests -- it is encoded at least twice in sequence without decoding -- and a real challenge. The DHNRDS does a super-plausible job on it... If it didn't work pretty darned closely to an actual DolbyA -- it would definitely sound worse, pump, surge, etc -- but it doesnt.
I have logical issues with this.

1. How do you know what was done on your Dreamworld recording? Documentation, or just subjective listening and assumption?

2. If DHNRDS does a super job decoding a real twice-Dolby A-encoded recording, then it cannot be a replica of actual Dolby A. None of the recorded compressor curves would be a match for a single decoder. But then, you may just be tweaking your decoder until it sounds right, which is hardly enough to verify it's conformity with Dolby A, much less prove anything about what was actually recorded.

3. The drawing of firm conclusions based on unknown test material and unverified post processing. You need a real hardware Dolby A encoder, properly functioning, against which to test.

Dolby made a single-ended NR device that utilized a Cat. 22 card
called a Cat. 43. It was essentially permitting tweaking of individual expander bands for use in noise reduction of material that wasn't encoded. You might find that interesting.

Quote:
Originally Posted by John Dyson View Post
If the testing/proof was done ONLY on generally 'easy' material, it would porobably suck on most material -- but the testing is being done on the WORST material, and then verified on very high quality jazz and classical material.
Test tones -- does pretty well on those also -- but the dynamics are the hard part, the static frequency response is difficult to achieve, but not the hard part on sounding plausible. I am NOT claiming perfection in the static behavior either -- but no decoder is perfect. I do fret over 0.25 dB errors... And, there are some places where the errors are greater (esp at LF, where it is about 0.5dB or so too hot in general.)
Don't fret over .25dB. The hardware that encoded the recordings wasn't that good. The hardware spec is +/- 1dB 30Hz-20kHz. And no tape recorder on this planet is even that good.
Quote:
Originally Posted by John Dyson View Post
Again -- the static frequency response issues are NOT the hard part. Matching the dynamics is mind-numbingly detailed and DOES require careful nonlinear dynamics (not fixed R/C) on top of the normal exponential attack/release. Tricky thing to accurately emulate --
This is confusing. The above makes it sound like you do have Dolby hardware, but all other descriptions make it sound like the development project is based on listening to rogue material only. So...which is it?
Quote:
Originally Posted by John Dyson View Post
DBX would be much easier.
Yup. I've already done it adequately with available tools in any DAW. No need to do a command-line dbx decoder. Now, an actual plugin with GUI? That might be handy.
Old 30th June 2019
  #39
Gear Head
 

(Answering comments about testing and comparison with DolbyA results -- I am very primitive in using Forum tools -- so I am not using quotes.)

I didn't use one single method of testing, but MANY methods of testing, with material where we have comparisons with cat22/363/etc deocodes and also open ended tests. Some material had tones, some material didn't have tones. I have a super-wide variation of sources, and rock-solid material to compare with (not just the results of the decoding of in some cases multiple DolbyA units, but the original material before being encoded.)

The effort has NOT been entirely scientific, but there have been well controlled aspects of the testing. For example, I only have about -- maybe -- 5-6 well controlled recordings/albums -- gigabytes with tones and decoding examples. In some cases, also have the pre-encoded source on other material -- where it is just single cuts from an album. My test colleciton isn't just the several recordings above, but also have other less-well controlled material with tones -- still very useful.

Additionally, to widen the test material, I have numerous examples of DolbyA leaked into the consumer world -- such material makes the most aggressive test source, not from a scientific standpoint, but from a subjective result standpoint. Some (pop) recordings with the dynamics in the 6-15kHz range become wonderful sources of IMD -- truly difficult to decode -- per examples as released when they are released actually decoded -- and the often superior results when using a more advanced decoder (using the leaked versions as source material.) I do have a set of subjective criteria that help to determine non-decoded, leaked material -- one is HF compression (difficult), but also stereo depth (very obvious when it is applicable.) Non-decoded material, but EQed, will tend to have an overly flat/weak stereo image. There are a few other spectral measurements that increse the accuracy also. My hit-rate is NOT 100% on the selection of leaked material -- but it definitely happens fairly oftne on 1967 through early 1990's recordings. Such leaked recordings DO make good test material -- from digital distributions, the calibration levels are usually in a narrow region of about +-0.5dB. Some of the consumer-leaked stuff is incredibly challenging to determine the best calibration -- so I just use that for a subjective "did the decoder splat at all?" type question.

Some recordings are NOT a challenge to the DolbyA encoding scheme (mostly orchestral type stuff, or single vocals.) Other (more challenging) recordings have a subtle loss of detail in choral group type recordings, or in the case of pop recordings -- splat city -- for fun, look at the output spectrum after encoding. Do a 20kHz brickwall first, then do an encoding... You might be surprised to see the hash generated above 20kHz, and that hash is NOT limited to above 20kHz -- the hash will exist down into the 6kHz range. The DHNRDS doesn't do hash like that (if the encoded copy is brickwalled at 20kHz -- because unlimited freq response material will have hash before decoding.) I do suspect that the hash comes partially from the two 3k-20+k and 9k-20+k compressors (and then expansion in the feedback loop), where the compressors overlap and compensate in their attack times -- thereby causing a random noise -- a kind of intermod of sorts. The DolbyA attack/release times are not fixed R/C, but very dynamic, and very dependent on the current state of the material -- so any time-based interactions will cause excess modulation effects -- beyond the simple modulation distortion. (Below, I explain an example of the simple modulation distortion causing a loss of detail -- it is NOT severe, but it is a loss of accuracy from the unencoded original source.)

I am NOT interested in criticizing the DolbyA scheme -- rather it is less perfect than might sometimes be realized. (DolbyA had a really important purpose.) Maybe the best test for people who do have before encoding and after encoded/decoded material -- listen to complex choral groups. You'll notice that the encoded/decoded material using DolbyA HW sounds plausible -- but the original un-processed version will tend to sometimes have a LOT more detail. That loss of detail is one example of the IMD created by the DolbyA scheme. The DHNRDS does recover more of the detail -- without sounding bad, when decoding. The detail is NOT lost in the on-tape encoded copy, but rather the HW decoding process does loose the detail.

There are three worlds of results in testing: 1) material w/tones (or at specified calibration levels) that provides test tones/noise bursts/etc. In most of such basic test material I have both original and test decoding results from DolbyA HW of various kinds 2) material wtih tones, commercial recordings(albums), in some cases original material, but in most cases just encoded material & decoding results from DolbyA HW 3) feral DolbyA material distributed to consumers as 'product' -- useful for testing more extreme dynamics.

All kinds of test material have been used -- and it has SOMETIMES been whack a mole to make sure that correcting one misbehavior doesn't cause other misbehaviors. The DHNRDS is not just an open-loop weekend design, but is a reasonably well tested, admittedly imperfect, piece of audio processing software. (It is also ugly to use for some people -- I don't do GUI/Windows. So, its UI is a bit primitive for those not used to command line.) :-). The Windows copy is actually a port of the original Linux version, and final tested on Windows. The Linux version is really SMOOTH for the Linux world - not so much for Windows, but it works well. I am a very honest critic of my own work -- REALLY.


PS: to clarify about decoding 'Dreamworld'... The original is very damaged by multiple encoding cycles (it is obvious once you hear the recording), and the fact that the dynamics are kept sane in two sequential decodes tells me that the tracking for decoding is pretty good. Are the results very good for 'Dreamworld'? I'd say not, but sounds better than the original that is often distributed in the ABBA lovers realm.


John
Old 30th June 2019
  #40
Gear Addict
 
jaddie's Avatar
 

Quote:
Originally Posted by John Dyson View Post

I didn't use one single method of testing, but MANY methods of testing, with material where we have comparisons with cat22/363/etc deocodes and also open ended tests. Some material had tones, some material didn't have tones. I have a super-wide variation of sources, and rock-solid material to compare with (not just the results of the decoding of in some cases multiple DolbyA units, but the original material before being encoded.)
So, unverified material then.
Quote:
Originally Posted by John Dyson View Post
The effort has NOT been entirely scientific, but there have been well controlled aspects of the testing. For example, I only have about -- maybe -- 5-6 well controlled recordings/albums -- gigabytes with tones and decoding examples. In some cases, also have the pre-encoded source on other material -- where it is just single cuts from an album. My test colleciton isn't just the several recordings above, but also have other less-well controlled material with tones -- still very useful.
How are these recordings verified as to Dolby level or encoding passes?
Quote:
Originally Posted by John Dyson View Post
Additionally, to widen the test material, I have numerous examples of DolbyA leaked into the consumer world -- such material makes the most aggressive test source, not from a scientific standpoint, but from a subjective result standpoint.
Again, precisely to my point. You are attempting to "decode" something that you cannot verify has been encoded, at what level, or the calibration of the recording machine. The evaluations are totally subjective, sighted, and biased.
Quote:
Originally Posted by John Dyson View Post
Some (pop) recordings with the dynamics in the 6-15kHz range become wonderful sources of IMD
You cannot quantify IMD with music.
Quote:
Originally Posted by John Dyson View Post
-- truly difficult to decode
You aren't including the rest of the path, the recorder characteristics and tape characteristics are undocumented, levels unknown...on and on...
Quote:
Originally Posted by John Dyson View Post
-- per examples as released when they are released actually decoded -- and the often superior results when using a more advanced decoder (using the leaked versions as source material.)
Again...subjective, assumptive, based on biased listening of undocumented material. It would actually be hard to be more vague.
Quote:
Originally Posted by John Dyson View Post
I do have a set of subjective criteria that help to determine non-decoded, leaked material -- one is HF compression (difficult), but also stereo depth (very obvious when it is applicable.) Non-decoded material, but EQed, will tend to have an overly flat/weak stereo image. There are a few other spectral measurements that increse the accuracy also. My hit-rate is NOT 100% on the selection of leaked material -- but it definitely happens fairly oftne on 1967 through early 1990's recordings. Such leaked recordings DO make good test material -- from digital distributions, the calibration levels are usually in a narrow region of about +-0.5dB. Some of the consumer-leaked stuff is incredibly challenging to determine the best calibration -- so I just use that for a subjective "did the decoder splat at all?" type question.
You really need to get your hands on some Dolby A HW. All of these are subjective and assumptive.
Quote:
Originally Posted by John Dyson View Post
I am NOT interested in criticizing the DolbyA scheme -- rather it is less perfect than might sometimes be realized. (DolbyA had a really important purpose.) Maybe the best test for people who do have before encoding and after encoded/decoded material -- listen to complex choral groups. You'll notice that the encoded/decoded material using DolbyA HW sounds plausible -- but the original un-processed version will tend to sometimes have a LOT more detail. That loss of detail is one example of the IMD created by the DolbyA scheme.
But the actual IMD of the end-to-end system is negligible. Again...you need hands-on here.
Quote:
Originally Posted by John Dyson View Post

There are three worlds of results in testing: 1) material w/tones (or at specified calibration levels) that provides test tones/noise bursts/etc. In most of such basic test material I have both original and test decoding results from DolbyA HW of various kinds
So...only test tones originally and verifiably encoded with cal tones?
Quote:
Originally Posted by John Dyson View Post
2) material wtih tones, commercial recordings(albums), in some cases original material, but in most cases just encoded material & decoding results from DolbyA HW
Again...verifiable? Got recorder performance data? Even EQ specs?
Quote:
Originally Posted by John Dyson View Post
3) feral DolbyA material distributed to consumers as 'product' -- useful for testing more extreme dynamics.
I think it's obvious what I'd say about that one.
Quote:
Originally Posted by John Dyson View Post
All kinds of test material have been used -- and it has SOMETIMES been whack a mole to make sure that correcting one misbehavior doesn't cause other misbehaviors.
The real whack-a-mole mechanism here is the unverified test source.
Quote:
Originally Posted by John Dyson View Post
I am a very honest critic of my own work -- REALLY.
Not being critical of you with this statement, just stating fact. The tests are subjective and biased.
Quote:
Originally Posted by John Dyson View Post
PS: to clarify about decoding 'Dreamworld'... The original is very damaged by multiple encoding cycles (it is obvious once you hear the recording), and the fact that the dynamics are kept sane in two sequential decodes tells me that the tracking for decoding is pretty good. Are the results very good for 'Dreamworld'? I'd say not, but sounds better than the original that is often distributed in the ABBA lovers realm.
How do you know that all wasn't intentional?
Old 30th June 2019
  #41
Gear Head
 

I'll repeat -- I used lots of verfiied, standard DolbyA encoded material for testing,.. I also have a lot of unverified material that is used for testing. I am using EVERY possible source, and providing almost every possible kind of both impairment and high quality material that I can find. Also, have been provided test suite of encoded tones, noise bursts at various levels, along with 363 decoded versions of each.

I am being provided whatever material needed by a moderately well known person in the recording/restoration industry, and the work product is being verified by individuals other than myself.

Sure, SOME material is unverfied, but it also increases the spectrum of test results -- which are generally excellent. Perfect -- no, excellent -- yes.

About Dreamworld -- listen to it -- judge for yourself. It is really a mess, and really good to test with because it is so bad. Are you suggesting that I claimed that the results were perfect? I sure hope not -- but the dynamics are consistent with proper decoding.

I certainly hope that no-one advocates using only pristine, easy-to-process material for testing sw? Avoiding the use of out-of-spec source material tends to create cherry-picked test results. The DHNRDS test results and the rather good performance (again, NEVER perfect) wouldn't mean anything if only carefully selected, single instrument (or single voice) source material was used. I hunted for the nastiest, plausibly encoded material to add to the test suite. As a software developer/EE of 40yrs experience, I learned early on -- it is best to do aggressive testing to avoid as many embarassments as possible. There is at least some testing beyond my own -- it is NOT rubber stamped.

BTW -- the development delays are all due to my own very critical testing -- otherwise, the decoder would have been availble 3 months ago. I am not tolerating ANY serious problems that I can detect -- and I am using the worst/most difficult material in addition to pristine commercial material with tones.

Is it perfect? No Is the DA decoder pretty good? yes, really...
Would the decoder be any good if the only critieria was perfectly encoded symphonic pieces with moderate dynamics? No.
Does using a very diverse set of source material help to achieve the best possible quality? Yes
Are there BOTH subjective and objective measurements in play? Yes
Did the development depend solely on subjective measurements? Absolutely Not.

Is the software decoder (I mean, REAL, full functioning decoder, with full noise reduction) a simple piece of software? No
Is the software decoder commercially viable? Well, only if the software development time isn't considered. A complete, fully noise reducing decoder would NOT be done commercially -- because there is NO real market, except for archives and just a few professional recording engineers who need the BEST POSSIBLE (but not perfect) decoding performance.

Probably most important: is there much ad-hoc coding/filtering/etc in the DHNRDS DA decoder? NO. There are some cases (which is a reason for the small amount of tuning), where modeling diodes at the correct Is level (competent EEs who know semiconductors would know what I mean), and also the choice of where dynamics filters and the needed delays required to emulate the decoding feedback design as ideally as possible. Almost all timeconstants are based on a few critical values -- and everything is based upon those. There are very few individual time constants or ANY ad-hoc delays needed. The most difficult is determining which factors change based upon the state of the compressor gains. It is NOT simple stuff, and being able to design a FET feedback compressor does not qualify an individual to develop a DA decoder like the DHNRDS. Any talk about the concept of an R/C time delay for attack release (like the Sony patent does) doesn't really represent the reality of what R Dolby did back in the early 1960s'.

I added a decoding example snippet -- kind of random out of 1000 or so test cases -- very reasonable source -- no EQ, nothing like that. These are NOT cherry-picked. Even the 'singles' collection release has as many or more artifacts in Karen's vocal sibilance. Sorry that the material that I have that has tones is proprietary and I cannot give out examples. When comparing (if you do) PLEASE only use sources that sound like the vinyl -- do not try to use digital releases, they are often passed through without decoding... That example is a feral DolbyA encoded release... Luckily, the copy that I have still has the levels intact -- one of my recording engineer friends had a copy that had been normalized.

EDIT: The 'decoding' of the feral copy of Sing was a little bit in error. This is one major problem when trying to deal with feral material -- no calibration tones. I only use such material for dynamics testing or my own listening -- and I truly did NOT cherry pick the example. I re-decoded the 'Sing' example with a better calibration level (and a slightly newer version of the decoder -- probably minimal difference because of that.) So -- just as a matter of integrity, I have added a slightly better decoded copy of the 'Sing' snippet -- not removing the original decoding example. IMO, the major difference is the ambiance is more natural -- I do NOT recommend trying to decode 'feral' material unless really necessary. It is the best when an entire album has been played out from the master -- with maybe a small amount of EQ that can be undone.


John

Last edited by John Dyson; 1st July 2019 at 12:53 PM.. Reason: Feral file calibration error.
Old 30th June 2019
  #42
Gear Head
 

Example of IMD mitigation.

I found a simple example of the modulation distortion from DolbyA that doesn't sound 'bad'. Most people wouldn't recognize it as distortion, but it is actually the signal being temporally smeared -- causing a loss of apparent details. All of the energy is in the signal, but it is kind of smoothed out.

* Listen to the seperation in the voices. Vocal chorus really benefit from IMD mitigation.

The DHNRDS example has anti-IMD processing engaged. The DolbyA version is from an apparently properly DolbyA decoded release (Polar Music -- don't have the source details online.) The undecoded source material is from Japan Discomate/CDP-106. Note: I really have been collecting source material for many years... Proper research is important to me.

So -- the DolbyA version is a properly decoded commercial copy, with the spendor of IMD in the audio. The DHNRDS version has been decoded at not quite maximum quality. The max-DHNRDS version has the maximum quality tweaks enabled.
Attached Files
Old 1st July 2019
  #43
Gear Guru
 
Thomas W. Bethe's Avatar
 

Verified Member
I, for one, cannot wait to hear this software. Anyone who takes as much care in doing the research must be able to do an outstanding job on writing the software. FWIW
Old 1st July 2019
  #44
Gear Head
 

Quote:
Originally Posted by Thomas W. Bethe View Post
I, for one, cannot wait to hear this software. Anyone who takes as much care in doing the research must be able to do an outstanding job on writing the software. FWIW
Thanks!!! My project partner, Richard, has estimated that it might be released at/near the end of July. As everyone knows -- writing software isn't just 'coding'. Also, I deeply regret my lack of expertise developing GUI software - so, it is command line based. We are trying to cobble together some aids to make it a bit more pleasant for people not accustomed to command line mode. Also, even though we are focusing on the professional who uses Windows, all of the decoder specific development is done on Linux. If there is a demand, we COULD produce a Linux version. (It is all a matter of logistics, not technology or anything like that.)

<Info below provided to show that it isn't really a science project to use -- esp when one gets used to it, and decides the basic usage pattern>
<Richard has written a really nice manual -- so usage details are documented>

As an idea on using the decoder -- it works pretty much like this (assuming not using a pipeline like one would normally do on Linux/Unix):

da-avx --tone=-13.00 --extra --info=2 --inp=infile.wav --outp=outfile.wav

--tone is the calibration level (can be read from a measurement given by the program, and/or Richard has developed a procedure to determine.)
--extra is one of the advanced modes, enables many of the extra anti-IMD mechanisms
--info=2 is one of the options that provide a running 'pacifier' (could be running gain lists, or just dots, depending)

--inp/--outp are obvious. It supports 48k->96k,176k->192k sample rates, 16bit/24bit/floating-point input, 24bit/floating-point output, bext/rf64/bw64 supported.
(16bit output not supported for obvious reasons -- dither.)
(it works with 44.1k, but not recommended -- no space for modulation sidebands.)
(if using 44.1k sample rate, then the decoder internally runs at 44.1k -- not mathematically a good thing.)

----
I really-truly hope for the DHNRDS DA decoder to do some good.

John
Old 1st July 2019
  #45
Gear Addict
 

Hello John,

I don’t know what your goal is for this SW Dolby decoder. If all you try to do is to make so called “Leaky Dolby material” sound “better”, then what you are doing is perfectly legitimate. Whatever float your boat is fine.

However, if you have any intention to develop a true SW decoder that can replace HW Dolby decoder, you have to be much more honest to what the HW decoder does. All suggested methods here have to be used to double and triple check the results. Again, having a pair of well calibrated and maintained HW units as your reference is a must. Afterall, you have to show that your results can meet that of a HW decoder, from every conceivable angle, under whatever the test condition.

You shouldn’t have to worry too much about frequency extension going out to 40KHz. Tape recorders from those eras didn’t have that extension. If your decoder has to rely on the 40KHz return from tape recorder in order to work correctly, you are asking for trouble. 40KHz frequency response from any analog tape recorder is extremely unreliable.

In theory, SW decoder can work better than that of HW decoder, I sincerely believe that. The original Dolby A 360 series units are almost 50 years old by now, the parts value drift, crossover filter value drifts, the level calibration drifts. Switches and relay contacts get dirty. The mechanical meters used on those units are not even consistent. Power filter caps die. I have four of those units, they were different from each other when I acquired them. I do have the complete service manual and I am able to service those machines, so they perform close to what they supposed to do. By the way, I found the meter on the unit is not very trustworthy. I have mine connected to external level meter for calibration.

When I use Dolby decoder, I have to hope the tape was made with a well calibrated and maintained encoder, the recorder was also in top calibrated shape, the test tones the recording engineer left on the tape were legit and agreed with the documentation that was written on the tape. (Provided they were there at all at the first place) The usage of compander could be so messy, and so inter-dependent it is a miracle that it worked at all. So many things can go wrong in the loop. A SW decoder can cut down a lot of these issues. SW decoder should be a lot more consistent.

I wish you all the best and success in developing your SW decoder. It won’t be easy knowing all the difficulties associated with using Dolby noise reduction systems.



Best regards,

Da-Hong
Old 1st July 2019
  #46
Gear Head
 

Quote:
Originally Posted by dseetoo View Post
Hello John,

I don’t know what your goal is for this SW Dolby decoder. If all you try to do is to make so called “Leaky Dolby material” sound “better”, then what you are doing is perfectly legitimate. Whatever float your boat is fine.

However, if you have any intention to develop a true SW decoder that can replace HW Dolby decoder, you have to be much more honest to what the HW decoder does. All suggested methods here have to be used to double and triple check the results. Again, having a pair of well calibrated and maintained HW units as your reference is a must. Afterall, you have to show that your results can meet that of a HW decoder, from every conceivable angle, under whatever the test condition.

You shouldn’t have to worry too much about frequency extension going out to 40KHz. Tape recorders from those eras didn’t have that extension. If your decoder has to rely on the 40KHz return from tape recorder in order to work correctly, you are asking for trouble. 40KHz frequency response from any analog tape recorder is extremely unreliable.

In theory, SW decoder can work better than that of HW decoder, I sincerely believe that. The original Dolby A 360 series units are almost 50 years old by now, the parts value drift, crossover filter value drifts, the level calibration drifts. Switches and relay contacts get dirty. The mechanical meters used on those units are not even consistent. Power filter caps die. I have four of those units, they were different from each other when I acquired them. I do have the complete service manual and I am able to service those machines, so they perform close to what they supposed to do. By the way, I found the meter on the unit is not very trustworthy. I have mine connected to external level meter for calibration.

When I use Dolby decoder, I have to hope the tape was made with a well calibrated and maintained encoder, the recorder was also in top calibrated shape, the test tones the recording engineer left on the tape were legit and agreed with the documentation that was written on the tape. (Provided they were there at all at the first place) The usage of compander could be so messy, and so inter-dependent it is a miracle that it worked at all. So many things can go wrong in the loop. A SW decoder can cut down a lot of these issues. SW decoder should be a lot more consistent.

I wish you all the best and success in developing your SW decoder. It won’t be easy knowing all the difficulties associated with using Dolby noise reduction systems.



Best regards,

Da-Hong
Most importantly: I cherish well-wishers... Below, I am going to try to explain a few things about the project, how it started (kind of a random walk), and how it became serious. Also randomly discuss some of the challenges. I am bordering on compositionally illiterate -- so bear with my 7th grade writing skills :-).

Second disclaimer before reading the body of the discussion:
Even though I am sometimes a bit overly enthusiastic -- I admit that the decoder is NOT perfect... If I wrote anything rude -- please accept any apologies -- I did try to review the note for any serious transgressions, but I am sometimes clumsy.

Body of the note:
Believe me -- we are going through a huge amount of verification. It will replace DolbyA HW for decoding purposes as well as different DolbyA units do -- perhaps with better quality in SOME regards. I fully expect that there will be times where the current DHNRDS DA decoder doesn't do very well -- I haven't found them yet myself. Whenever there is constructive criticism, it WILL be acted upon. (I have a total of perhaps 500-1000 recordings that are tested, usually running about 200-300 per night.) I don't like to listen to ANY of my test material any more -- the fun is gone :-).

We understand that older DolbyA units might be failing (getting long in the tooth) -- but that isn't really the motivation for the project. The original (3yrs ago) motivation (no longer operative) was to simply do a DolbyA because I had noticed feral DolbyA material in the consumer realm. I had written 'plausible' decoders for the last few years, but not of commercial quality. True commercial quality is what professionals REALLY need. But, considering professional DolbyA users needs, my initial SW 2-3yrs ago, was total cr*p.

The project actually was started when Richard found out about my beginning hobby project, and started providing me with all kinds of nice measurements (gain curves) -- I had previously been using some gain curve estimates that weretn't very accurate -- but were close enough for listening purposes. Trying to infer gain curves from DolbyA schematics that used selected components was not very productive.

Once Richard got involved, and he showed a lot of interest -- then I (as someone with a huge amount of programming and analog EE experience/expertise) started accepting the responsibility that the DHNRDS (didn't have a name at the time) would be a REAL project, that would eventually meet professional needs (to help Richard, who was nice enough to help me.)

As of about 2yrs ago, the DHNRDS effort really started, and it has been long and difficult. First -- reading all of the literature about DolbyA results in more dyseducation than actual learning. There is a lot of conjecture in some documents, other documents are unclear and imprecise. The Sony patent -- for example -- has some good ideas, but has many innaccurate 'factoids' in it -- best to ignore the patent entirely. Also, even though a parametric feedback scheme is 'easy' -- esp as described in the Sony patent, it is also frought with subtle challenges for the DSP/computer realm. I chose to 100% avoid anything in the Sony patent. It was a good thing to effectively ignore. This is because even though the techniques that I ended up using were bordering on impossibly difficult, the results have been lightyears beyond what is possible in the straightforward parameteric feedback scheme (the Sony method is a parameter driven version of the direct audio feedback scheme as used in real DolbyA HW.)

My scheme is a hybrid feedforward & feedback scheme. A straight audio feedback scheme is impossible in the DSP world, but a pure feedforward scheme cannot work esp regarding the level detection/attack&release... So, the DHNRDS uses a hybrid feedback and feedforward scheme. The really cool thing about the DHNRDS approach is that some interesting&useful aspects are fully decoupled from the feedback -- allowing flexibility in modifying the gain control signals. This abiltiy to modify supports both trajectory shaping for the gain control and IMD slicing (difficult to describe, but some concepts are similar to that used in the Orban patents on hard limiter IMD mitgation.) These 'cute tricks' support the IMD mitgation as shown in a previous post.

Patents mentioned:
Sony patent: US 5,907,623
One of Orban's patents: US 6,205,225

About the attack/release and basic emulation. That has been tricky, but untili I had realized that a DolbyA can NOT be emulated by a super-good (artistic :-)) RMS detection scheme and there needs to be some parameters fed-back from the gain control (part of the hybrid feeback/feedforward), the results kept being that of a cute-toy. After a lot of time-wasting pain, the attack/release calculations are now so close that the ambiance on most tricky material is maintained as correct -- if the calibration is set correctly,.

The final breakthroughs started about 6months ago, when the attack/release emulation starting be pretty much exactly what a DolbyA does.... Everything started coming together. All of the previous anti-IMD work and trajectory shaping started being more useful. Even though the DHNRDS sounded like a DolbyA almost 1yr ago, it wasn't 'good enough' until 3months ago. The last breakthrough 1-2mos ago -- which I am still uncomfortable talking about -- has also done wonderful things from an IMD standpoint in the detection phase itself. (There is a natural occurance of IMD in the DolbyA detector design -- most decoding examples of SuperTrouper from ABBA seem to manifest it... That IMD is now *gone* in the DHNRDS.) The front end IMD does wonders even without the other IMD mitigation schemes. (I use ABBA for certain kinds of subjective testing because of the kind of strain that the material places on the attack/release and IMD behavior.)

The last 3months have been final polishing -- I have actually delayed the release because I strongly believe (and I thinking that Richard does also) that the DHNRDS must be as good as can be done -- period. As of the last few weeks, the DA decoder is at that point. Right now, there are some adjustments in the IMD mitigation that create very strong tension with the attack dynamics. (I use certain subjective test material to make sure that percussion is persuasively accurate... hint hint...) The kind of tweaking being done now is an optimization -- an almost 100% exact DolbyA decode has been possible for at least 1-2 months, but I am trying to squeeze out the very last bit of IMD that I can find. Frankly, I might even deviate a little bit from an ideal DolbyA decode -- if it improves the sound in an obvious and universally detectable way. Right now, the code is pretty faithful (I hope :-)).

It has been horrific making the DA decoder work as well as it does (which is really well) without significant objective specifications. (Schematics full of selected components do not count...) The next project -- Telcom C4 has the benefit of an expert on the C4 participating on that phase of the DHNRDS series of decoders.

Summary:

The DHNRDS is NOT a new project just starting to be cobbled together last week :-), I have talked about toy versions of it over the last few years. The current code is capable of decoding DolbyA about as good as can be done using the techniques that are being used. We'll all find out if people who try it, do like it :-).



John
Old 1st July 2019
  #47
Gear Head
 

Mildly off-topic comment, but it is about the DHNRDS...

Anyone who can tolerate/likes ABBA -- some of their recordings are REALLY hidden gems. When properly decoded some are really amazing... I am listening to the Voulez-vous album right now -- the DHNRDS is so very relentless in removing IMD, that all of the temporally (time wise) confusing aspects of the decoding are gone/disappeared...

The ambiance is almost as if a quad matrix (e.g. SQ/QS) is being used -- but not. The temporal details aren't being swamped by IMD. Anyone curious, I could make an arrangement for a demo.

If there were objective specifications -- the task would be much easier!!!

PS: after reading this again -- I wanna make it clear that I wasn't meaning to imply that the sound was somehow 'matrix decoded', but rather the ambiance was recovered very naturally because of the very clean/precise dynamics recovery. I was NOT claiming quad sound per se!!! :-).

John

Last edited by John Dyson; 2nd July 2019 at 07:01 PM.. Reason: Clarification of 'matrix'.
Old 5th July 2019
  #48
Gear Head
 

For funzies -- I am making some examples available here that were referred to in another forum. Basically, was dicussing the poor quality of some of ABBA's albums. I agreed that was probably true, but also offered some opportunity for improvement using the DHNRDS IN SOME CASES. The DolbyA (Polar music) CD example snippets are typical quality -- better than most releases -- also not hyper compressed like some remasters. Essentially, the DolbyA/Polar versions seem somewhat representative of a typical ABBA release. The V0.9.XYN indicates using most-recent DHNRDS pre-release software for decoding. The Polar released versions have 'DolbyA' in the name.

st prefix -- SuperTrouper (time range indicated in name)
Waterloo -- first snippet of Waterloo.
Livingstone example -- seems-to-me unfixable garble in chorus (still problematic)

The examples (I also supply mp3 for quicker/easier review for those just casually interested.)

(st prefix) Super Trouper (song), 0 to 45secs
At the beginning, lots of IMD that upset the HF0/HF1 gains on released version. Mostly resolved by decoding upgrade to DHNRDS. This severe IMD at the beginning was resolved by a improving the input detection -- making IMD from freq above 3k much much less strong.
(st prefix) Super Trouper (song) 63 to 96secs
More clean/distinct vocal chorus
Waterloo (English version) 0 to 45 secs
More distinct vocal chorus
Livingstone (section of garbled chorus, seems unfxiable.)
Example of something that DHNRDS doesn't fix
Above are examples of why I sometimes use ABBA for test material on the project. Also use lots of other source material also (including objective test tones/basis for objective measurement.) I use Brasil'66, Tijuana Brass, Nat King Cole, Burt Bacharach collection, Bread, Carpenters, Linda Ronstadt and sometimes Carly Simon (some of her recordings are gritty some how, even including some of the MFSL disk that I use for reference.) Also, have lots of others that don't really have qualities that are helpful for testing -- the Petula Clark collection in my archives for example. The music/entertainment is great, just haven't found it to have characteristics that are useful for test material.

I was myself surprised about the relative cleanness on SuperTrouper, but also have tried to figure out what might be wrong with the DHNRDS on the Livingstone example. JUST MIGHT be a flaw in the source material? It seems like some ABBA material is very challenging.

Example archive (will probably be maintained for a few weeks):

https://www.dropbox.com/sh/8qlb91db5...ePqH39EWa?dl=0

John
Old 1st August 2019
  #49
Gear Head
 

FYI -- the production version of the DHNRDS DA decoder is *finally* software complete and is starting to trickle SLOWLY out. It sounds really good when used with proper calibration, etc but Its use is not for the faint of heart when there are no calibration tones. When all is right (that is, proper calibration, proper EQ), the results can be astonishingly clean. Vocal chorus are one of the obvious improvements over the admittedly good DolbyA HW.

(I am not a sales/marketing person -- just talking about my technical toy) -- I don't sell anything, but a little search/research or private messaging with me to connect with my project partner might be super beneficial for when super high quality DolbyA decoding is desirable. The software is being sold ONLY for nuisance fee -- it is a 99% altruistic effort. In an emergency I can make a timed-out version until the 'official' version can be distributed again. (Project partner is hyper tied up right now, but I can fill in for emergencies -- I don't accept money though... I Only make 1-2month timeout versions for the interim or sometimes for evaluation.)

I am willing to discuss some of the mathematical techniques -- the math is NOT straightforward like for a normal compressor or expander, and a lot of work was done to mitigate modulation distortions (the results have significantly greater detail.) Lots of stuff like 'Hilbert transforms' and 'Kaiser-Bessel' window functions were used.

Olivia's 'If you love me' example is a bit metallic sounding, but better sounding than any other copy in my collection. The other Olivia example is just nice, normal sounding.

Most important, listen to the SuperTrouper example and try to find ANY examples anywhere nearly that clear -- I doubt anyone but Bjorn or Benny would have one. Listening to some of the ABBA albums fully decoded sound amazingly clear when compared with typical CDs or even vinyl.

These examples were decoded from feral DolbyA sources, but true master tapes are even better. I don't do 'mastering' or even 'recording' for a living, so there are likely mistakes in my results, but NONE of the mistakes are from the decoder.

Sorry about the snippets being truncated -- gotta do that, and I have already stretched the rules a bit too far... Also, the Dropbox player sometimes creates a phasing sound -- a download/local play will clear that up.

https://www.dropbox.com/sh/60ys7ouob...OpPfTd8ha?dl=0
Topic:
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Forum Jump
Forum Jump