Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Possible reason why many old music releases sound harsh... (Read 12079 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Re: Possible reason why many old music releases sound harsh...

Reply #25
Apologies about the version of the program that I produced...  It is missing some of the expansion needed to match the spec.  It still works well, but will be missing about 1/3 of the expansion.  This has to do with the behavior of digital filters/feedback vs. analog.  I am working on an improvement -- but a straightforward fix will produce an oscillator (not what any of us want.)  The first phase fix that I produced seems to improve a copy of ABBA that I have by another 'half step.'   The problem is NOT with the expansion ratio, but rather the expansion depth.

Again, the version that I uploaded makes a big improvement when it works, but doesn't expand to the depth necessary to fully remove the compression.   I screwed up a little - but converting ones thinking from the analog world to the digital world can be a bit tricky (because I am a full analog engineer and about 75% of a full DSP programmer.)  Since the program appears to work pretty well, I kind of fooled myself into believing that it is 100% working -- it is about 60-70% working.

John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #26
You are really over-thinking this.  While I am totally in awe of someone whipping of a software Dolby decoder, it may be time to work smarter, not harder.

First of all, you're trying to create a complimentary system using only information from literature.  Then, you're trying to fine-tune it without ever having heard what encoded NR of any kind sounds like (they are all quite different).  I've asked some of this before, and the reason is to help to determine what you are hearing using more of a forensic approach.

Here are some critical questions:

1. What kind of tape is this recorded on?  1/4 reel to reel, for example?

2. What is the track format...1/2 track, 1/4 track, etc.

3. What size reels are these on?

4. What is the tape speed?  7.5ips, 15ips, etc.

5. You called the recordings "anthologies".  Do you mean something compiled for commercial release, or compiled for some other purpose, like a consumer's own personal use?

If you would supply answers to these, I could suggest the highest probability of what you're hearing, and then you could properly decode it if it is indeed NR.

I could provide you with actual Dolby A encoded material, but before we go there, let's figure the rest of this out.

Re: Possible reason why many old music releases sound harsh...

Reply #27
Approximately, that's exactly the reason why many engineers of my generation had avoided to use it at all costs.
No, that's not correct.  Dolby A tracks itself remarkably well, and was very widely used.  The reason some engineers avoided it was a classic case of sighted expectation bias.  Once you knew what was going on you'd swear you heard all sorts of things, like reduction in ambience, less presence, dulling, brightening, etc.  Dolby A was responsible for none of that.  It is one of the most transparent noise reduction systems ever produced, second only to Dolby SR.  But Dolby gear was expensive, and high output tape combined with using tape saturation as an "effect" negated the need for it to some degree.  Engineers who actually used it frequently recognized its purpose, value, and efficacy. 

Re: Possible reason why many old music releases sound harsh...

Reply #28
Approximately, that's exactly the reason why many engineers of my generation had avoided to use it at all costs.
No, that's not correct.  Dolby A tracks itself remarkably well, and was very widely used.  The reason some engineers avoided it was a classic case of sighted expectation bias.  Once you knew what was going on you'd swear you heard all sorts of things, like reduction in ambience, less presence, dulling, brightening, etc.  Dolby A was responsible for none of that.  It is one of the most transparent noise reduction systems ever produced, second only to Dolby SR.  But Dolby gear was expensive, and high output tape combined with using tape saturation as an "effect" negated the need for it to some degree.  Engineers who actually used it frequently recognized its purpose, value, and efficacy.

It would appear that the above are claims that need to be illustrated with relevant DBTs..

Before and after tracks that have been round-tripped through  Dolby processing without the insertion of an analog tape machine would be the evidence of choice. 

Re: Possible reason why many old music releases sound harsh...

Reply #29
It appears that my processor MIGHT have been doing 1/2 of the expansion (in dB) as it should have been.   Everything that I have read has talked about '10dB' for the amount of compression (including the patent information that I provided to you all.)  However, I just read a setup manual a DolbyA unit, and found that there is 20dB of compression being implied by some of setup numbers.  So, I retuned the unit for 20dB (slightly modification of parameters) and now the expander is doing more.  Initial results are showing that it might be doing ALOT more than what it was doing before.   Again -- not sure, I just made the modification about 1HR ago, and read the set-up manual last night.  Previously, I was only looking at the schematic, and used the idea that there was only 10dB of gain control gonig on.  In fact, the patent information seemed to agree with that idea -- but the setup manual?  Not sure yet.

BTW -- I am not getting any information directly from ANY tape, but simply stuff that I have purchased from HDtracks, and other places over the years etc.   I have always been quite disappointed with the apparent excessive compression that I definitely know that wasn't there before.   Also, I have another datapoint with a Carly Simon download -- the hum/buzz in the first cut obviously shows that there was compression in the release that wasn't there many years ago (I remember that buzz from working with HW designs in the 1980's) I remember that my gain control systems would make that buzz more audible -- just like the release in my possession today. My interest/involvement has nothing to do with any 'tapes' or any commercial interest.  My only interest is in dealing with the poorer-than-desired quality materials (recordings) that are being distributed, and what to do about them.

Unlike most -- I have the real tools to deal with audio processing -- I am not dependent on 'plugins', toys or whatever (in fact, there is an ancient plugin based upon a toy that I was playing with about a decade or two ago -- but it was only a toy compared with what I did and have been doing.)   I am willing to cooperate when everyone is willing.

John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #30
Quote
I could provide you with actual Dolby A encoded material, but before we go there, let's figure the rest of this out.
All I need is a flac file with the appropriate 'dolby tone.'   I can deal with any standard or non-standard flac file (mp3 would work, but the mp3 compression would distort a small amount of the proverbial 'hints' left over from the compression process.)   I guess a good mp3 file or a good opus file would be much better than nothing at this point.  I'd even be happy with a short raw wav or even a raw file with the parameters associated with it (e.g. encoding format -- float,64float,24bit signed/unsigned, sample rate, channels,etc.)  If sox won't handle it, as long as it isn't compressed with an exotic compression scheme I can decode it.

Frankly, I could learn ALOT even without the reference tone!!!  I'll happily publically provide my results whether or not I find myself being a crackpot :-).

John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #31
It appears that my processor MIGHT have been doing 1/2 of the expansion (in dB) as it should have been.  Everything that I have read has talked about '10dB' for the amount of compression (including the patent information that I provided to you all.)  However, I just read a setup manual a DolbyA unit, and found that there is 20dB of compression being implied by some of setup numbers.

 So, I retuned the unit for 20dB (slightly modification of parameters) and now the expander is doing more.  Initial results are showing that it might be doing ALOT more than what it was doing before.  Again -- not sure, I just made the modification about 1HR ago, and read the set-up manual last night.  Previously, I was only looking at the schematic, and used the idea that there was only 10dB of gain control gonig on.  In fact, the patent information seemed to agree with that idea -- but the setup manual?  Not sure yet.

BTW -- I am not getting any information directly from ANY tape, but simply stuff that I have purchased from HDtracks, and other places over the years etc.  I have always been quite disappointed with the apparent excessive compression that I definitely know that wasn't there before. 
But, IIR, you don't own the reference copy anymore, right?  You're depending on a "memory"? 
Also, I have another datapoint with a Carly Simon download -- the hum/buzz in the first cut obviously shows that there was compression in the release that wasn't there many years ago (I remember that buzz from working with HW designs in the 1980's) I remember that my gain control systems would make that buzz more audible -- just like the release in my possession today.
Again, your reference data point is a memory.
My interest/involvement has nothing to do with any 'tapes' or any commercial interest.  My only interest is in dealing with the poorer-than-desired quality materials (recordings) that are being distributed, and what to do about them.
But that's a subjective opinion that flies in the face of industry procedure and standards.  Dolby A would never get out on a release!  Neither would any other encoded NR type.
Unlike most -- I have the real tools to deal with audio processing -- I am not dependent on 'plugins', toys or whatever (in fact, there is an ancient plugin based upon a toy that I was playing with about a decade or two ago -- but it was only a toy compared with what I did and have been doing.)  I am willing to cooperate when everyone is willing.

John Dyson


I'm willing, but we must deal with real data, not memories.  And we must deal with specifics, not ambiguities.  Like state the specific track, for example.  Some of us might just have the actual vinyl!  And then, you'd have a real reference datapoint.

Re: Possible reason why many old music releases sound harsh...

Reply #32
Quote
I could provide you with actual Dolby A encoded material, but before we go there, let's figure the rest of this out.
All I need is a flac file with the appropriate 'dolby tone.'   I can deal with any standard or non-standard flac file (mp3 would work, but the mp3 compression would distort a small amount of the proverbial 'hints' left over from the compression process.)   I guess a good mp3 file or a good opus file would be much better than nothing at this point.  I'd even be happy with a short raw wav or even a raw file with the parameters associated with it (e.g. encoding format -- float,64float,24bit signed/unsigned, sample rate, channels,etc.)  If sox won't handle it, as long as it isn't compressed with an exotic compression scheme I can decode it.

Frankly, I could learn ALOT even without the reference tone!!!  I'll happily publically provide my results whether or not I find myself being a crackpot :-).

John Dyson

To get that you'd need a FLAC file of a studio master with the original Dolby tone on it to establish its level relative to the encoded material.  I can give you a file of Dolby tone, but I have no idea at what level I should record it.  My best guess would be something like -10 relative to 0dBFS-RMS, but that would not at all be guaranteed to be within the 2dB window.

I can also give you Dolby A encoded samples, but to make that meaningful again, I'd have to establish a new arbitrary reference level.  Dolby tone was to be at 185nW/m, the old Ampex reference fluxivity for analog tape.  Some studios used different levels, but that was what Dolby recommended.  We don't have that reference in digits.  My "guess" was based on having 10dB above 185nW/m of clean headroom for tapes like Ampex 456, but that changes with tape speed the individual tolerance for distortion, as tape saturates slowly compared to digital 0dBFS clipping. 

We also need to start with old familiar material confirmed (in your memory) to NOT have any NR encoding on it.  The reality is, that would be anything, but again we're dealing with memories. 

I know my opinion on this may not carry weight...I can assure you, the sample clips uploaded so far do not have Dolby A on them.  In fact, no flavor of Dolby pro NR was ever meant to be listenable without decoding.  The only Dolby type that could be listened to...sort of... with no decoding was Dolby B, and they did that because they knew the greater number of players would have no on-board Dolby processing.  But if you're dealing with HD Tracks, you don't have Dolby B either!  That's because it was only applied to cassetted dub masters, which were then dubbed to bin-loop masters, then high-speed duplicated to actual cassettes. 

You also need to understand, letting non-decoded Dolby A out on a release would be a ridiculous blunder.  All Dolby A tapes had the reference tone at the head, and that tone is distinctive, could not be mistaken for any other set-up tone.  It's 850hz with essentially an FSK "nick" every quarter second or so.  Tapes, both boxes and reels, were tagged with a yellow Dolby A sticker, or in some other way clearly marked.  To think you have a release of any artist as prominent as the Carpenters or Carly Simon with undecoded Dolby A is frankly ludicrous.

If you want to hear some Dolby A, post a clip you are convinced has no NR on it, or any other compression, and I'll give it a shot.  My take a day or so, busy week, but I have the gear to do it. 

Re: Possible reason why many old music releases sound harsh...

Reply #33
Approximately, that's exactly the reason why many engineers of my generation had avoided to use it at all costs.
No, that's not correct.  Dolby A tracks itself remarkably well, and was very widely used.  The reason some engineers avoided it was a classic case of sighted expectation bias.  Once you knew what was going on you'd swear you heard all sorts of things, like reduction in ambience, less presence, dulling, brightening, etc.  Dolby A was responsible for none of that.  It is one of the most transparent noise reduction systems ever produced, second only to Dolby SR.  But Dolby gear was expensive, and high output tape combined with using tape saturation as an "effect" negated the need for it to some degree.  Engineers who actually used it frequently recognized its purpose, value, and efficacy.

It would appear that the above are claims that need to be illustrated with relevant DBTs..

Before and after tracks that have been round-tripped through  Dolby processing without the insertion of an analog tape machine would be the evidence of choice. 


We'll get back to that in a bit...(said the owner of one of David Clark's (and your) ABX systems)...

Re: Possible reason why many old music releases sound harsh...

Reply #34
I am not dealing with memories when seeing the improvement -- my comparisons with the past were ONLY TO SPUR ME ON, not used as the basis of any success/failure today.  I don't think that I have published any results based solely on comparisons with the past, but only to chat about comparison with my memories of the sound of the music in the past.  Any improvement found today exists today and not in comparison with the past...

Additionally -- the main reason for coming back into the group just now wasn't to answer the fact that I am living in the past -- but rather there was very good benefit from the retune to the new parameters based upon the setup information that I read last night.  So, I am using the same curve, but effectively doubled the resulting dBs before converting to linear and doing the gain multiplication.  Also I had a bit of trouble with the HF (9K+) channel, where there appeared to be some instability -- which didn't reach the output due to plenty of brickwall filter.  I found that the instability might have come from the gain filter which emulates the large resistor on the gate of the gain control FET (figure about 5-10pf capacitance with about 330k of resistance.)  I removed the filter, and it seems like some of the restrictions of parameter choice have been lifted...  (A filter means phaseshift, and can cause a loop to go unstable.)

So, it is working better than any decoder than I have previously released, but still need a day or so of listening/testing with this configuration.

John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #35

You also need to understand, letting non-decoded Dolby A out on a release would be a ridiculous blunder.  All Dolby A tapes had the reference tone at the head, and that tone is distinctive, could not be mistaken for any other set-up tone.  It's 850hz with essentially an FSK "nick" every quarter second or so.  Tapes, both boxes and reels, were tagged with a yellow Dolby A sticker, or in some other way clearly marked.  To think you have a release of any artist as prominent as the Carpenters or Carly Simon with undecoded Dolby A is frankly ludicrous.


It seems to have happened.....
http://tapeop.com/interviews/93/brian-kehew-bonus/

Quote
Have you had any crazy technical issues?

We were going through Yes' Tormato tapes. Producer/engineer Eddie Offord had started the album -he had done most of the Yes records and I know from working on his tracks that he used Dolby A a lot. These tapes don't say Dolby A, but Tormato is a famously bad-sounding record. They parted ways with him mid-course and somebody else finished the record. So I'm looking at the tapes and it doesn't say Dolby A anywhere on them — it's typical that they note that when encoded — but I said, "Hold on a second, let me put Dolby on this." And everything — except for some of the later overdubs — sounded amazing. I went, "Aha!" I think we realized what happened. They went to somebody else and the other person didn't see Dolby on the tapes.

Because that record did sound kind of murky...

Thin, flat and terrible. I accidentally discovered the bad secret of it: that it could have sounded a whole lot better.


I can testify that Tormato 's production  has indeed always sounded 'off', going back to its original LP release in 1978, and that hasn't been remedied by numerous remasterings since its first CD release. 

Re: Possible reason why many old music releases sound harsh...

Reply #36

You also need to understand, letting non-decoded Dolby A out on a release would be a ridiculous blunder.  All Dolby A tapes had the reference tone at the head, and that tone is distinctive, could not be mistaken for any other set-up tone.  It's 850hz with essentially an FSK "nick" every quarter second or so.  Tapes, both boxes and reels, were tagged with a yellow Dolby A sticker, or in some other way clearly marked.  To think you have a release of any artist as prominent as the Carpenters or Carly Simon with undecoded Dolby A is frankly ludicrous.


It seems to have happened.....
http://tapeop.com/interviews/93/brian-kehew-bonus/

Quote
Have you had any crazy technical issues?

We were going through Yes' Tormato tapes. Producer/engineer Eddie Offord had started the album -he had done most of the Yes records and I know from working on his tracks that he used Dolby A a lot. These tapes don't say Dolby A, but Tormato is a famously bad-sounding record. They parted ways with him mid-course and somebody else finished the record. So I'm looking at the tapes and it doesn't say Dolby A anywhere on them — it's typical that they note that when encoded — but I said, "Hold on a second, let me put Dolby on this." And everything — except for some of the later overdubs — sounded amazing. I went, "Aha!" I think we realized what happened. They went to somebody else and the other person didn't see Dolby on the tapes.

Because that record did sound kind of murky...

Thin, flat and terrible. I accidentally discovered the bad secret of it: that it could have sounded a whole lot better.


I can testify that Tormato 's production  has indeed always sounded 'off', going back to its original LP release in 1978, and that hasn't been remedied by numerous remasterings since its first CD release. 


There are always a few exceptions.  That doesn't change the norm/average.  I mentioned earlier that it was not unusual for TV to broadcast non-Dolby audio with Dolby A decoding, resulting in pumpy/dull audio.  Mistake, clearly, avoidable if anyone was more conscientious. 

Re: Possible reason why many old music releases sound harsh...

Reply #37
I am not dealing with memories when seeing the improvement -- my comparisons with the past were ONLY TO SPUR ME ON, not used as the basis of any success/failure today.  I don't think that I have published any results based solely on comparisons with the past, but only to chat about comparison with my memories of the sound of the music in the past.  Any improvement found today exists today and not in comparison with the past...
So sorry if I misunderstood this:
I believe that there  the results of decoding sound CLOSER to the original vinyl type sound, but there is a certain warmness to vinyl (they had to use lots of eq/feedback cutting amplifiers  to make vinyl work correctly.)   i don't think that an exact emulation of vinyl sound would be possible (or in some cases, desirable -- click/pop :-)).   However, when using the decoder on appropriate source material, the resulting general balance of the sound is MUCH closer to the original 'Carpenters' sound.    
It's the above quote that gave me the idea you were using the "original vinyl" as your reference.  Perhaps I was mistaken.

My offer to provide actual Dolby A encoded material still stands, you pick the material.

Triva: the original name for the Dolby A301 unit was "S/N Stretcher", which is shown in the upper left corner of the unit.  I'm sure after Ray Dolby got a marketing department together they changed that one.   IIR, Dolby made Time Magazine with his "stretcher".



That thing was a beast!  Mid 1960s analog gear was heavy.  I replaced one of those with a brand-new set of 361A units. 

Re: Possible reason why many old music releases sound harsh...

Reply #38
LINUX versions:
da32-sse3 -- works on most machines made since the middle 200X timeframe.   Might not work on the earliest Atoms. 32bit or 64bit
da32-p4 -- works on practically all machines, but is also the slowest version. 32bit or 64bit
da-avx -- works on most recent CPUS, 64bit only -- the fastest version so far -- often works in realtime.
(when running the 32bit versions on 64bit systems, you have to already have the 32bit shared libs installed also.)
Also, before running the file that you downloaded, you'll need to rename the program (remove the .prog extension), and then add execute permissions, e.g.:  'mv da32-p4.prog da32-p4", then "chmod +x da32-p4".

It would be much better if you could provide source code - or even publish your programm on github or simillar.

Re: Possible reason why many old music releases sound harsh...

Reply #39
GANG -- I created several major bugs when doing the release.  I am DEEPLY EMBARASSED.  AN earlier version worked much better than the released version, but the new version (not yet released) appears to be 'dead nuts on', if not still a little more 'gating' effect than it should have (not much/not ovious -- but it is there.)  It will be improved yet some more before releasing again. NO MORE EMBARRASSMENT.  When I did the FreeBSD VM code, I made lots of mistakes due to 'distractions', but it ended up being the best VM system for many years.  I work on anything that I do until it is best possible -- however, this was a stupid punt and I am not happy with myself.

  Because of some health issues, I haven't been able to address the expander issue very much until now.  Still haven't read any email about this -- but here are some ideas about the stupid bugs:

1) I miscalculated the gain (a major part of the problems.)  This means that it was only doing a little bit of expansion on the higher levels.  It did improve some of the sound, but not sufficient to make it worth while.
2) the output/feedback calculation was dependent on timing -- fixed that 100% so that the gain is being carried around instead of the signal differences.  I had screwed up the phase in the feedback, basically nullifying 70% of it.  It only required a few minor changes in choice of delays to cause that to mess up (that cannot happen anymore in the new code.)
3) my display of gain was screwed up.

THIS IS WHAT I GET FOR not doing a complete test suite and making 'last minute' changes.

Right now, no matter if this is a DolbyA decoder or not -- the version under test is successfully expanding and doing noise reduction (quite a bit in fact), and also doing more than just helping mitigate excess ambiance of the sound.

The noise reduction is now very, very obvious -- and the only real problem in that area is that this new rewrite originally did MORE than DolbyA can do for pushing the gain down.   I had to tune it down so that it basically does (for most music sequences) on each band, between -6dB(about -7dBin) and 16dB(quietest -- below -40dBin) of loss, while the gain increases to about 0dB as the signal increases, but is programmatically limited to 0dB gain.   Most of the time, however, it NOW sits between 6dB and 16dB of loss (down to just below -40dB in relative to threshold) -- and much of the time the 9k range will maximize the loss (when there are no HF.)

I am still EXHAUSTIVELY testing, doing a staged release rather than just putting out something that I have tricked myself that it works.  I am doing both subjective tests (certainly not adequate), and doing tests with various test tones at various levels -- and also doing broadband noise tests to make sure that the expander works properly.  It will be at least two more days to provide something new.

I did have a version working better that the released version (not nearly as good as this version) earlier before sending it out, but I didn't do adequate release/code control -- which is a mistake that I know that I shouldn't have made.   The earlier version was dependent on the delay of the filters, and was doing a normal feedback scenario (where the full signal was being sent back in negative feedback.)   This new version does things much closer (in limited ways) like the patent that I talked about -- but they are missing some other important things.   My method is probably 100x better than the patented method - and actually requires simpler code.   But, the way that I am implementing the expander is that it doesn't feed back the signal, but rather calculates the feedback -- eliminating the phase/time delay issues (which were causing me major headaches.)  (IN THIS CASE, GAIN has less phase/timing criticality than the signal itself -- it is kind of like the color subcarrier on the old TV system -- it was very fragile when residing in the composite signal, but when in demodulated form, was relatively more robust WRT the timing.)

So, now I am truly seeing full -16 to -6dB gain changes (with corrected measurements), while the loudest sections might move up a few more dB in gain increase perhaps up to 0dB gain.  Before, on the version (stupid) that I released, my display was fooling me because I had used the wrong calculation, and about the maximum that version COULD do was about 3-6dB of signal level change -- how did I fool myself?  (earlier versions worked -- and I stupidly made last minute 'improvements'.)
I am doing hours and hours of testing/review/etc now.   Not touching the code for tweaking purposes and being more serious about the version control -- which I already do -- but more serious.

This new version does do proven gain control.
The handling of above 9kHz appears to be better (more complete), but will be tested to at least have reasonable behavior.
Hiss disappears with less background noise in some cases on some materials.  Much better than the sometimes doing NR at all.  I actually had fooled myself into thinking that it was working -- because it did in the past!!!

Sorry, you'all
John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #40
Good news!!!  The 'expander', DolbyA decoder or whatever is working beyond my best imagination now (it means that it simply works -- and does REAL noise reduction and sometimes reasonable removal of compression artifacts.)  IT really does do 10.5dB of expansion in each channel, and appears to be REASONABLY flat.

This version works at 44.1k, 48k, 96k and 192k...  It only needs one pass, because it does everything that I planned and it does everything that it should.  I am distributing the files 'in the open' without using zip -- apparently some systems are worried about executables in zipfiles, so I am uploading the files in-the-clear.  The following files are being made available:  da-sse3 for Intel 64 bit Linux on anything recent, da-win.exe (and cygwin1.dll support library) for Intel 64bit on Windows, dainst.txt (lots of info about what has been done), and the source file for the math used in the decoder (audioproc.cpp.)  At least, if you can read moderately advanced C++, it will describe what is going on.
I have tested, tested, verified, tested (and a few minor mods) in the last 2-3days -- trying to avoid making the mistakes that I made last time.  I DEEPLY APOLOGIZE FOR THE MISTAKES.  I haven't had time to read any messages lately (and afraid of reading about the mistakes that I made...   Basically, it ONLY NEEDS ONE PASS, so the command line is like (you'll have to translate for Windows command line -- I am using cygwin command line on Windows for my own purposes):
da-sse --info --cmd="(l,-1,0)" <infile.wav >outfile.wav
or
da-sse --info --cmd="(l,-1,0)" --inf=infile.wav --outf=outfile.wav
on windows, one would use the command da-win, and need the cygwin1.dll file in the same directory.
REMEMBER, If on Linux, rename the da-sse3.prog to da-sse.  Also, add execute permissions by 'chmod +x da-sse'.
For windows, depending on your environment, you might need to add execute permissions for da-win.exe.

The source code might be interesting to computer/EE techies, but is NOT complete WRT the more elusive support
code.

Good luck, and best wishes, John


Re: Possible reason why many old music releases sound harsh...

Reply #41
Warning about something that works, but only have one of the two choices that work correctly!
Regarding using the da-win command under the command line shell on Windows64 -- you have to use the --inf/--outf switches.  Apparently the stdin/stdout redirection doesn't work as expected.
so the command line this is the only thing that works:,
da-win --cmd="(l,-1,0)" --info --inf=infile.wav --outf=outfile.wav

The 2nd cmd arg (-1) is the threshold, and that is the only thing other than the filenames that really need to be changed.   The use of --info is optional,. but  helpful to know that it is working.

I have previously been using the cygwin shell which handles the redirection as expected.  Apparently, there is an incompatibility with redirection on the norma/standard windows command line shell.

I just tested it with the shell -- I had a thought (panic) that was one thing that I didn't check.  LO AND BEHOLD -- that one thing that I didn't test was a problem!!!

Under the WINDOWS command line, not under cygwin command line -- I just did the following exact command that worked:
da-win --cmd="(l,-1,0)" --info --inf=218-raw.wav --outf=218-out.wav

I am not a windows person -- only have done kenel level programming 15yrs ago, and only a casual user and very minimal programmer on it.  SORRY about the SCREWUP again, but the code does work!!!
The only reason why I am using Windows now is that much of the world lives on windows, and I am REALLY trying to help!!!

John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #42
More good info:
I have two examples (Scarborough Fair -- S&G) -- one decoded, one not decoded.  This is definitely in need of some decoding -- note that when you hear 'sses' and other high freqs, they are there -- but the HISS PART OF THE VEIL IS PRETTY MUCH GONE.  SOME OF THE VEIL ITSELF IS ALSO IMPROVED.   When watching the decoder, it says that it did about 7dB NR at 9kHz+, and about 6dB NR at 3kHZ+ in the long term.   There was enough other content at those frequencies that it couldn't gain-down more than that.  When there were 'sses' or other significant HF material, then the gain went up to approx the same as 80-3k.   There was also significant inter-syllable NR at LF.   It tried to reach down a little at 80-3k in between syllables, but it only got about 1dB of gain dip from time to time.  During the longer piece, it might have dipped a few more dB at 80-3k, but it mostly sit between -1.5dB and -0.80dB.

I normalized the volume of the entire piece before chopping off these 30second snippets.  There is actually more dynamic range in the sense of 'longer term' also.  So there is probably a full 10dB of NR at HF when all is said and done (when considering the processing the entire piece and entire dynamic range), and perhaps 3-6dB of NR at LF (0-80Hz.)   The NR at 80-3k is pretty much nil -- perhaps 2-3dB during quiet times.     THIS IS ACTUALLY WORKING CORRECTLY.

The song itself actually sounds better with just a little less excess ambiance and a little more expression during the entire piece.

John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #43
More good news---
I have noticed that when using the decoder,there had still been some 'ducking' left during sibilant 'sses' (during some of Karen Carpenters' singing for example).   The decode did help with some of the ducking, but it was incomplete.  That proved to me that in the recently released version, I am still not handling the 'attacks' 100% correctly.  If the processor handled the attacks correctly, then by far most, if not all of the apparent ducking would be cancelled out.

After a bunch more experimentation (testing, listening, research), etc -- I fixed that specific ducking problem.  It is amazing how sensitive the adjustments are to fully emulate the behavior of some old hardware!!!  That ducking (like during KC's singing) is fixed in all but one example that I know of --and that example is significantly better suppressed.  The HW cannot practically directly emulated -- I mean, I don't normally worry about the good old Is(1+e^(v/(vt))) (APPROX) type equation when programming -- that is something reserved for working with real semiconductors.  But, good old Ray Dolby took advantage of that kind of behavior a few times in the attack/release circuit design.)  BTW, that equation is a stylized version of the diode equation (current as a function of voltage.)   Dolby didn't use the diodes just as a one directional conductor, but also used them as a nonlinear resistor, taking advantage of their dynamic resistance decreasing with increasing current.   I am amazed about how much Ray Dolby got out of each component.  Nowadays, we'd likely use some kind of ASIC to do the work.

To fix the residual ducking, it required modifications in two sections of the code -- the secondary filter needed some rework -- the attacks for HF/hF are now fast & fixed, but the attack times for the LF/MF freqs are dynamic.   Also, the structure of the main filter code needed a very 'fiddly' modification where the first 2msec decay time constant filter ended up needing to be a symmetrical 2msec attack/2msec decay (for LF/MF) and 1msec equiv for HF/hF.  Then, there needed to be a slight change to the filter algorithm where it needed some (ideal) diode-type nonlinearity elsewhere in the math.  IT IS AMAZING HOW DIFFERENT THE RESULTS ARE WITH JUST A MSEC or TWO DIFFERENCE.  This is really sensitive stuff to correctly emulate an aspect of a 50yr old piece of HW. 


John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #44
More good news (yet.)  Much better decoding -- fewer left-over compression artifacts.  A bit less noise.

Regarding the improvement of the attack -- I have an example of EXACTLY the same piece (scarbourough fair).  I normalized the level/etc, so later on in the song (when it gets louder), the levels of all three (the 2 earlier examples) and this one ARE EQUIVALENT.

This new example results from again -- exactly the same piece -- processed by the decoder (latest version, not released yet -- but archived), the results normalized, and the take the first 30seconds as before.
Because of the better transient behavior, etc -- then my ability to set the threshold more appropriately, the results are better yet.
If you compare the undecoded with the first decoded -- there is an improvement.
If you compare the new one with the previously decoded version, you'll hear even more improvement!!!
This test is being done fair -- (normalized version of original source , render first 30secs),\
(first decode example: decode original source, normalize entire result, render first 30sec),
(second decode example: decode original source using improved program, normalize entire result, render first 30sec.)

I think that the left over compression artifacts are much better handled yet!!!

Re: Possible reason why many old music releases sound harsh...

Reply #45
Progress report -- found yet another bug in the decoder (still havent tested against a controlled source - and don't have a real spec, so mistakes a likely to happen.)
The expander/decoder up until yesterday (about 1wk after the most recent upload), had a problem with the lack of COMPLETE handling of the feedback expansion needed to fully decode.  This meant that a few dB of dynamic expansion was missing.   The static amount of expansion was pretty much correct, but how fast the expansion occurred and the rate was a bit slow.   I haven't been able to get a feedback design working correctly, so I am doing an explicit calculation of what happens during the feedback (really tedious.)  However, I have gotten another quanta of improvement.  Luckly, the DolbyA has a pretty well defined attack/release rate -- so doing the math to recursively correct the missing expansion is pretty reasonable.   It actually ends up being a fairly common series expansion (like a Taylor series -- shaped something like a log Taylor series), but cannot be expanded directly like a simple 'expansion in --> better expansion out', but rather it needs to know the historical behavior and filter out the static gain, and building up the changes.   It is still impossible to be theoretically perfect, but I can come close.  I am trying desperately to stay away from the patented method for DolbyA decoding -- simply implementing that isn't difficult at all -- I don't want any intellectual property issues to occur.
This makes a 'hell' of a hobby.

John

Re: Possible reason why many old music releases sound harsh...

Reply #46
If you're trying to actually decode real Dolby A, I can't image why you're avoiding their methods.  The patents expired years ago.

Re: Possible reason why many old music releases sound harsh...

Reply #47
For fun -- I have some demos that I produced and downloaded to a site.   Most are ABBA oriented, but there is some Olivia Newton John and Carpenters. I have been getting VERY GOOD reviews as to the quality.  Had lots of breakthroughs and my face has been "RED" several times with embarassment.  ANY OF MY PREVIOUS DEMOS HAS SUCKED COMPARED TO THESE.  I also have a dolbyA decoder that really works well (there were continual bugs regarding the dynamics until recently.)   You are welcome to the source if you PROMISE to try to port to MACOS, but Linux binaries reside on the site below.  Will do Win64 if told that someone will really attempt to use it.

I am a good software person, but not really a DSP expert, and even less an audio expert (I know DSP and know a little bit about audio, but dont' claim total expertise like I do software.)

REALLY!!! -- HAVE FUN LISTENING:

https://spaces.hightail.com/space/bOPBXTkeeT

There are many 10's of MB of good sounding stuff in there.

PS:  the patent that I am avoiding is a more recent Sony patent that does some things that make it better for computer DSP.   I have the code really working well now, so no biggie.  Thanks for the feedback!!!


John Dyson

Re: Possible reason why many old music releases sound harsh...

Reply #48
I have some useful/interesting updates available.
Firstly, the expander needed some retweaking -- the thresholds and compensation needed to be fixed:
sos.mp3, nog.mp3, st.mp3, uatk.mp3 (all ABBA -- corrections made after input.)

Fairly noticeable example(s) of noise reduction:
hiss reduction:  carp1970-01-orig.mp3, carp1970-01-exp.mp3 (listen for the his much decreased on -exp version)
rumble/rush reduction: SandG-orig.mp3, SandG-exp.mp3 (listen for less rumble, slight less 'rush' sound
Significant improvement in dynamics:
Original:  SandG02-orig.mp3  (LOTS of compression and some distortion -- UGLY)
Simple pseudo-DolbyA:  SandG02-exp.mp3 (Compression is better -- more listenable)
psuedo-DolbyA and full expander:  SandG02-proc.mp3 (Almost normal sounding -- very listenable)


I have gotten some feedback that the 'decoder' or 'pseudo-DolbyA expander' might not really be DolbyA or some of the files might not be really DolbyA encoded.  I don't necessarily disagree or even have direct knowledge, except:

THE RESULTS SEEM TO BE VERY GOOD!!!  Definitely NOT perfect, and I always accept and apply constructive criticism!!!

Gonna release the psuedo-DolbyA (or whatever it is) SOURCE CODE and additional binaries  when I get enough feedback that I am sure that it does what I want it to do..

John Dyson