|
Post by jsdyson on Jan 17, 2018 14:21:44 GMT
Gang -- I'd been reading other threads on this board, and noticed a thread 'dear' to my heart -- the abba sound. Didn't want to follow directly with that thread since it is long dead, but I might have something to offer. Even though my 'restoration processor' has been helpful in decoding the ABBA sound, it really isn't the key step. Geesh, even the Carpenters have used some of the processing that my processor can undo -- the compression isn't really the only key to 'the ABBA sound'. I have a formula which is expressible with a SOX/shell script which can be helpful to see what they have done. I am not going to provide it unless there is a lot of interest (it isn't long, but it is technical.) The key has been related to applying a 90deg phase shift and a 10msec time delay. The 90deg phase shift kind of allows mixing the sound in a way that it doesn't cancel, and the 10msec delay causes the very dense sound when summed back to the audio. I haven't derived the exact formula, but I created a decoding version that does work in most cases (as you can hear by some of my examples.) It is important to start with a pristine copy or the math probably won't work -- so the Polar releases or other obscure releases which weren't mis-processed are probably your best bet. I have an ancient, non-english release which contains about the cleanest copies that I have heard -- and that is what I use, very likely because the cost of finalizing that version wasn't deemed to be worthwhile. I haven't tried the Polar release yet for the test.
Even though I don't know their exact procedure, it is probably like: (take the signal delayed to match the hilbert 90deg transform -- but don't use the transform), then (take the signal witih hilber transform PLUS 10msec), add the signals together -- and voila -- you have close to the ABBA sound. I haven't checked the 'encoding' procedure, but I am using a similar inverse with good results. These tests and associated scaling parameters take time to find out (by trial and error), so that is the only reason why I haven't provided an exact encoding script.
I am willing to share my decoding script, but undoing the general steps above are the way that I have cleared up the voices. It is NOT fun using these tools, but it is nice to be able to hear the voices so clearly.
ADDON: Just tried the Polar release -- there are some hiccups, but comes close to working okay. The biggest problems is that they did some enhancement and some compression which messes up some of the math, but the algorithm still seems to apply.
John
|
|
|
Post by jsdyson on Jan 16, 2018 22:51:43 GMT
I worked REALLY hard on Dreamworld for the last hour. The version that I started with was passed through DolbyA HF emphasis (or severe compression with HF emphasis) too many times, so I had to run multiple decoding passes. Plus, I ran the restoration processor to get some of the transients back. It still isn't perfect, but is better than the original that I started with, better than the version that I found deep in my archives -- but NOT perfect by far. Please enjoy -- the filename is "DreamworldCleanBest.mp3". I am not claiming that it is the best possible, but is the best that I could do for right now. It might have required another pass of DolbyA and less lowpass filtering. As it is, I still added slight eq boost at 6.5kHz, and a low pass filter (one pole) at 9k -- which maintains the 'sses' fairly strongly while getting rid of the harsh HF.
John
|
|
|
Post by jsdyson on Jan 16, 2018 19:15:14 GMT
No -- I don't have the Voulez-vous deluxe edition, but will look again through all of my archives (I have many, many versions -- but will look them up again to see what I can find.) On the other hand, I have turned the restoration processor up to 20 (on a scale of 1-10), and posted the new version. It is pretty good, except still has some HF problems, and I pushed it a little farther than the algorithms are written to support. The voice has been cleaned up to be almost transparent. It is a WONDERFUL song, right up there with Cassandra. The new version is in the repo below -- DreamworldCleanNew.mp3, and I am going to search around again (I have probably 50GB of ABBA stuff.) spaces.hightail.com/space/pG4t4ZFnyBGOOD NEWS -- follow-on, I did find a copy of Dreamworld in my archives. When listening to it -- I think that it comes from the same place as my copy, but they have done a pretty artful job of minimizing the HF smearing -- it still has the smearing and is (like alot of ABBA stuff) hyper-compressed. I suspect that there are better copies out there than what I have. Since I now know what at least one copy of cleaned-up Dreamworld sounds like, I think that I can re-equalize my copy and avoid the hyper compression that I have on the copy that I just found. I am going to work on it -- nice little project. John
|
|
|
Post by jsdyson on Jan 16, 2018 18:42:40 GMT
Interesting that you mention 'Dreamworld.' I have a damaged copy of it only -- nothing really very clean, and have been working on cleaning up my copy (electronically.) I really do enjoy that song -- refreshing like there is indeed a new ABBA song. I just had some more ideas about removing some more of the damage from the copy that I have, and might give them a try tonight. Take a look at my other postings, and on my repo site I have my latest copy so far. I just thought about some more things to do last night -- the current version still has too much HF compression and I know that I can mitigate most of that, just don't know how much I can improve it.
Dreamworld is a really nice, pretty song. Wish/hope they can find some more jewels like that in their archives.
|
|
|
Post by jsdyson on Jan 15, 2018 13:58:56 GMT
For fun & enjoyment -- I applied my technology to a few songs. These are going to reside online for a few days, but they are probably as clear as anyone has heard outside of a studio (or without a tape smuggled out.) These are from a Spanish or Portuguese release from many years ago, and cleaned up as well as the current software design can do. The song names are SOSClean, TakeChanceClean and a few others. These were processed for vocal quality, but sound pretty good anyway. You'll notice a LOT less 'squeezed' sound. This technology works on other material also, but less dramatic (e.g. can take some of the raspyness from Karen Carpenter's recordings -- smoothing the voice out significantly -- showing the depth.) The software is definitely planned for source release (free) soon after binary in afew weeks. If you compare even with Polar releases, the sound will be much smoother. The site also seems to provide for direct play of mp3's again, so hope it works for easy use: spaces.hightail.com/space/pG4t4ZFnyBThe copy of Dreamworld comes from a terribly distorted/destroyed copy. The processor did a fantastic job of reaching into the signal and pulling out something that sounds almost correct -- even I was amazed. The results here come from the very latest version -- intend to freeze so that I can produce a distributable & fully tested version of the processor. John
|
|
|
Post by jsdyson on Jan 12, 2018 13:20:01 GMT
Just for fun, on the repo-site, I dropped a super cleaned up copy of 'Name of the Game'. It is EXACTLY the same was what is on ABBA Gold, but has been cleaned up to the point where the leading voice is quite clear. It has a bit of a stereo image problem -- and is definitely not commercial quality even though it is very clear sounding without losing its music value (in my opinion.) This rendition appears to have exposed a bug in my 'restoration' processor -- I eliminated the pseudo-DolbyA as the cause. The reason why I am making it available is simply to show how much can be done with a little signal processing. NOTHING has been added artificially, and nothing has been completely removed. The result of Name Of the Game (and I think, SuperTrouper) came from the various restoration processor pieces and a little bit of careful manipulation, but without the restoration processor, the final stages could not have been done.
Frankly, I found listening to these (and comparing with whatever other release) to be a lot of fun.
John
|
|
|
Post by jsdyson on Jan 11, 2018 18:40:18 GMT
I am intending to be very respectful -- even though I initiated this discussion, and interested in various peoples ideas/input, I am not meaning to criticize 'ABBA' nor downplaying the enjoyment that I (we all) have gotten over the years!!!
The input so-far has been helpful to me, and in fact pretty much agrees with my sensibilities that pretty much is: "I got lots of enjoyment from ABBA, and cleaning up some technical details won't change that fact.". My sensibilities also include (which all might not agree), that a little "clean up" as long as it doesn't go too far might be a good thing.
Of all of the music sources that I am testing/experimenting with WRT cleaning up the audio, I have probably spent about 1/3 of my time on ABBA stuff. I am working with approx 5-10 different groups of music sources (Carpenters, Anne Murray, Simon & Garfunkel, Queen, Carly Simon, Suzanne Vega, Some orchestral material) plus some newer ones for reference (Taylor Swift, EVescience(sp), and other relatively more recent groups), and some generic sources (Bangles, Bananarama, Chicago ) for simple verification to make sure that the intense processing doesn't break any of the other music. All of the groups/entertainers have different amounts of recording innovation, and final processing to produce the recording to be distributed. In some cases I really enjoy the music, and in other cases the music might be tedious or borderline irritating to me -- but the goal is to at least SLIGHTLY improve the listening experience from a technical standpoint WITHOUT causing noticeable damage. I have NO interest in hurting/changing the artistic choices.
If I do have a 'criticism' per-se about some of the ABBA distributions, it is not with ABBA nor the recording engineer, but rather sometimes the production (finalizing/final processing) of the records/CDs. Even though I have presented clearer copies of some of the material -- there is NO criticism of the ABBA team, because all of the stuff that I am showing is IN THE RECORDINGS. I didn't add or take away anything (other than some emphasis, similar in concept of tone control.)
Again, my ONLY possible criticism (if I was playing a judgmental game, which I AM NOT), is in the production of the records/CDs. The great variability between records/CDs appears to come from the choice of the music distributor. I do forget where the original ABBA version that I have used came from, but I seem to remember it was a Spanish or Portuguese language distribution -- and they seemed to have the RAWest version of any distribution that I have heard. I have had that copy for 5+yrs -- and has languished in my archives from back with the storage space was expensive. Since I am lucky enough to have a relatively RAW and relatively un-processed copy of some ABBA recordings, I can sometimes make more obvious changes because I have a relatively less processed copy to begin with. Some of my other work actually uses very processed copy, and have had pretty good results with those also.
If I had a goal about ABBA -- it would be to help produce a distribution version of some of their recordings with more of a 2000's technical sensibility.
One note -- I mistakenly placed two full versions of 'Under Attack' on my repo-site (218.mp3, and 218-remix.mp3). I am going to have to remove them before the ENDOFDAY -- wanted to keep the clips short. However, when I first listened to these versions of 'Under Attack', I was amazed with how much of the music is 'unavailable' to one's ears in the normal commercial recordings. You just might want to grab and listen to these versions (esp the 'vanilla' 218.mp3) -- they are a musical experience. I actually like the 218-remix better. I have about 30-40 of these songs reprocessed, and it they are just so pretty. I won't be able to distribute all of them, but I hope to make my tools available (hopefully soon/almost definitely free) for others (you'all) to be able to clean-up old recordings also.
My bigger (more global) goal is to produce a set of general purpose tools that support the gentle, but important cleaning up of older recordings (without changing their character.) My scope is limited to take recording, but someday if I had the right hardware for testing, I'd do some things to help with records also (but the tick/pop issues have been pretty much solved -- the nonlinearity on the other hand probably still could use some work.)
Given a film analogy, I would NOT be colorizing the old films (for example), but rather removing the scratches, mitigating some of the grain, and rebalancing the color to be more accurate.
Sincerely, and interested in more input and more ideas!!!
John Dyson
|
|
|
Post by jsdyson on Jan 11, 2018 15:40:24 GMT
I am hoping more for a discussion instead of 'survey', but this matter has been interesting to me since day one. I have two versions of 'name of the game' (trucated to 30 seconds, but I can supply those really interested in full versions if already have copy of ABBA Gold or whatever... Here it is -- there are two versions -- one is a 30second snippet of a version from a Polar disk. The other is a processed and somewhat roughly cleaned up version where the voice is clearer/more open sounding. My big question is if the very 'mixed' sound of ABBA is of integral importance, or would it have been 'better' to have had a clearer rendition of the voices. It seems like about 1/2 of the songs have the voices very mixed, while others have relatively clear renditions of the singer (usually A or AF.) Song files: nog-orig.mp3 (original), nog-remixcomp.mp3 (clarified, but probably inferior version) spaces.hightail.com/space/pG4t4ZFnyBNow, in my opinion, this attemp/demot of producing a 'clearer' version of the voice is inferior (from both technical and artistic reasons), even though the technical aspect can probably be made to be equivalent given a significant investment. Would the clearer version have been better from the beginning (assuming better mixing/creativity/etc), or is the somewhat 'foggy' sound of some of the ABBA songs really a part of the expected personality of the group recordings? Given the Polar (or Polydor) versions are usually considered better than the others (e.g. various US manufacturers -- which I agree with), this is a comparison of one of the best normally available versions with a hacked-up (admittedly) but clearer version in some respect. I guess the real question is -- if it would have been done more clearly to begin with, would it have been better/happier/more enjoyable? I don't know the answer to the question myself, but this has been one of the imponderable questions that I have had almost the first time that I listened to ABBA. I do really like their music, and enjoy their recordings as-is, but could they have been 'better'?If 'clearer', more voice-transparent versions could have been made available, would hit have been worth it? Please note -- I am not criticising them or claiming that I could have been better, but otherwise I have so much respect and have historically enjoyed their music so very much -- this is simply interesting to me... John Dyson
|
|
|
Post by jsdyson on Jan 10, 2018 4:24:15 GMT
This remix isn't major, but is slightly less garbled in the chorus and the voice is a little clearer (not orders of magnitude -- just a little.) This came as a byproduct of some of my own curiosity about how they got the intensity in the music. Frankly, I figured out how to clearly isolate some of the pieces (including the lead voice) without doing a simple L+R cancellation -- but really getting a stereo copy. When I got done playing with that, I did a little experimentation about how to put it back together and slightly improving the clarity of the voice and slightly lessening the density of the chorus. The best improvement of the chorus is nearer the beginning. The voice is a little better (just a little) all the way through. The 'remix' is on my repo site, and the name is 106-remix.mp3. It is encoded at a pretty high quality and the original copy comes from my processed copy from the pseudo-DolbyA/restoration complex. spaces.hightail.com/space/pG4t4ZFnyBHope you enjoy. Don't know if it sounds better to other people than myself. EVEN TO ME IT IS FAR FROM PERFECT, and know that after a day or so it could be cleaned up a lot more, but I just don't have time to play with it. Also, I am downloading a cleaned-up Dreamworld. The orignal that I have is really messed up -- but I should have this new one ready by about 2Hrs from now. Sincerely John PS: I tried the same parameters on Under Attack. Where I got maybe a slght improvement on the chorus in SuperTrouper (if changed SuperTrouper too much, it would not sound correct.) On Under Attack it sucks badly -- really badly. The foreground voice is overly emphasized and sounds unworldly (not phasing or anything like that, but just WAY too present.) So much for fun and games -- just don't have time to tweak for each song.
|
|
|
Post by jsdyson on Jan 6, 2018 17:41:11 GMT
Demo music is definitely ready now. I was WOWED with these results. VERY CLEAR -- and no garbling or time delay distortions between the gain and the signal. Bottom line -- unless you heard a studio copy, I doubt that you have heard these releases so well. In fact, I am going to promote some of my ABBA results into my set of demos for my audio processor -- before I only had some Carpenters, Brasil'66, Anne Murray, Simon & Garfunkel -- but now these ABBA results sound great and SOUND LIKE ABBA!!! MOST ARE READY AND AVAILABLE NOW...Let me know what you think. The music names are based upon the track numbers for abba gold and abba goldII (simpler names than the long english ones.) All are high quality MP3 encodings -- everything was originally processed as 24bit flac at 96k and converted to normal very high quality mp3. Gonna keep on adding songs until later tonight. I have three or four more definite selections, but there are several good ones on there right now. The repository location: spaces.hightail.com/space/pG4t4ZFnyB
|
|
|
Post by jsdyson on Jan 6, 2018 13:01:11 GMT
I do have a 100% clean copy of the Visitors, plus a Polar & Polydor copy. The Polar & Polydor appear to sound very pretty -- I'll see if I can produce an essentially unprocessed version that sounds good. Sometimes, they (ABBA) did a really good job of processing, then sometimes they seemed to have jammed too much into the music. I understand their market -- but I do wish that they let the 'girls' voices be more prominent. I'll work on the Visitors right away and see what I can do.
John Dyson
|
|
|
Post by jsdyson on Jan 6, 2018 9:17:19 GMT
Finally -- the repository now has REALLY REALLY better versions of the reprocessed ABBA songs. The names of the songs on the site are based upon their track number of ABBA GOLD & ABBA GOLD II. The sound is more present and generally less distorted than the originals (and better than my previous tries earlier today.) Also, they are less compressed. For those who really like the 'compressed sound', I have produced a few songs as 'finalized', where I have enabled my finalizing processor. I didn't spend a lot of time tuning the finalizer, so the compression might be a little odd. MUCH MUCH BETTER. AGAIN -- this has LOTS of bugfixes when compared with the original copies that I produced earlier today. The attack times are better synchronized, and shaped more like DolbyA attack times. Also, the attack times are better synchronized with the music. The total effect is that the garbling effect should be totally gone. Also, I am going to distribute the latest version of my psuedo-DolbyA decoder later on today/tomorrow. It has some attack time bugfixes, and the DolbyA decoder is doing about 1/2 of the processing chores of the 'restoration' processor suite. I hope to distribute the rest of the restoration suite within a few weeks. That thing is COMPLICATED, and might still have some lurking audio processing bugs. ABBA songs on my repo site: 103.mp3 (Take a chance on me), 104.mp3 (Mama Mia), 106.mp3 (SuperTrouper), 110.mp3 (SOS), 117.mp3 (Name of the Game), 216.mp3 (You are the City), 218.mp3(Under attack.) A few of the songs have 'finalized' versions, where they have more of the traditional ABBA sound, but are not nearly as severely compressed (103-finalized.mp3, 110-finalized.mp3, 216-finalized.mp3 and 218-finalized.mp3.) I am probably going to add Dreamworld soon also -- but the copy that I have is pretty rough, so my processor has troubles finding the hints in the structure to reconstruct the sound. I am adding Dreamworld because it is so uncommon. If you have any favorites, I might be able to clean one of those up a bit (bring the sound forward a bit), but I only have ABBA GOLD, ABBA GOLD II and a few others. The purpose of this exercise is primarily as fodder for my audio processing software, but I enjoy ABBA and I hope that you'll enough the music before I have to delete it (I don't want to keep it all on the site for very long.) PLEASE ENJOY IT. Here is the respository again: spaces.hightail.com/space/pG4t4ZFnyBJohn Dyson
|
|
|
Post by jsdyson on Jan 6, 2018 1:12:45 GMT
UPDATE -- I have fixed the harshness/transient balance problem (to a large extent). I am getting good transient behavior and a lot less harsh sound. I made several mistakes -- actually forgot to enable the full transient recovery mechanism -- so all I could do is to make it nastier sounding instead of more clean/intense. Also, I had an incorrect sync between the audio signal and the audio level signal -- they were more out of sync than desirable (everytime I change a certain set of parameters, I have to change some corresponding delay parameters -- but forgot to do that.) One more thing -- I wanted my version to sound more like Polar's version but without so darned much compression. Well -- I figured out that the distribution that I have has a matrix difference between typical ABBA releases and what my version has. SO, I have tried to rebalance the matrix (hard to explain, but has to do with the spatial presentation of the singers/instruments/etc -- and can actually be tuned.) After these changes, the results are better than anything that I have demoed here before.
It takes approximately 20minutes per song for the audio processor to do the cleanup -- the hard part is the transient recovery (reshaping the audio signal so that sharp/intense sounds appear more clear.) When all of them are complete (probably tomorrow morning), I'll post the updates. After this, I intend to leave this specific matter alone for a while to complete some other things. However, I'll happily take specific processing requests (favorite songs or whatever), and make them available fairly quickly.
This is very tricky to try to complete a phase of a project -- there are always BUGS -- and all of the bugs have been my own screwups.
FIgure on something on the repository and an announcement here in approx 12Hrs or less.
John Dyson
|
|
|
Post by jsdyson on Jan 5, 2018 20:25:57 GMT
I have added some more examples to the repository. Firstly, I have a very-difficult-to-process copy of 'dreamworld.' I only have a very heavily compressed copy of dreamworld, so I gave it a try and uploaded it to the repo site. It sounds pretty good, but is still pretty damaged. Also -- I have uploaded some 'finalized' copies of the previously processed songs. These 'finalized' copies come from the processed versions that I had previously uploaded -- from EXACTLY the same files (except that I locally work in 96k/24bit flac, and not mp3.) One note -- the quality isn't really high enough to hear much of a difference between 200+k mp3s and flac, but I started with the original flac copies of the processed versions simply because those are the versions that I traditionally use. So, I have 'finalized' versions of 'under attack', 'take a chance' and some others... These have been compressed and sweetened a bit... I didn't spend a lot of time doing careful finalization, but you can probably hear that ABBA is really meant to be listened to in compressed form.
I REALLY hope that you enjoy. I am still working on the restoration processor and psuedo-DolbyA decoders. On most ancient music ('60s through '80s), the software does really help. ABBA is definitely a worst case, and very difficult to process my probably one of my very favorite bands!!!
|
|
|
Post by jsdyson on Jan 5, 2018 14:45:57 GMT
Since you seemed to be entertained by SuperTrouper and MamaMia -- I just added I am the city (216.mp3) and Under Attack (218.mp3). I am working on Take a chance on me or maybe another one -- not sure which one sounds best... I think Take a chance sounds pretty good -- so I should be uploading it (103.mp3) in about an hour. I am actually working on a super restored version of SuperTrouper -- cleaning up the choruses better -- but it is VERY tricky. They really hyper compressed some of that, and did some spatial matrix things that make things challenging. Everything that I am doing is meant to be minimally changed -- just enough to modernize the sound a little. So -- probably expect 'take a chance' about 1Hr from now.
One caveat -- the resulting sound is very RAW. I didn't 'finalize' the sound at all -- so there is no treble enhancement, bass boosting or even normal compression after the restoration/recovery processing. Normally, one might add some strategic compression/limiting and some EQ here or there. I have done NONE of that. I do have some last ditch limting enabled so that if the signal gets too loud that it will keep it from clipping. The limiters (there are several passes of limiters in the processor that I wrote) haven't been triggered, however. I carefully made sure that the output doesn't go into limiting.
Frankly, the music might have sounded a LITTLE better with some dynamic range control -- but it takes a lot more time to tune the right amount (the processor has several kinds of compressors and limiters -- each one does something special.) I haven't been focusing on that stuff right now, and working solely on signal clean-up/restoration/noise reduction/peak restoration/etc.
John Dyson
|
|
|
Post by jsdyson on Jan 5, 2018 3:57:59 GMT
Okay -- i have felt a little guilty about the 'tease', which I didn't mean to do. Here, I am posting a repository, where you can download a few mp3s which are examples of some of the work that I am doing. Both are processed copies of the best example of each song: SuperTrouper and MamaMia. you'll notice that on SuperTrouper that there is absolutely no intermodulation at the start -- many of the released versions do have an ugly intermod at the beginning. These have NOT been finalized, but have been processed by my processor (from the best possible source), with approx 1:1.50 expansion -- with huge amounts of NR and transient recovery. Unfortunately, much ABBA stuff has been so very processed that my processor has terrible troubles during reconstruction of the music. This stuff is far from perfect, but is very minimally compressed (perhaps 1-3dB of limiting) only because the large amount of other processing had left too much dynamic range in the music. These are not the best examples of my 'restoration' processor output, but does sound very different than most ABBA releases. Again, these are from the VERY BEST source that I have been able to find. The filenames are 104.mp3 and 106.mp3 (appropriately named from their position on the ABBA Gold album.) Here is the url for download: spaces.hightail.com/space/pG4t4ZFnyBJust select and download the files. Sometimes, you can play the files directly, but it is best to just download and play. Give me feedback, if you wish. I happily accept constructive criticism. Also for demo purposes for the audio processor (in some easier cases), I am adding 'Rainy Days and Mondays' from the carpenters and also Brasil'66 example. They should appear within an additional 1Hr after this posting update. The ABBA examples are already there. John Dyson
|
|
|
Post by jsdyson on Jan 5, 2018 0:48:00 GMT
I'll put together some of the results. Because of where i am on other things, might be a few weeks (well, I think that I can promise under two weeks.) Much progress has been made -- I have been focusing only on DolbyA-style decode and general expansion project right now. So, the results of 'undoing' some excess compression are the most prominent right now. The 'advanced' mild recompression phase is on hold -- which is mostly not really needed. I have been searching out all of the very best copies of ABBA music that I can find, and then starting with those. Havent been just focusing on ABBA, but ABBA is one of my most important (and VERY challenging) test subjects. They did a really good job of tightening up the music so that it is very difficult to seperate (definitely sounds like much was done as a radio mix.) On the other hand, others of my test subjects 'Brasil'66', and Carpenters 1971 album, the results are fantastic. (You can actually see the tape hiss/noise reduction on the spectrograms, while also achieving tasteful amounts of transient improvement.) The ABBA stuff is beautiful as it is, but definitely a bit eccentric to clean up so as to sound like a more current production. I do wish that they (B&B) could muster up the energy to get into their master tapes and bring up the results to something that we'd normally expect to be done today. When I critically (VERY critically) listen to a few sources that are pre-commercial-distribution, I can hear that there is A LOT of 20XX's quality in there, and I am afraid that it will be lost to eternity. Again, as it is, the ABBA stuff is beautiful in its own right, and alot of good could be done (I believe) by doing a pristine remix and remastering. Even though I am pure techie, I do know that if I had the 20-40 tracks of music (say, 20 songs with 16-32 track master tape on computer media), and a year of time, I could do a pretty damned clean job of it... As it is, I have a few copies of Polar releases, Polydor releases, US releases, and a few bootleg copies to work with (I very honestly forget where I got them, but it was from many years ago -- in my later years and greater experience might guess that some might be two track DolbyA copies.) My goal has been to produce an optimized listening experience of what I have found to be available. (This is beyond my interest in generally cleaning up old recordings -- removing hiss, truly replacing damaged transients, etc.) All of my software actually reconstructs and doesn't just 'transient enhance' or whatever -- so it takes lots of CPU, but the results are amazingly uneventful, but music sounds like it should (not over-enhanced, but just 'cleaner.')
|
|
|
Post by jsdyson on Nov 21, 2017 7:49:15 GMT
I have re-configured the processors (still in progress -- approx 3wks before I can release it.) I have done a good amount of clarification/seperation of the voices. I know that supertrouper is notorious for the chorus being a little too dense. This demo really clears it up -- but it still sounds like the same song!!! Filename: abba-106-st-voicesclearer.mp3. The repository location: spaces.hightail.com/space/KzA53ZKeFcLet me know how it sounds to you. I have almost all of AbbaGold and Abba GoldII processed this way, and so far my listening hasn't detected many major mistakes yet. John
|
|
|
Post by jsdyson on Nov 17, 2017 1:45:23 GMT
I have made progress in fixing the demos -- I have gotten the normal demos up to waterloo (up to 119 -- I am not providing all of the AG songs, just a few.) Also, I enabled one of the features that I held back that decodes more info out of the music... So, if you try playing some ABBA songs in the repo with the name 'BETTER' in them -- you might really enjoy. I am building approx 1/Hr for right now (it is really timeconsuming to run the very complex algorithms.) spaces.hightail.com/space/KzA53ZKeFcThey will disappear in approx 2-3days... I would really appreciate reviews (either way) as to the sound quality. All abba processing was done with the entire processor (not just the 4band.) The updated source for the 4bad resides in: spaces.hightail.com/space/bOPBXTkeeTThe 4 band processor filename is da.zip -- contains Linux binaries and all of the C++ source code. Might be fun for DSP/EE/software types to see what SIMD/vector audio processing looks like. John
|
|
|
Post by jsdyson on Nov 16, 2017 16:32:11 GMT
gang -- I made a big mistake in the da decoder -- I commented out an important filter, and it caused a hashness in the results. I'll post fixed versions of the music by the endof day 16nov2017. I HATE IT WHEN I MAKE MISTAKES!!!! John
|
|
|
Post by jsdyson on Nov 16, 2017 3:57:53 GMT
Been working hard on the audio restoration software. The software takes the best copies of what I can find and tries to remove excessive processing -- this isn't just an ABBA project, but I do use ABBA distributions/songs for test subjects. I have two styles of results -- and even though I think that I have decided on one of the styles, I have left the other style available on my repository for comment. The processor consists of two different kinds of reprocessing (in a way -- deprocessing) software. Which combination of software packages to use is dependent on the status of the source material. The source code of the reprocessing software is intended to be released, where the first (and simpler) one is currently available for free in source & binary forms. I am most interested in listening resuls -- positive or negative. So, I am providing two pointers to the material -- one contains mostly the audio/music demos. The demos will be disappearing within days for obvious reasons. The other is the source code and binaries for the first (simplest) music clean-up package. The bigger (very complex/CPU intensive) package will be forthcoming within days or few weeks. The music demos (for feedback) have obvious names on the repository like: abba-106-supertrouper.mp3. The files with 'NOA' in the filename have a slghtly different form of reprocessing, and I'ld like feedback that talks about the difference between the NOA versions and the non-NOA versions. The music URL is: spaces.hightail.com/space/KzA53ZKeFc and the files will be disappearing within a few days -- so both evaluate and enjoy. Feedback is very welcome. For those who want to do the simple/first phase of processing testing, the binaries/sources reside elsewhere. The processing is a very simple, very gentle expander that is great for cleaning up over compressed material, and it can help to get rid of the 'bouncing backgorund' levels, some of the pumping, excess ambiance, etc that comes from too much compression. There are few adjustments because the scope/purpose of this simple processor is quite limited. I am distributing Linux and Windows64 binaries righit now -- and they are command line because frankly, I am a really good DSP, Operating Systems and Embedded programmer, but suck at GUI. Source code/binaries: spaces.hightail.com/space/bOPBXTkeeTThe Linux X86 binaries reside in files like da-avx, da-sse3, da32-avx, etc. For recent workstations, the da-avx is probably correct. For recent laptops, then da-sse3 or da32-sse3 are probably the best. For older machines, then the da32-p4 is the last chance :-). Refer to the readme.txt & damanual.txt for more info. The windows X86-64 binaries reside in the da-win.zip file. This zipfile contains the .exe file and some needed .dlls. The zipfile also contains a readme similar to the one in the repository itself. Feel free to contact me if you have troubles and/or even complaints. This piece of software IS NOT aggressive in its processing, but it can work miracles when it is appropriate. The processing done to the music examples is more aggressive, but mathematically and technically correct. I need feedback both from people with younger ears than myself plus someone who might have different tastes than myself. Thanks!!! John
|
|
|
Post by jsdyson on Nov 13, 2017 13:51:39 GMT
Good news -- I have been getting feedback on some technical groups -- and seeing darned good results for the audio recovery software package... Making huge progress. I have some results that might be of some interest to ABBA lovers -- I have been running the recovery software on select ABBA songs (adjusting the recovery software/equalization/etc to match the Polar EQ to some extent.) Don't get me wrong -- it does sound 'different' than the Polar release -- it has taken reference to all of the various releases to produce these results. I believe that these results are as good as public copies can get... I have a pointer to the repository below, and right now I have three of the songs 'recovered' -- abba-103-chanceonme.mp3, abba-104-mamamia.mp3 and abba-106-supertrouper.mp3. I was very generous with the MP3 encoding parameters -- so there is really no noticeable loss from my .flac processing output copies. The 'recovery' software has been promised to be released, and also the pseudo-DolbyA decoder will also be released (sooner than the recovery software) in source form. The ABBA results that I am pointing to here is NOT using the DolbyA decoding software, even though some of the other examples on the site are indeed using psuedo-DolbyA decoding also. The sound is VERY musical... I will have to remove these examples soon (within days), but I hope that you might enjoy the music. *the recovery software is VERY CPU intensive -- much more than just a simple expander -- and running off the songs takes LOTS of time. I hope to be able to present perhaps 2-3 more songs before the end of the day. (I am not using the entire gamut of the recovery software -- the ABBA recordings are a serious challenge and sometimes cause the software to make mistakes... These copies are essentially correct relative to what the recovery software can do now. spaces.hightail.com/space/KzA53ZKeFcJohn
|
|
|
Post by jsdyson on Oct 22, 2017 1:41:28 GMT
Just for fun, I added some compressed/enhanced versions of some songs (filenames ending with -enh). This is simply an extension of the Decoding/recovery effort. I added in some additional stages of my processor/finalizer including a few different kinds of compression/limiting stages. (my compressor/limiters don't sound quite the same as most.) The results sound like 'ABBA' -- but the changes are MUCH MUCH more conservative than some of the real CD releases. (Some of the real CD releases then went to heroic efforts, but mine are simple mods to make them more listenable.) Frankly, some of the real releases are 'prettier' than mine, but mine are more 'real.'
The reside on the same site -- so far I have mmia-enh and nog-enh.
John Dyson
|
|
|
Post by jsdyson on Oct 20, 2017 22:41:29 GMT
I have updated some of the demos -- got some feedback that the EQ was wrong on the ABBA examples, so I fixed it to better match the Polar sound. Also, the dynamic range expander has been improved. So -- the normally decoded/conservatively improved versions are; st.mp3, nog.mp3, sos.mp3. The more aggresively restored versions are: st-proc.mp3, not-proc.mp3, and sos-proc.mp3. It all resides on the same site: spaces.hightail.com/space/bOPBXTkeeTSome of the examples are truly very pretty... I have tried to do the best that I can do, and when I get enough feedback I intend to release source. I am going to try to release a new version of the programs for both Windows64, Linux64 and Linux32 (the Linux versions will support i3/i5/i7 processors, ATOM processors and old P4 processors... The Windows version will be optimized for the ATOM, but will also work on the i3/i5/i7 -- but not the P4.) The new programs should be available by the end of the weekend. All will be command line. I am still considering the development of a plugin implementing the pseudo-DolbyA decoder ONLY. The full expander set might be freed, but I truly need listening test feedback. I have definitely made a commitment for releasing the simpler decoder ONCE I KNOW THAT IT IS STABLE. John Dyson
|
|
|
Post by jsdyson on Oct 18, 2017 17:31:35 GMT
I'll consider the plugin thing. Previously, I hadn't been thinking 'plugin' because my other (main) project is incredibly more complex and uses LOTS AND LOTS more CPU. The pseudo-DolbyA 'expander' on the other hand might be using perhaps 1/10 of one Intel Haswell CPU core (Intel 4770) and is single threaded (for real-time.) My other project uses approx 3 CPU cores (it is highly multi-threaded.) Used to be the psuedo-DolbyA expander was part of the bigger project -- but I realized that there would be a lot of use for those who don't want or need the 'monster.' :-). I'll look into what is needed to do the plug-in thing. I was also thinking about Java - even though it is 4X slower in this case than the C++ vector oriented code that I wrote. (Normally, Java is not that much slower, but Java doesn't know much about SIMD/vectors yet.) Java would give access to those who aren't running on Intel.
Thanks for the input, and I am (right now) thinking it through. John
|
|
|
Post by jsdyson on Oct 17, 2017 4:24:18 GMT
Thanks for the interest and your positive feedback!!! I happily accept criticism also -- we all need feedback to learn...
Anyway -- have downloaded a few new fun examples onto the same site. The 'dreamworld' one uses more of my technology.
As you know, my copy of 'dreamworld' is fairly mangled. I applied my full processor (very judiciously) to help correct some of the defects. My processor (this is the full one -- NOT just the DolbyA program) has several capabilities, including 3 full expanders, several compressors of different types, and several kinds of limiters. Not going into a lot of detail except to talk about the expanders (and 4 of the various limiters are kicking in, but I doubt that you will notice them.) The expanders probably need further description, because they are NOT simple expanders. (one of the dreamworld defects either appears to be TOO MUCH mp3 type compression or maybe mag tape saturation -- not sure -- it has been 20yrs since I have heard reel-to-reel. I cannot fix the 'swirls' in the freq response, but can help the dynamics a bit.)
These expanders which are helping to clean up the over compression in 'Dreamworld' have both 'linear-in-dB' and also a nonlinear expander that mathematically (PARTIALLY) undo normal linear compressors and some kinds of limiters with time varying characteristics calculated from a Volterra series kind of math._ So, the expanders have a normal longer term gainup/gaindown capability, but also boost-up and reach-down the gain very aggressively without distortion. The goal is to maximize the expansion without the ear being able to hear 'SURGING.' Surging is one of the common defects of expanders, but these expanders can have EFFECTIVE expansion ratios beyond 1:1.5 or greater without surging. This copy of 'Dreamworld' is getting a peak of perhaps 1:1.8 dB/dB expansion, but a base of about 1:1.3dB/dB or so. (the expansion isn't really dB/dB, but the numbers are approximate.) The total amount of dB gain increase/decrease over a couple of seconds often repeatedly have a 7-10dB range. Some of the changes are several dB UP&DOWN both within a 10msec timeframe for midrange on up.
The name for this new 'Dreamworld' demo is dreamworld-fix.mp3.
Also, there are a few more morsels in the repository, all might be fun and interesting to listen to. Have fun hearing somewhat better dynamic range and clarity. I am actually enjoying the music without being critical for a change, because the cleaned-up music MUCH BETTER MEETS MY OWN STANDARDS. Some of the songs are prettier than I had ever thought that I would ever get a chance to listen to.
If you have ideas for improvement, I'll certainly try to figure out how to convert the suggestions into software.
John
|
|
|
Post by jsdyson on Oct 16, 2017 14:14:39 GMT
Regarding the MacOS thingie -- oddly, I wrote much of the original FreeBSD kernel whose associated userland was a basis for early versions of the MAC stuff. I haven't done much OS work in the last 15yrs though (too much Testosterone in the kernel-kiddies like Linus who was writing a competing piece of software). My nerves just could not stand it anymore. My life was never 100% FreeBSD even though I had often been making lots of money while being one of the original authors of the code.
The best that I can do is to strip down the code enough that you could port it to MAC OS. The DolbyA decoder isn't objective-C, but instead C++. Current version has LOTS AND LOTS of extra stuff left over when I used the infrastructure for my finalizer to produce the DolbyA decoder (a lot of the needed infrastructure is theoretically similar.) Then, I removed the multi-thread support because it isn't needed. Also, the finalizer uses 8 bands of audio, yet the DolbyA only needs 4 bands -- so I had to change some variable sizes/etc. As it is now, with a half day of work, I can probably get the program down to 2-5k lines of code. Right now, it contains lots of stuff not needed (e.g my dynamic attack/release classes, and unneeded linear filter classes, and even my multi-threading communcations classes.) Stripping that stuff out might make a porting effort for someone who hasn't worked with the code for years much easier.
Probably the biggest challenge for even an expert C++ programmer is dealing with all of the vector math. Luckily, the main DolbyA decoder code is actually quite short once the support infrastructure is in place. Let me start an effort of shrinking the code down to the necessary bits (which needs to be done anyway), and I'll let you (or someone who wants to) do a port to MAC/OS.
John Dyson
|
|
|
Post by jsdyson on Oct 16, 2017 5:19:44 GMT
I have added a potentially interesting program to the repository. It works only on Intel Linux boxes -- but I am providing both a 32bit and 64bit version. The best one is meant for recent CPUS, but I did build a 32bit Pentium4 version. I don't normally do windows, but do have a windows64 build that I can produce within a day or so (if REALLY desired.)
It is a DolbyA decoder that is doing 75% of the work on the music that I have been showing... I have LOTS more stuff with sophisticated algorithms, so giving this away is meant as a helpful gesture (hope it doesn't waste anyones time.) IT ONLY WORKS ON DolbyA encoded material, and most ABBA stuff is not DolbyA encoded, but some stuff that I have purchased online is decodable by the decoder.
It is a Linux command line program, and takes several standard wav file types in and out. There is a short blurb on the site that shows how to use it. (damanual.txt)
Names: da-avx, da-sse3, da32-p4, damanual.txt.
This version of the code is literally only about 4Hrs old -- so can be quite buggy. ONLY THE da-avx version is well tested. This is BY FAR the best version that I have produced (you can hear the results), even though earlier versions did help.
It resides on the same site.
Good luck. Hope you enjoy.
John
|
|
|
Post by jsdyson on Oct 16, 2017 4:50:11 GMT
ANOTHER MAYBE kind of cool thing...
I looked around for something truly interesting --- did a cleanup of my best copy of 'dreamworld.' It resides on the previously stated repository.
Alll of the examples that I am showing have the least possible processing that will clean up the sound as well as I can. I have other kinds of expanders and carefully crafted compressors available (self-written), but don't want to taint anything very much.
John
|
|
|
Post by jsdyson on Oct 16, 2017 4:18:01 GMT
Guys -- take a look at the same place as before. VASTLY improved copies. I made an algorithmic mistake in my expander -- and some good stuff 'just popped out'. Been trying hard. Also, in the little 'online' repository, I have a Carpenters and Olivia Newton John piece. The main reason why the sound can be so good is that it is now almost 100% determined (by me) that alot of 'old stuff' available for download is DolbyA encoded (apparently, they are just running off tapes and transferring to media). This is likely only for the high quality versions -- because I recently in the last few years bought a total cr*p copy of ABBA also. THE SOUND IN THESE NEW ONES WILL KNOCK YOUR SOCKS OFF. (the sound did knock mine off :-)). The names are the same, but you'll notice the date/time are different.
spaces.hightail.com/space/bOPBXTkeeT
|
|