Jump to content

John Dyson

  • Content Count

    218
  • Joined

  • Last visited

1 Follower

About John Dyson

Recent Profile Visitors

The recent visitors block is disabled and is not being shown to other users.

  1. Thanks -- there were probably three reasons for chatting about these decoding examples -- lossy compression in ideal conditions, the mastering differences being MUCH more important than the last small amount of sample rate or bit depth, and trying to show the potential improvement available (using my decoder or perhaps some superior one that might yet be invented by someone who really IS an audio expert :-)). After all of this -- still off topic, but an example of the h*ll that I am in -- the demos that I made were done by a version of the decoder with a bug remaining (just found the bug, and the decoder *IS* complicated - probably 10X more complicated than a typical compressor software package -- running in 20threads.) I still think that the general quality is still well represented even with the bug. We have a LONG WAY to go before some of the finer grained improvements are really needed -- this first and most important (and where we have least amount of control) is in the source material itself -- that is, the mastering part of the process. The bit-rate (or with MQA, the bit-rot) is important, but not the biggest problem. I'll probably end up updating the repository (it will be the same location) early tomorrow with the corrected material. As it is, again -- the examples -- for looking at possible quality improvements, and comparing with what mp3 can do -- that is still valid. (I could have probably kept the defect hidden -- but that is NOT in my nature -- I truly believe in transparency and attempting to be honest.) I *think* that starting with the highest quality material possible -- mp3 (or opus) can handle any case where bit reduction is really needed nowadays. Otherwise, we are VERY WELL COVERED with .flac and other lossless compression schemes. The in-between data-reduction doesn't have a place anymore -- time has come and gone. There is *ZERO* advantage of a lossy compression scheme that controls our freedom also. All bases (for our needs) are already covered nowadays. John
  2. I am so sorry that this took so long to do. I get distracted and forget things. Lately, been writing some licensing software, and that is so far afield from what I know that I am becoming frustrated all of the time. I put together some chopped (sorry, I might have forgotten to 'fade' them instead of chop them) demos of freshly and unEQed decoding results by the DolbyA compatible decoder. First -- I did NO EQ or ANY KIND of compression of the decoding output. This is RAW output, and I have just rebuilt the decoder correctly (I think...) Been working on some very subtle performance improvements -- which often totally break the decoder for a while, but I think that it is put back together correctly. There are also equivalent .mp3 results at 320k. Listening carefully, you can tell that the output directly from a DolbyA encoded (then decoded) copy can be pretty darned high quality -- when compared to the original source material. IF you do compare with equivalent commercial releases -- make sure to start with properly mastered releases -- a lot of releases are DolbyA encoded or badly EQed. I usually try to use original vinyl for comparison. The vinyl is not higher quality, but is a good basis of comparison for EQ. ABBA is doubly problematical -- their music is consistently difficult to properly decode, but there are often flaws in the recordings also. Another thing about ABBA -- I have 3 undecoded DolbyA copies of some ABBA songs, and each one sounds different. I think that the SOS example that I uploaded is not mutiply encoded/decoded. DolbyA starts getting ugly after multiple encode/decode cycles, so I have tried to find the best sources in my collection. One other thing -- it is INCREDIBLY tricky to distinguish between poor mastering and leaked DolbyA material. I can often be tricked into thinking that something is leaked DolbyA, when it really isn't. I use multiple criteria to demonstrate the high liklihoood -- including looking at the noise spectrum -- because DolbyA creates certain patterns and shapes in the spectograms. Also, I am pretty good at detecting the HF compression when using DolbyA, but pop material is already compressed -- so it can be confusing to make the correct judgement call. So, when I don't actually KNOW that I am providing properly decoded material, I have super high confidence that this started with leaked DolbyA material. (Legally leaked/properly purchased is what I mean.) All said, there is almost the best possible copies of some of the material, and the mp3 version -- encoded with lame in 'insane' mode. I will also be trying to compare these today (again), and others might be interested in trying to detect differences. THESE ARE SHORTENED to try to be a good citizen, but to avoid any kind of long-term distribution of even the snippets, I'll be deleting these examples within about a wk or so. These are shortened enough that the material is a tease and might motivate purchase of the music :-). I am notoriously bad at swapping the L+R channels also -- note that in any comparisons with normal releases -- sometimes the DolbyA encoded versions have swapped channels, and I don't know why that sometimes seems to be true. Here are the demos (this is all 1960's/1970's stuff, I think pretty good for the timeframe): https://www.dropbox.com/sh/ab9nhtqjforacd8/AABvt7IYgoob7VXxpN0ekK6ra?dl=0 John
  3. Discussed this before elsewhere -- at the 44.1k-48k/16 bits level, other factors than just the sample rate/bit depth tend to be the problem. Even in the most extreme cases, assuming perfection in the recording itself, then 48k/20bits is all that could possibly be needed. At the very high SNR/low distortion/flat response/perfect transient response in the hearing range (Gibbs has been discussed elsewhere), things other than the media data rates tend to be the limitation. (I know that most of the pure techies might agree with the general statement -- maybe not the details.) IMO -- the biggest defect that the CD rate is good enough -- I do believe that 48k is really better. Also an extra bit of bit depth might help a little for those who have very very good hearing. No matter what, CD is darned good -- especially for REAL recordings. I am sure that there are a few specialty recordings that might approach the CD limits of actual quality -- VERY FEW recordings. I also believe (know) that 96k/24bits is extreme overkill for listening purposes -- creating more burden on a system/memory-disk space/etc than necessary. Admittedly 96k/24bits feels good. 44.1k/16bits is good enough and for all practical purposes perfect for listening. (Again, I still claim that 48k is marginally better than 44.1k, but cannot prove it.) John
  4. I used my own cobbled together research engineering type stuff. I very very seldomly use stuff like Audacity, even though it is sometimes convienient for certain things. My DolbyA decoder is as advanced as ever produced -- does do amazing things to correct the distortions, even some from encoding. Software that was a 2wk project (for the audio processing portions -- usually it is GUI that takes the time) just cannot compare with carefully crafted material. The DolbyA compatible decoder isn't just a single band fake (or a sloppy multiband) like some other decoders available for purchase. This decoder is SUPER sophisticated and does mostly match as closely as various versions of DolbyA HW match (which is specifically not too difficult a goal.) Except for improved distortion characteristics, the DHNRDS DA decoder mostly sounds like a DolbyA. (My bragging results from a multi-year project, but my ego is just barely big enough to avoid Napolean complex type things. After a fairly competent engineer sucessfully develops a difficult project like this, he/she deserves a bit of self congratulation. 🙂) Even if the decoder is bad (which all feedback that I have received says that it is very good), I would self congratulate because I tried very hard. The mp3 encoder that I compared the results with 'lame', which I have heard is one of the better encoders. mp3 using that encoder left a lot to be desired even at 320k. * I do have to admit that sometimes the DHNRDS decoder has produced especially difficult material to mp3 encode because of suboptimal quality -- and erroneously producing difficult-to-encode audio. I don't think that this instance is such a case -- but at least I am trying to fully inform about possible mistakes. I will try to post some comparisons today or tomorrow. I found another quality optimization in the DolbyA decoder (also busy doing the security mechanisms -- my project partner tells me that the decoder (DHNRDS project) will be stolen otherwise.) For only 10dB,10dB,10dB and 15dB processing for each band, there are LOTS of mistakes that the hardware, software and basic design can make... This has been both a fun and frustrating project. I will do EVERYTHING that I can to put together a clean demo today. When I upload the results to Dropbox, remember that the default mp3 decoder used when on Dropbox is insanely bad. It would be unfair against mp3 to depend on the Dropbox decoder. John
  5. geesh, I have been putting together a beautiful demo for the ABBACHAT group, and tried to encode it into mp3 format (lame, in insane 320k mode, with all of the checks enabled, and carefully limited bandwidth to avoid any kind of possible artifacts.) Mp3 sounds kind of bad (actually fairly bad, definitely not great.) The material that I can produce is SIGNIFICANTLY better than mp3 (with 40yr old ABBA), let alone the more recent studios that could totally blow it away. (I did careful processing & mastering -- much better than any demo that I might have done here.) Mp3 might be okay for cars or listening while running, but it is really not all that great IMO. And I do NOT have 'golden ears' at all. John
  6. You are dead right about the spatial image (e.g. stereo image.) Compression flattens the image -- that is probably the second major improvement (other than frequency response) that proper DolbyA decoding does. I would place correcting the stereo image being more important than the HF compression itself. (Even though, the HF compression is the biggest cause of the damage to the spatial image.) However, some people DO like compression, and a small amount of JUDICIOUS compression can make material more listenable FOR SOME PEOPLE. I generally dislike the sound of compresison myself (and I have written a free-to-use compressor -- MANY TIMES/KINDS/VERSIONS.) There is a free one on Hydrogenaudio FWIW. John
  7. I am NOT claiming that it sounds exactly like the original SACD, but the difference between the SACD and the direct decode from a DolbyA copy is extreme enough that the differences are still shown. mp3 doesn't do enough damage to obscure the difference, but does damage nonetheless. I am SUPER familar with mp3/opus/etc damage (not familar with the damage caused by AAC -- but at least I am honest about it.) You would NOT believe how clean the direct, 24bit flac 96k immediately decoded DolbyA copies sound. The sound is SO clean that the damage from mp3 is much greater than with the SACD version -- because the direct decode starts with cleaner audio. The temporal damage from mp3 is pretty severe (a chorus merges together too much -- lost detail.) It is more difficult to hear the damage when starting with lower quality material (like traditional commercial copies -- even the Carpenters stuff from HDtracks.) Lower quality material tends to be fuzzier, over-mastered, etc. (The temporal damage using Opus seems significantly less, but there is a graininess that I am not really sure what is going on.) . John
  8. Effectively it seems like a quality tradeoff (probably mitigated by some games) to get more bits for DRM type stuff. There is NO benfit to the end consumer. Anything talking about games that improve the quality to compensate are secondary to the actual DRM purpose. There certainly is little/no benefit to a marginally effective lossy compression scheme. (I mean margainally effective in the sense of saving space.) Any loss of quality is not a good thing anymore -- when it can be allowed, then we have mp3/opus/aac/etc. John
  9. You must not be listening to DolbySR or DolbyA? They both have 'kinks' -- you can even hear them in the remastered HDtracks Carpenters album -- they are deep down because of the significant compression, but you can still hear the kinks. (They have a more rounded sound because of the subsequent processing.) The kinks are more obvious on the non-remastered material though -- because that material tends to be more clean. If you are interested, I can cobble together an example -- doesn't even challenge the high end reproduction, it is apparently a burst of IMD or perhaps the uncompensated timing dither (it is more of a dither, it can be compesnated if the processing is sophisticated enough.) One form of the kink happens when the calibration is set too high (that means the gain is too low), then the expansion curve misses part of the needed processing. That is certainly ANOTHER kink, but it seems like the kink is very common on older material. Maybe, just maybe it is a distortion problem on the earlier equipment -- all I know, it is there and not dependent so much on the playback equipment. This is not the worst case -- but listen to the sibilance -- disturbing. (BTW, that album IS dolbyA encoded, so I didn't even decode it to eliminate the variable.) https://www.dropbox.com/s/p829w8ts1oc60sq/stutter.mp3?dl=0 John
  10. My own 'problem' is accurate sibilance. It is so hard to find (especially on older noise reduced material.) Sometimes the NR systems just couldn't handle sibilance very well. Too often, the sibilance would have a 'kink' in it, or something seems time-retarded. On the other hand, I do happen to like the overly sweet sibilance sound, as long as it isn't overbearing. The sweet sibilance might be intentional, but the kinked sibilance is very likely due to NR system or poor compressor. (Kinked -- my term -- sibilance is akin to a kind of lsip.) Bass -- I can take it or leave it, as long as it isn't repressive/overbearing. Much material from vinyl are cut below about 40Hz (there is still stuff down in the lower bass, it is just attenuated to make the material fit on the record.) John
  11. When I hear about 'audio guru this' or 'audio guru that', I don't quite understand. All too often, I hear nonsense from people with that label. I am not claiming that they are always wrong, but there is a lower signal to noise than should be. Everytime I hear about 'audio guru said', I start thinking about 'Gibbs ringing' or just thought about 'pre-ringing' as another interesting misunderstood thing. (Linear phase DOES sometimes sound different from minimium phase, but often the diagnosis is a little off.) On the pre-ringing thing, it is more accurately that some frequencies are delayed differently in the minmum phase, and sometimes might sound better. (When dealing with filters other than linear phase in complex signal processing -- it can be much more challenging... When signals are split with linear phase,, and then re-assembled, there is no real difference other than delay -- so there is really NO ringing. It is more of a difference sound and sometimes Gibbs effect -- NOT Gibbs ringing. Most of the time, the 'ringing' is actually removal of expected signal components, thereby causing a residual near-sine-like wobble in the signal.) Point being -- beware audio 'gurus', you really want to hear from EEs with long term expertise in DSP (actually writing code and/or designing -- NOT COPYING -- circuits.) They might agree with an effect looking the same -- but the manifestation might be from a place different than common knowledge. John
  12. Just realized that there might have also been a defect in my comparison -- the copy that I produced was NOT mastered. It was just decoded from a fairly close to master tape. By the time that the material gets to vinyl it might have some attenuation on the high end. I noticed on some ABBA stuff about -3dB in the 15kHz range. It might vary from recording to recording (or album to album.) It seems like a lot of albums are done all at once --- probably the reason why the average/peak levels of each song is sometimes different. Also, some albums seem to have one or two of the songs 'corrected' because of an unexpected quality to it. On my album 'original', I just now, experimented with a simple 2nd order lowpass at 17.5k with a Q of 0.707, and it probably sounds closer to what most people might expect. By doing the lowpass, NOTHING is really missing, but it does remove some of of the intensity. I don't know why they do the intense HF, other than actual DolbyA HW tends to mute the high frequencies because of some kind of HF distortion -- kind of like tape itself, and recording engineers might have tried to compensate for it. My decoder is very faithful to what is in the recording. I found out that one of the natural distortion mechanisms doesn't appear to be so much IMD, but rather a kind of uncompensated timing non-random-dither in the attack/decay curves. I don't know where it comes from -- might be the delay in the feedback loop when encoding. My decoder does compensate for that wobble -- lines up the highs (in a time sense) more accurately. The decoded material is probably as close to what went into the DolbyA encoder as been heard in 40yrs. That does NOT mean that it produces the expected sound -- because the muffling is what is expected. I have also gotten used to non-mastered material. (Not all material has been modified, but some material DOES sound more normal with a bit of massaging.) In some cases, the actual mastered material does sound worse (IMO), and sometimes it sounds better. Lately, there has been a habit of improving the competitiveness by compression (decreasing the peak-RMS ratio), allowing the materail to be played more loudly given the same amplifier power (or modulation of a transmitter.) I agree about compression though -- -it is not necessarily bad. I *do* think that it should be avoided if possible, and should be used in careful amounts. (No loudness wars by using broadcast processors on recordings -- like appearntly 'The Complete Studio Recordings' from ABBA. That was a VERY destructive thing. Sounds 6-9dB louder -- kind of amazing, actually.) John
  13. The DolbyA version IS the original version. The original version was DolbyA encoded. (In fact Karen's vocals were enhanced -- which give that overly 'crisp' sound. Some of the earlier albums were NOT as enhanced, you can tell by listening to the original recordings.) The SACD version IS the compressed version (or at least, somehow processed beyond DolbyA decoding.) This is what I was commenting on -- the preference for the ADDITIONAL compressed sound. This is now a common thing -- I am not judging either way, just interested (that is why I was confused.) Even people who like audio also like compression from time to time. Here is the undecoded version from the album (I wasn't really trying to talk so much about DolbyA decoding -- notice that the peaks are the same as the undecoded version but also the more extreme overly crisp sound). I happened into DolbyA encoded versions of the Carpenters albums -- that is what I am starting with. Even the normal vinyl might have been further processed -- that is called *mastering*. The SACD version surprised me -- I had expected an excellent DolbyA decode, not an additionally compressed version. HDtracks remastered versoin is similar (more compression). https://www.dropbox.com/s/k0t4grg530znywy/SuperStar-undecoded.mp3?dl=0 John
  14. My version derived directly from a copy of a DolbyA version is the first one (natural.) The 2nd one to me sounds compressed and distant. Each person has their own taste, but it seems (to me) that the general taste is going towards a compressed sound. The natural version is about as 'naked' as the recording gets. (You can actually hear the distortion from the original -- the first version above -- in the remastered HDtracks version. They apparently, for remastering the HDtracks version, only had the distorted DolbyA versions to begin with, and simply added some compression/accompanyment/EQ to the original.) * I was initially incredulous that the SACD version was so very far from the original recording... This general taste for compression is the reason why I asked the question. It is interesting to me -- I actually used to like compressed sound, but compression can almost never be done without distortion. Now, since I am working on audio processing -- I have become hyper-sensitized because I know *exactly* how to listen for the gain control distortion (it is all part of the audio h*ll thing.) The real problem with the orginal (not-compressed-additionally) stuff is that ALL of the original distortion and vocal enhancement is obvious. Sometimes it seems that some compression helps to take the edge off of the sound. Sometimes, mellowing out the sound with some EQ or compression might be helpful. Hearing the super clear, raw DolbyA decoded versions of Linda Ronstadt (stuff like 'Just One Look') using the Aphex distorter (keep forgetting the real name) is a real trip to audio hell. Potentially really clear, nice sound -- all mangled by phase/time corruption. John
  15. With the 'CD' player that you mention -- you JUST MIGHT be manifesting the same syndrome that has been driving me batty for the last few years... My hearing changes vs. time of day/listening duration/fatigue/mood/etc. This makes using my hearing very tricky when using as a measuring instrument. I have made NUMEROUS mistakes when trying to use my hearing for measurement (esp in A/B comparisons with ANY time delay.) Point being -- in my own case, practically any of the changes that I hear over time result from changes in my own hearing, not changes in my equipment. It was incredibly frustrating to solve 'equipment' or 'software' problems until I finally had realized - the actual problem was the variability in my hearing. One especially frustrating situation -- my hearing can become sensitized to certain kinds of distortion, which APPEAR to get worse and worse as my hearing trains more and more to listen for the defect. Then.... All of a sudden (like this morning), everything is sounding perfect, and feeks like I couldn't discern 1% distortion right now. The sound from my project is SOOO VERY BEAUTIFUL -- until probably later on today, then it will sound like h*ll. (Actually, sounding the same in reality.) This is another data point that supports my SUGGESTION (and suggestion only) to try to focus on the enjoyment, not so much the perfection (or not.) If someone wants to join me in my listening h*ll, one last suggestion -- avoid joining me, but you are very welcome to vist -- with regrets. John
×
×
  • Create New...