Popular Post John Dyson Posted March 24, 2021 Popular Post Share Posted March 24, 2021 Here is an interesting datapoint. This is about HF hearing, not about the software that I am working on... I don't know what this means, but before making any assertions -- my hearing is blind to tones > 14kHz. Note exactly the language that I am using -- 'blind to tones', not cannot hear >14kHz!!! During testing, I found that I made a mistake on one of the equalizers in my SW -- caught by hearing an anomaly in the sound... I was wrongly using an equalizer that is an HF shelving filter of 21kHz with a rolloff of -3dB. (Yes, such small details are important.) Instead, I should have used 18kHz, with a gain of 18kHz/21kHz.or -1.34dB. ( I happened to choose the wrong filter spec because I was not thinking clearly, and should have chosen the 18kHz EQ to begin with...) Even though when considering the normal hearing rolling off at 18-20kHz, and my hearing rolling off at 14kHz, I could hear a profound difference between the 18kHz, -1.34dB 1st order EQ vs. 21kHz, -3dB. The 21kHz, -3dB rolloff seemed to leave more highs in the sound, but the highs seemed distorted. These kinds of distortions or 'tells' are why I can do EQ in certain cases in frequency ranges where I cannot hear well. Oddly, this scheme doesn't appear to work at freq <1kHz or so.) Now -- about this strange HF hearing. A 1st order EQ of 18kHz, -1.34dB has a small effect at 14kHz, and even at 14kHz, I my hearing of tones is more weak than at 10kHz. The difference that I am hearing is more significant than the sub-dB change that technically is happening in my normal range of hearing. This is a guess, and for EEs they will understand this hypothesis: I believe that my hearing, blind at 14kHz is sometimes getting the effect of parametric amplification, perhaps the nonlinear effects at frequencies above my normal range of hearing. So, with the co-incedence of energy in the HF region, a kind of parametric amplification or biasing of my hearing can boost my ability to hear transients above 14kHz. I don't think that I can ever hear 'tones' at 18kHz, but I am hearing some kind of distortion as a result of co-incedental energy. This biasing/parametric effect thing is the REAL reason why analog tape bias works. It makes me wonder if the same kind of thing is happening, but with a different result or purpose, for hearing >14kHz? For those into 'measurements' -- this might be an interesting project, and one reason why some 'golden ears' honestly complain about differences when BTW is wider/narrower than 20kHz, yet most of 'em cannot hear much above 18kHz or even less.... I really don't think that this is much about 'filter skirts', but might be a contributing factor. Also, there can be issues associated with time delay (e.g. high order analog filters or not-linear phase digital filters causing time delay differences.) Just an observation -- but also a helpful odd artifact of human hearing (or at least, my own human hearing.) ADD-ON: I noticed the 'distortion' on processing the Supertramp recordings this morning, but heard it again when processing 'Olivia's' recordings. I did a review and found the bug described above. Once the 'bug' was corrected, the distortion went away. Another ADD-ON: This has manifest on different headphones & different situations in the past. I just finally realized that this might be an interesting issue for someone to research. I am sure that at the levels that we are speaking, this isn't a headphone nonlinearity matter, and I am using high sample rates in the 88.2k/96k range for the ultimate D/A conversion. sandyk and Josh Mound 2 Link to comment
John Dyson Posted March 24, 2021 Author Share Posted March 24, 2021 41 minutes ago, pkane2001 said: John, the simplest explanation is that you're just hearing IMD. Energy at or above 14kHz can be significant on some recordings, so you don't need to hear those frequencies, as they add and subtract with frequencies below 14k. THANKS FOR THE REMINDER: I thought about IMD also. but that would have to be one heck of a lot of IMD if at 18kHz/21kHz directly in the local analog/digital circuit. Note that the 'distortion' difference that I am hearing isn't the result of significant signal level difference either (both EQ result in similar levels), and the energy at those frequencies shouldn't be high enough to cause significant IMD. I was hearing a PROFOUND amount of distortion -- but read further -- I was reminded of what was really going on. There is ONE thing that it could plausibly be, and it COULD be IMD, but not generated directlty by the signal at 18-21kHz. There is a vast amount of anti-distortion phasing in the signal procesing on the decoder, and if the phases and cancellation isn't done perfectly, some of the latent distortions left over from the FA signal processing can be let through. I really doubt that is the source of distortion at the high frequencies in the local circuitry, but could be the modulation products created at the lower frequencies, in the mutiple dynamics processors needed to produce FA and the phasing at the lower frequencies didnt' cancel as completely. So, I didn't initially think about IMD because of the energy levels at 18kHz are so small (not enough to directly create noticeable IMD at lower frequencies), but there IS already a lot of IMD in the FA signal, and the incorrect EQ at 18kHz could cause more profound phase effects at lower frequencies than the effects on levels. A senior moment where I forgot about an important feature in the decoder that appears to exist nowhere else. The phasing matter has been a major improvement over a straight-forward design with simple EQ. But, I just didn't think of it in this specific instance (again, transient senior moment) In fact, the EQ being done at 21kHz and/or 18kHz had the anti-distortion phasing in it also. This just might have been a senior brain fart about what was really going on. ADD-ON: explaining where the phasing modulation is happening... Think about multiple dynamics processors with a dynamically changing phase, but instead f the output of each unit being hand-in-glove decoded, the processing goes further on to another step. The anti-distortion cancellation does an approx 1st order cancellation of the PM that is happening in the signal. So, the audible effect is anti-IMD when actually it does significant cancellation of the PM. This happens at each step in the process. The major effect of the anti-PM EQ when it comes to frequency response is pretty much nil, but the effect on phase is profound, and along with the anti-MD code, makes a MAJOR improvement for distortion reduction. I should have remembered -- but I get tripped up once in a while, then get locked into wrong-headed thinking for a few days. I REALLY WISH I HAD A CO-WORKER ACROSS THE DESK LIKE I USED TO... I was a member of a sub-team of two people at AT&T Bell Labs, myself (the crazy person), and the exquisitely sane person who was mildly on the 'spectrum'. Both of us together created a superhuman, but by myself, I am incredibly mistake prone. I really love the good old days!!! John Link to comment
John Dyson Posted March 25, 2021 Author Share Posted March 25, 2021 (sorry about not properly quoting -- I am a very primitive user of GUI stuff. I still think that there is the *possibility* of SOMETHING going on about the parametric effects. I am not fully convinced (NOW, after this recent experience) that there isn't something usefully nonlinear in the auditory process. My current position is still about the engineering norm right now. An experiment that eliminates the variables, also considering the parametric effects, SHOULD be done. Most of the time, the thinking is limited to simple nonlinearity instead of the more dynamic parametric nonlinearities. However, for now, I am still a believer in the 20kHz orthodoxy. My mind is open for other things though -- once there is true, more than anecdotal evidence. Parametric amplification usually entails some kind of IMD, but isn't the typical audio IMD, and the output from the desired signal can be greater than the input (given the amplifier or mixer is pumped by a signal, like tape bias, for example.) However, I also agree with @pkane2001(in a circumspect way) tabout my issue, hat there are other things that are going on -- I temporarily forgot them. Some day, I'll do some demos after turning off the 'special sauce' entirely. (The 'special sauce' is a bunch of distortion cancellation stuff.) The difference is extreme. The trivial, straightforward EQ sounds significantly worse than the circuitous EQ that I do for the 'sauce.' The cancellation EQ is substantially different than the norm, and is a bit non-intuitive. Since the special sauce works so well, I quit using a special switch to disable it, and the mode can only be disabled in legacy parts of the code. This special EQ makes me think that there MIGHT be equipment that doesn't sound as good as it should -- because the 'trick' isn't well known? Some actual measurements might also be useful in the future. Link to comment
John Dyson Posted March 25, 2021 Author Share Posted March 25, 2021 1 hour ago, firedog said: Because hearing also involves processing in the brain. It's not just your ears. That's why I can still (with typical age compromised ears) still tell mp3 from CD. Many younger people can't. It's not their ears, it's their brain, which hasn't been taught the difference, so they don't notice it. You hit the nail on the head.... (pun intended.) We can accomodate a lot of things about what we hear. Some of us make hiss go away, some can actually ignore the FA compression, some people's hearing adjust down their bass or treble sensitivity when there is too much. There is a 'kind of' agc going on in our hearing system. Likewise, when we start missing the highest highs, or the lows disappear because we have troubles hearing those frequency regions, there is no real indication that we are missing those Our brains adjust to the fact that we don't hear those sounds any more, and all seems normal, but not really. I used to be very sensitive to frequencies well above 18kHz, and moderately strong signals above 20kHz, but now I have troubles starting in the 14kHz region. I don't feel like I am missing anything, but I really am. I can also sometimes hear the temporal distortion caused by MP3, but in my case, I can only hear it on already defective material that goes beyond normal audio signals -- and that hint tells me that other people with more acute hearing in that area just might be able to detect mp3 in less extreme conditions. John pkane2001 1 Link to comment
John Dyson Posted March 27, 2021 Author Share Posted March 27, 2021 54 minutes ago, lucretius said: I would love to have been there for the blind testing. 🙂 About the 'age compromised' ears... Sometimes my hearing goes berserk (as you have seen me complain before), but oddly -- SOMETIMES when my hearing goes defective, I seem to be better at hearing defects that I couldn't easily hear otherwise. I will NEVER make strong claims about always being able to hear differences between different recording formats, but IN SOME WAYS, 'age compromised' hearing might actually make it easier to detect certain impairments in a recording?!?! (punctuation means, not sure....) lucretius 1 Link to comment
John Dyson Posted March 27, 2021 Author Share Posted March 27, 2021 6 hours ago, davide256 said: mmm, don't experience that with mp3 any more than I do with PCM rates...improvements in source to DAC chain are what have gradually whittled my digital fatigue down But prefer to only listen to mp3 if its music I don't know well or fairly simple piece; seems like a lot of musical detail gets blenderized/lost if I play familiar complex music in reduced data density mp3 format on main system. Its like settling for generic vs artisan prepared food, obvious quality reduction. You must have better hearing than I do -- but that isn't very difficult to achieve nowadays :-). However, that almost 'blenderized' sound is a good way to describe the temporal distortions that I hear on MP3, WHEN I CAN HEAR THEM. More often than not, especially on less dense material, at most I detect only a purely subjective and probably non-repeatable difference. Where I do hear a true difference with MP3, again -- very seldom a casually noticeable difference -- is on complex material with intense detail. I have NEVER detected anything more than a 'feeling' (unrepeatable) difference on non-dense recordings. (Hear it more on moderately intense pop-rock, less-so on typical classical recordings.) For casual listening, I hear NO difference, but again, my hearing is much poorer than when I was 30-40yrs old or younger. I respect the amazingly acute hearing that some people DO have. Link to comment
John Dyson Posted March 28, 2021 Author Share Posted March 28, 2021 4 hours ago, jabbr said: Your surprise suggests that you are assuming your hearing is linear. There is every reason to believe that human hearing, like every type of neural processing, is highly nonlinear. There is no way to characterize your nonlinearlity in a mathematical fashion without extensive testing, if its even possible. I didn't suggest that hearing was linear, but 'someone' is forgetting about masking also. Actually, with additional tones, hearing should become MORE blind to lower levels assuming the log relationships and masking. I am hearing the opposite. Also, the effect that I am hearing is more like 'pumping' or 'tape bias'. (which is what I had stated earlier here or somewhere else). which is NOT linear and also NOT what one would normally expect. (Masking is used to hide some of the defects in mp3 encoding, for example.) Again, NO WAY did I suggest that hearing is LINEAR?!?!?!? Where did that come from? Link to comment
John Dyson Posted March 28, 2021 Author Share Posted March 28, 2021 4 hours ago, The Computer Audiophile said: In addition, frequency sensitivity depends on sound levels. At lower levels, our ear sensitivity in the bass and treble areas, drops significantly. We are more we have wider response for loud tones, but we become more blind to tones nearby -- that is, masking is what should be happening, but I am hearing 'pumping' (an effect similar to tape bias). ADD-ON: Whoops, I forgot about the term 'pumping' in the audio processing sense, because I have my EE/math hat on when I am speaking up 'pumping' here. Best way to describe it -- look up 'pumping' in the sense of parametric amplication. THAT IS DIFFERENT than AGC pumping... Sorry if there is confusion in the usage of terms -- seems like we re-use words in odd and confusing ways at times. Link to comment
John Dyson Posted March 28, 2021 Author Share Posted March 28, 2021 10 minutes ago, jabbr said: You have an assumption that your hearing of a complex signal should be predicted by pure tone response, or that masking should behave a certain way etc. * I won't mark 'disagree' in the previous message, because sometimes that also suggests 'disapproval', which I do not. I believe that there is some misinterpretation going on here. Interpeting what I was trying to say, in less technical terms... Pumping a nolinear system can result in amplification, it can also just result in distortion -- but the amplification side of things is what I think that I was seeing. I don't think that I was really specifying pure tones, and if I was -- then I didn't intend to. (Parmetric amplification can be a lot like using an AC signal for a 'power supply' instead of 'DC'... Well, in a way.) Paramps were used in early microwave receivers because the gain wasn't really based on a resistive type transconductance (noisy) but instead of a reactive kind of transconductance -- lower noise. Anyway -- my vision of things is somewhat complex, so communicating is sometimes tricky -- do I start writing more deeply technical things in my long paragraphs that are already too boring, or try to explain in more reasonable terms -- perhaps taking shortcuts? Link to comment
John Dyson Posted March 28, 2021 Author Share Posted March 28, 2021 19 minutes ago, jabbr said: As an engineer you are tempted to analyze the human hearing system using tools in the engineering toolbox. We can analyze an electronic circuit pretty well using SPICE. The field of computational neuroscience is an attempt at understanding the brain in a similar fashion. Humanity isn't there yet. Thats for the normal human brain. You aren't normal. Elon Musk's Neurolink aims to record from 1000 neurons simultaneously but 100 billion? I use Spice when I need it. All too often, if I need to emulate a nonlinear LF circuit, I can do it straight away in C++. Spice doesn't do the kind of parametric nonlinearities like paramps very well. It can, but not so good. I used to buiild my own spice 3e2 back in the 80's, but I don't jump on spice every time I run into something. A quick run for an amplifier of some kind, or a lower frequency nonlinear circuit - it is great. For higher frequency stuff -- it is hit or miss. However, it is super-duper good for a 1st order design, getting an idea of what is going on with things. For using Spice for other things, like the DA decoder, you need reasonable models. There are NO models for the selected components, so it is a cr*pshoot. Instead of worrying about spice modeling, I simply translated the circuit into C++ software (easy for me -- been doing semiconductor circuits since the '60s) -- and then did a bisection scheme to find the correct parameters. Frankly, I'd have to desolder a few of the diodes to get an idea of its magic characteristics for direct modeling (is it a low or high is diode?) Obviously, they have good exponential characteritics, because that is a critical design feature of the attack/release circuit. (I did a straight exponential model -- it worked great, but curve-fitting was a b*tch, becuase there are 3 circuits where the diode parameters are criticial, and they interact!!!) Reallly messy. Also, imagine modeling (fully) a DolbyA HW unit -- yikes!!! Need-a-supercomputer time -- that is, if you have the models for the parts (which I NOW DO, after lots and lots of work -- I know some secrets!!!) :0> yea, that'd be fun -- run a Spice emulated DolbyA unit to decode commercial recordings -- YEA -- that's the ticket!!! So -- spice is really good for certain things, but there are better packages for certain nonlinear analysis -- e.g. for parametric nonlinearities. IT CAN DO IT -- but I don't have any projects like that right now. However -- Spice can be a lifesaver when a circuit might be hard to conceptualize... John Link to comment
John Dyson Posted March 29, 2021 Author Share Posted March 29, 2021 11 minutes ago, semente said: John, what speakers or headphones are you using for monitoring your work? Speakers are an anathema for the kind of subtle distortions that I have to listen for. I don't even care about speakers -- I just don't like speakers for the highly technical work -- but for listening, they are great. I use the most evil headphones available and selected for-purpose. Beyerdynamic DT770/80. You might claim -- there are prettier sounding headphones, yes that is correct. Take a look at the characteristics and what I am doing -- the DT770 is perfect. A peak near the 7-10kHz range, and bass down to DC with almost no rolloff. This gives me a picture of two of the critical parts of the spectrum. OTOH -- I found that I need 'pretty' headphones that are closer to what people normally use. I am thinking about some Sennheisers with a ltitle more flat HF, and the nice LF rolloff that keeps the air pressure from hurting my ears. (My LF hearing disappears, but I can still feel the air-pressure.) When headphones have response that go down as low as the DT770, one needs to be VERY careful about the LF EQ. This does give me an indication of too much LF, but also takes away from the enjoyment when the flat LF sometimes causes some misjudgment. So, I am NOT mastering -- I am hunting for 'tells' in the signal that show that there are messed up joins across the frequency range. Until I gained this experience, I didn't even realize that there is a LOT to EQ that I was clueless about. I had only encountered multi-pole EQ, but most of this DolbyA and pro-stuff is single pole, and it acts WAY WAY WAY different than multi-pole. There is no concept of adding bass by increasing the single pole response below, say 50 or 100Hz. The single pole EQ is about attaining a flat response without any 'tells'. If you want to EQ and master for listening purposes, except for some very limited cases, 2nd order, parametric are the best bet. I have added some EQ modes to the decoder that change the 'balance' between the highs and lows, but not 'bass boost' or treble boost/cut per-se. BTW, the decoder has some really high tech 2nd order EQ with the anti-distoriton attribute. I doubt that most people would want to dial in the EQ forumulas that cancel distortion artifacts -- and it really DOES work. With my 2nd order EQ technique, it is so easy to control 'rough' sibilance by even a 0.375dB tweak. (It is about RELATIVE levels, you boost/cut.) OTOH, sibilance IS a problem on some older recoridngs -- Simon & Garfunkel recordings DO have too much sibilance (as mentioned by someone directly hearing the tapes, and myself dealing with them.) So, there is some dynamic automatic parametric EQ going on if you ask for the anti-sibilance. (There is also a fixed ant-sibilance -- mostly useful if there was a mastering peak in the recording.) SO -- speakers -- NAY. They are mostly an anathema to 90% of what I need to do. I even now know the tells for stereo image issues -- and since I am only replicating a master tape, and not changing the stereo image -- I just don't need speakers in my living room or bedroom. (Sorry, I know that it is disappointing, but if I did get some speakers, I'd get some powered monitors -- but without mastering, why bother?) QUESTION: do you want the decoder adjusted for my listening environment. WIth the headphones, there is a direct coupling - no room effects, not dependent on seating position. The headphones DO have personality, but I have quantified them. semente 1 Link to comment
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now