Spatial and Dynamic Information #431
Replies: 4 comments 1 reply
-
|
You are mistaken about how lossless codecs work. What you describe is essentially how lossy codecs work. Lossless codecs do not discard any information. Saying that lossless codecs alter the sound of the compressed audio is equal to saying zipping a PDF of a Shakespeare play and then unzipping it again alters the text due to the zip algorithm dropping words which it considers redundant. If you hear any difference between lossless codecs or between different settings for the same lossless codec, your playback or listening equipment is flawed or broken. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you Robert,
It is the perennial dispute between mathematicians and musicians. There IS a
difference and, if you use good equipment it is clearly discernible. I am not
alone in thinking this, some of the best audio engineers in the world agree
with me. Of course, the differences are small and, to most, inconsequential.
If there were no difference, why do we have so many different algorithms? Take
a high quality original recording, encode it with flac 1.4.2 and APE then play
back with high quality pro headphones. You will definitely hear a difference.
As I regularly make high quality original recordings of live performances, I
am lucky enough to have really good files to work with and I have been
experimenting for some time. On my website you will find the SilkPad
encryption programme. I wrote several versions of this before settling on the
current version. I understand the mathematics. I was also a pioneer in the
field of biometrics and developed several interesting concepts here and wrote
several books for Springer on the subject.
I was curious as to if and how people would respond and I truly thank you for
your comments and admire the work you have done with fre:ac over the years. I
wish we were more local, then I could show you these differences. I am unusual
in being both a composer, audio engineer and software engineer, so I have a
different perspective. I also worked in the BA IT dept for 20 years, on
everything from mainframes to PKI (which I have also recently released a book
on, published by Taylor and Francis) so I do understand how these things work.
Check out also; https://link.springer.com/book/10.1007/978-3-030-62429-3
which is a gentle introduction into the art of recording. It has become
unexpectedly popular.
Take care,
On 29/01/2023 at 21:40, Robert Kausch ***@***.***> wrote:
You are mistaken about how lossless codecs work. What you describe is
essentially how lossy codecs work. Lossless codecs do not discard any
information.
Saying that lossless codecs alter the sound of the compressed audio is equal
to saying zipping a PDF of a Shakespeare play and then unzipping it again
alters the text due to the zip algorithm dropping words which it considers
redundant.
If you hear any difference between lossless codecs or between different
settings for the same lossless codec, your playback or listening equipment is
flawed or broken.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
This may be of value to explain the perceived audio inconsistencies of lossless compression; https://youtu.be/IIVx7Hlugq4?t=111 For others that are interested, this video is also informative - and may help to save disk-space and potentially some needless disappointment due to misunderstanding the technical details of the audio source format 😃 |
Beta Was this translation helpful? Give feedback.
-
|
By the way, if you can find some of the excellent Decca recordings made in the
early 60s using variations of the 'Decca Tree' microphone technique, you will
have some good material with which to experiment. There are also some very
good early digital recordings made by Denon and JVC. Denon were slightly ahead
of most.
If you can get to Berkhamsted for one of the Berkhamsted Jazz concerts, I will
make you a true stereophonic file to also experiment with. Alan Blumlein
worked all this out in 1928 (before tape recorders) because he understood the
physics of sound (as I do) and that much of what we hear is not obvious. There
are probably 1001 papers on psychoacoustics but we don't need those.
If you really study the maths on the lossless codecs, they DO discard what
they consider to be mathematically redundant. That is the issue, our certainty
in mathematics only reaches as far as our mathematical understanding. We don't
understand everything. Another book you should read is 'In Search of Nature'
where I identify the gaps in our scientific understanding and address the
stand off between science and religion. You would enjoy it.
Take care and thanks, as always, for your work on fre:ac
On 29/01/2023 at 21:40, Robert Kausch ***@***.***> wrote:
You are mistaken about how lossless codecs work. What you describe is
essentially how lossy codecs work. Lossless codecs do not discard any
information.
Saying that lossless codecs alter the sound of the compressed audio is equal
to saying zipping a PDF of a Shakespeare play and then unzipping it again
alters the text due to the zip algorithm dropping words which it considers
redundant.
If you hear any difference between lossless codecs or between different
settings for the same lossless codec, your playback or listening equipment is
flawed or broken.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID:
***@***.***>
|
Beta Was this translation helpful? Give feedback.
-
|
Where to begin? Much to address. For context, besides a long-time (decades) interest in high-fidelity audio capture+reproduction (I'm wary of using the audiophile term; it's become loaded and associated with pseudo-science quackery), and having become quite familiar with digital audio, I was also an AV tech in a former life. I'll start with the simpler remarks, before the deep-dive.
Analogue audio/electronics are, at best, imperfect (and at worst, messy). Quality (well-designed & built) equipment makes the difference. For clarity, I absolutely do not mean various silliness like vacuum-tubes/valves (no, thanks; transistors, please), or other gimmicks. You'll also want to ensure that your cables and interconnects are of quality, from a reputable manufacturer & supplier. I highly recommend Blue Jeans Cable; no-nonsense, science-based, professional-grade cables. I especially recommend the library of articles, which explain a great deal, and debunk many myths. Blue Jeans Cable also stood up to Monster Cable's patent-trolling (not to mention debunking the misinformation around it's silly over-priced nonsense cables (oxygen-free copper? WTF? there's none of that BS in the pro AV world, because the engineers are savvy enough to know that it's bunk) Another point about equipment; make balanced selections. By which I mean, there's no point in having a very expensive amplifier, if you're using it to drive crappy speakers. Same if you're trying to drive high-grade studio monitors from a crappy amp. Compare computer specs; unless you have a very niche application in mind, then there's little point in having one component be very high-spec, while another is low-spec. You're better off getting mid-range across the board to avoid wasting your money and ending up with silly bottlenecks which hinder the high-spec components from performing to their potential. Another general principle, is that you want to buy separates/components, and interconnect them. As apposed to all-in-one boxes. Various reasons, for this.
Modular systems are so advantageous, that I even designed & constructed (from major component assemblies) my own mains distribution electrical cabling, to be able to combine them to make any arrangement of length & socket-count I wished. I can elaborate, if desired.
I can only agree, entilely, with enzo1982. But, wait, it gets fundamentally worse …
What, precisely, is ‘true stereo’? Do you mean Real Stereo (as the original inventor defined it)? The modern approach might be binaural recordings (which, when played back through quality headphones, sound awesome).
Hmm, depends what you mean, precisely. You wanna be careful about avoiding phasing effects if they're cited too close together. If you're meaning real stereo (as opposed to merely 2-channel (essentially dual-mono)), then you'll actually want a moderate separation between them. Or, as I said, use binaural capture.
Erm, you'll need a pre-amp, at least, since a recorder expects line-level signals. Unless you're used to lower-end recorders which often include a pre-amp. Mic-level output into line-level input, will be inaudiable. That's what signal gain circuitry is for (gain is an increase in voltage, whereas amplification is more about an increase of current). Especially if using ‘high quality microphones’, which tend to be large-diaphragm, and need a lot of gain (like, 60dB, which is demanding for any pre-amp circuitry). But, back to what I said about interconnecting separates; so that you know what pre-amp is being used, what it can (and can't) do, how it's configured, and otherwise have knowledge+control over the quality (or other properties/characteristics) of equipment, and how the signal path is being affected. While a mic, pre-amp, and recorder would be the absolute minimum to function, here; at least in the pro audio world, this was never done (and the type of work I did, exposed me to a wide variety of situations). You'll want some degree of control, and thus some sort of console. A simple example would be in mic'ing a band; the drumkit; specifically (for this example) the kick-drum. For many reasons (especially in live audio situations), you don't want a simple open mic. Next to your mixing console, you'll have a (standard 19-inch) rack of other equipment. Among it, will be several gates, with connectors to ‘insert’ it into the signal path of whichever channel you have the kick-drum mic on. What's a gate? What does it do? Why have one on the kick-drum? I hear you ask. A gate is a (relatively) simple processing device which, essentially, controls when the source signal (in this case, from the mic) is passed through the gate (or not) based on the loudness/intensity/signal-level (yes, loudness, not volume; they're different, but that's another explanation), the threshold of which can be set by a variable control (if I recall, typically a potentiometer). Thus, what you want for the kick-drum, is that when it's being kicked, and the level is relatively high, for the signal to pass the gate, but after it decays back toward silence (well, quiet at least; between the booms of the drummer stamping on the pedal) for the gate to close and output silence to the console's channel (while you leave the fader at the general intensity level for the kick-drum (while the gate is open, at least)). Why? A drum-kit requires many microphones in close proximity. For being able to mix cleanly, you want clean separation of what's on each channel. Gates help to reduce cross-talk between mics. That is, when the kick-drum isn't booming, then you're not gonna get noise (unwanted sound) on that channel from any of the other drums, or cymbals. Only the boom of the kick-drum. This is especially important in a live performance situation, because feedback is your enemy. The more open mics, the more likely (unless otherwise mitigated/controlled) you are at risk from the lovely din that is feedback. Bad times. Eves in a conference type situation, with only several voice mics; while I'm listening to the audio of whoever's speaking (to ensure it's intelligible and otherwise sounds good, I'm watching the others who aren't speaking, for signs that they're about to speak (imagine a Q&A session; because if it's just a lecturer at a podium, then it's dead simple for the duration of his slide-deck). Why am I watching the non-speakers during Q&A? Because I have the faders of those mics at -5dB or so from unity, to reduce noise (no-one wants to hear amplified coughing, or similar) and to reduce the chance of nasty feedback (which will escalate very quickly; quicker than you can react, by which point it's too late and embarassment has already occurred; so it's to be avoided). Compare the concept of sound isolation in a recording environment. You may also want other signal management (as distinct from ‘processing’ in the sense of altering the nature/quality of the sound itserf (such as with an EQ; though there are legit management reasons for doing that, too; band-limiting the frequency range, for example; almost all mixing consoles have, per-channel, a toggle for a high-pass (‘low-cut’) filter, normally at the 100Hz mark, in order to minimise the noise of someone handling a mic, footsteps on the stage (on which the mic is ultimately standing), and so on; only specific things like a kick-drum will be down at 100Hz; voice (as a common, familiar source) starts several hundred Hz higher up, so strongly attenuating 100Hz and below is fine, and thus desirable for management purposes). Another reason you'll want other equipment, especially a mixer, is because there is no ideal microphone (flat frequency response across the entire 20Hz to 20kHz range of (unimpaired) human hearing). Hence, there are quite so many microphones; they have different properties, and thus different applications. So, entirely dependent on what you're recording, you'll likely want more than 2 mics. Sure, if you're simply wanting to capture what a single audience-member would hear, then yes. But, it's not that simple, and comes with problems. If you want immersion & realism (warts & all), then OK; but then you might as well do binaural capture. Often, literal realism isn't what's sought (but, clarity). Different sound sources are at different distances (and thus wildly different apparent loudness), some will obscure others, and I wont even get started about diffraction & acoustics based on the type of sound (largely frequently-dependent). A major sub-field of (pro) audio tech is sound reinforcement. So that all of the audience (including those right at the back in the cheap seats) can hear everything clearly. Because it's about actually hearing the speaker/band, rather than absolute realism (which, at the back, would include attenuation, noise (from the rest of the audience), echo, delay, and perhaps some other funky effects based on the mood of the techno-gods that day (or the phase of the moon, if you prefer), and multiple factors about the venue which you have little to no meaningful control over (I recall one venue, a popular hotel, which had the kitchens right next to the main event room/hall; yes, utterly daft; you could hear the clanging of utensils as lunch was being prepared; and this was not the worst horror that particular place). Strictly, if you want realism, then you'd be in an amphitheatre. There are reasons why we do things differently, today. Also, each audience member, ideally, wants their own performance (as if they're the only audience member). They could have that if they're willing to pay for it, but most can't/won't. But, we try to approximate it. So, those at the back of the audience don't want to be reminded that they can barely see anything (through all the other bodies), let alone would struggle to hear anything which didn't sound like a blurry noise off in the distance if going for absolute realism. Thus, we can give the illusion of something closer to receiving a personal performance, by at least making the audio clear for them, and either staggering the seating rows or having the performers up on stage, so that they've got a hope of actually seeing anything. Anyway; back to mics; let's imagine it's an orchestra playing classical pieces. Well, firstly, that's quite a large area to cover, with only a couple microphones. Remember the inverse square law (for sensitivity/loudness relative to distance). Different instruments will, to a degree, have different intrinsic loudness, compared to each other. Sometimes you want that difference, but sometimes a loud instrument can drown out a quieter one. The instrument-players will be keeping time with each other (and/or the conductor; I won't claim to know how that part works; how the sound is produced is a different matter than capturing the result), whereas the distance (and thus latencies involved) between them will be different than for the perception of an audience member (especially if you've got any reverb, or other echo/reflections; the Royal Albert Hall installed those inverted white domes/mushrooms in the ceiling, precisely to reduce this problem they're sound-absorbers suspended from the ceiling; so the reason for their non-uniform size and distribution should now be apparent to anyone with even basic knowledge of acoustics; different frequencies (because different natural resonance)). Then, back to that pesky problem of no ideal mic. Even a drum-kit involves several distinct types of mics (a cymbal sounds very different to a drum; different frequency-response). Vocals being different, again, to both. In something like an orchestra, the variety is even greater. And you'll want multiples of each type (because multiple instruments, and a large area to cover, if you want to capture all the complexity (harmonics, phasing, and the like) of multiple instruments playing together. So, you'll most definitely need a mixer, if only to combine many input channels into 2 output channels. Maybe the clearest example is with vocals (if you've got singers, rather than a purely instrumental recital). Strict realism would mean that they sound off-in-the-distance (even when up on the stage) to most of the audience (else, if you're down in the pit, then you've lost line-of-sight to the orchestra, and it'll sound muffled/distorted), and the direction of the source relative to a given audience member will vary as the singer moves across the stage. Not really ideal, or even close. Tickets aren't cheap, either. So, singers are miced. Either with individual radio-mics (if you have only a few singers), else floor-standing mics on the stage if you have a large cast (because only a few are singing at any one time, and your mixer has a finite count of channels, and you have only 10 fingers regardless of how many faders you have infront of you; so keeping mic-count sane helps a lot; don't forget about minimising cross-talk and the risk of feedback, either/too; it's a perpetual hazard). So, singers being miced (with vocal mics, different to instrument mics, hence why you see so many flight cases going in/out of the truck(s) after/before a given run of performances, because the amount of equipment, of various sorts, is staggering (more-so when you have to help shift it; most flight cases have wheels for very practical reasons). So, yes, singers miced; this almost eliminates traditional realism, but gains much clarity (speakers distributed wherever you want, for reinforcement (moderate loudness, evenly distributed, instead of a bank of speakers with intense amplification; because even at concerts, if large enough, there'll be multiple sets of speakers, at different distances, so the crowd at the back can still hear clearly (and the front-of-house tower is among the audience, so that the FoH techs can directly hear what the audience hears; instant feedback (in the constructive, control-loop sense) to whatever changes they're making through the show, rather than being hidden in some corner with headphones (somewhat like the mix operator for the stage-monitors (so that the performers can hear each other, to keep in time and otherwise play well together)))). Yup, tangents for days (I've got more). Clarity, because what good are vocals if people can't actually discern the lyrics (no commentary on genres like death metal, which to me is perverse because lead vocals is basically shouting into the mic; no wonder it sounds so distorted)? So, audience up high / at the back, can have a speaker nearer to them than the performers. They can still see (avid opera goers sometimes bring binoculars), since they can control optics, but they can also clearly hear the performance (they can't meaningfully control the acoustics). In a typical situ where the audience vastly outnumbers the performers (and the crew, combined), this is as close to the ideal (1-to-1 performance) as you're gonna get (for that price). It's the economies of scale which mean that the non-rich can afford to attend the theatre, at all. You'll likely also need various signal routing, such as if you also wanna record. You'll need to ensure that performers can hear each other (back to what I said about stage monitors, or equivalent; there are a few ways of doing it; the optical being in-ear monitors). Oh, and aside from mic distribution around the area, you'll likely also want them either directly on the instruments, or suspended above (to minimise noise from humans (including the audience), since classical instruments are relatively loud, and professional instrument players know to be quiet while they're not actually playing). Sometimes you'll have unusual requirements from the client. In some cases, I needed an aux output to feed into an amp which generated a T-loop for hearing-aid users. That was an interesting one; it's basically a special amp in a dedicated box, with signal-input and a couple output terminals. You run a loop of wire (relatively thick, akin to speaker cable) around the perimeter of the area where the audience will be, connect each end to the respective output terminals, and I gather this generated a varying EM field modulated by the input signal, which hearing-aids in T-mode (and within range) then pick-up and use to drive their internal amp (as opposed to the external mic), for the hearing-aid user to finally hear (clearly, for them; essentially, glorified & (very) expensive generic-wireless (non-Bluetooth) earbuds). This is without any effects you may want; a common one is a bit of reverb on the vocals. Yet more signal management (rather than to alter the musical quality of the sound), is likely to include (dynamic-range) compressors. Electronics can handle only a given range of signal intensities. Too high and you'll get distortion; too low and it's at greater risk of being lost in the noise-floor. A compressor (proportionally, smoothly; unlike a limiter) attenuates the loud, so that the difference between loud and quiet is less (but still quite perceptible). In terms of audience-experience (let's pretend, for a moment, that the equipment could handle an infinite range; human ears cannot); either the quiet will be difficult to hear, else the loud will be at risk of permanently damaging hearing. Compressors also mean that you're not having to jockey the faders, quite so much (imagine having, say, 40+ channels to manage with only ~8 fingers, and you'll quickly see the problem without some kind of management). Ah, but all this ‘processing’ makes the sound impure, y'say? Well, remember that feedback (the nasty kind) is ever a threat. That definitely sounds awful, to everyone, and must be avoided. So, what's worse? A negligible (with modern, quality electronics) amount of extra noise, distortion, and such, or the significantly greater risk of definitely-awful-sounding feedback? Now for the less obvious stuff; why mic each part of a drum-kit (besides being able to have per-channel inserts (like a gate, mentioned before)? Because experienced engineers change the mix depending on the type of music being played. Mixing used for artistic effect. For a studio album (let's assume perfect mastering, without any silliness like extreme dynamic range compression of the loudness war); what's actually on your copy of the final production, is different to what you would've heard if stood in the recording studio. Deliberately. Sometimes, different instruments will be (deliberately, for artistic/creative reasons) be emphasised over others. Listen closely, next time; sometimes you'll hear the snare-drum louder, or the kick-drum (if the track is intended to be bass-heavy). An obvious one is during a solo; of course that source (eg guitar) will be mixed louder than the rest of the ensemble; that's the whole point. Why should the relative (or absolute) loudness in the studio have to match the desired output? An electric guitar, just by itself, with no gain/amplification, is extremely quiet. It's designed to have gain+amplification applied (else it'd be an acoustic). At the time of performance, the important things are timing, and character of the sound. Loudness can be altered later. To make the point, imagine lead vocals being expected to be inherently loud. The poor human is gonna end up hoarse, or losing their voice. No need. They sing at a comfortable level (whatever feels & sounds best), and the relative mix can be tinkered with after. The take used will likely be the one which best fits with the other performers (eg timing, and character of sound; I won't pretend to fully understand this part; but, imagine something along the lines of which rendition best conveyed the intended feelings behind the lyrics). So, when recording in a studio, this is why multi-track recordings are made; to be able to change the relative mix in post, however you wish (often with several different candidates). Anyway, I could go on. I think receptive readers will have got the gist/point by now.
Y'mean, all the sound, including noise, no matter how bad? Well, sure, if that's what you're going for. Very few do, though. Again, think about why a recording studio has so much sound treatment, and often different (sound-isolated) rooms. To have a controlled environment. Same reason why a theatre, by default, should be pitch-black. So that the lighting designer has absolute control over what light ends up where. Back in Shakespeare's day, when none of that was practical or even possible, and so theatres were open-air, naturally-lit structures; that's why characters will explicitly mention in dialogue about how dark/cold the night is (because how else would the audience know, since they'll be watching it during daylight?). But, at least lighting designers don't have to deal with damned feedback!
Right; here's where it gets technical, and interesting. This is a misleading characterisation of how digital audio works. I'll attempt to give a summary, but I strongly suggest you watch the tutorial videos made by Monty via Xiph; last I checked, there were 2. The first is a primer; an excellent introduction to many of the fundamentals & concepts. The second is quite specifically about various myths related to sample-rate and quantising audio in general. The second is especially notable, because he gives real demonstrations, using a whole bunch of equipment (all of it old-school analogue (some of it featuring hilariously large micro-controllers), except for the USB audio interface to the computer, as the only digital part (acting as an ADC+DAC (that's, analogue, to digital, back to analogue, converter). He measures the output using old analogue equipment, too. I think you'll find the results surprising. Essentially, so long as certain criteria are met (eg your sampling rate is at least double the maximum frequency you want to record), analogue and digital are equivalent and interchangeable. This is as per the mathematical work of Shannon & Nyquist (and probably others). You also don't need >48KHz, or >24-bit-per-sample, either. Monty explains that, too. I do a fair bit of recording of speech, with a portable digital recorder (Olmpus LS-10; ~14 years old now), and for speech the lowest fidelity (44,100Hz, 16-bit; basically CD fidelity) is perfectly good enough for speech. I would use a lower rate (but not resolution) if it were available, down to a point. Human speech is typically around 400 to 8000+ Hz. At least, for it to be intelligible. A higher threshold at the top end (eg 16-18kHz) does sound rather clearer. My point is that at 44,100Hz (sampling), the maximum frequency that can be reliably captured is 22,050Hz (acoustic); that's significantly above human threshold, and well above the high end of speech. Recorder captures uncompressed (linear PCM), and at 44.1kHz & 16-bit, that's about 10MiB per minute. I have absolutely no need to increase any of those numbers, just to get a slightly higher fidelity (for dogs, I guess, at those >96kHz frequencies) of the various ambient noise I have no control over. Pointless, especially for all the extra data it produces (which reduces my recording duration (given storage & power limitations of a portable recorder)). There's a whole lot more to it, but you're better watching the videos, than me trying to rehash it. DAC doesn't (except in very crappy equipment) use a zero-order hold (to produce a stair-step shaped output waveform from the samples). It finds the unique sine-wave solution to fit the samples (which are best represented lollipop style; because we have no data between samples). Digital output (unless artificially crippled; eg 1 bits per sample and 10 samples per second, or something stupid like that) yields a sine-wave which matches the original/source/input (well, assuming we're not using lossy compression anywhere, and without any waveform manipulation (DSP, or similar)). So, not an approximation, at all. Not even close. I'd be rather more concerned about the much greater effects of the analogue parts of the system (adding noise, distortion, picking up RFI (unless you're using differential signalling; that's yet another explanation; pro audio uses 3-pin XLR connectors for very good reason; differential/balanced signals are clever engineering)). So, exactly how are you connecting those generic mics to that generic recorder? That's gonna matter much more than anything going on inside the electronics of a digital recorder. So, not only mistaken about types of digital compression, but about deeper fundamentals of digital audio (and somewhat/arguably analogue audio), generally.
Redundant data? Wut? When doing things sensibly, the recorder shouldn't be trying to do compression. Several reasons for that.
Why (depending what you mean by such ambiguous terms) would you want redundant data, anyway? It just takes up more storage. Redundant data (except for specific purposes, eg parity or other error checking, but that's to ensure that the core data still has integrity and hasn't been corrupted). Surely, if you're going for maximum fidelity, then by definition there is no redundancy; all the data is needed to preserve the nuances of the original/source? Unless this is precisely your intended meaning.
At this point I'm confused, based on the ambiguity of earlier words, of if we're before or after the ADC. So, even following your explanation is challenging.
Well, it's hardly a system; it'll be an algorithm implemented in an IC.
Nope; like enzo1982 said, this might be true of lossy, but not lossless. A really simple example of a lossless compression technique is a repetition-count. Imagine an image with large areas of uniform colour. Imagine we're storing the pixel values serially (starting at 0,0, and working our way across and down, just like a CRT would, to the opposite corner). When we hit a section of a different colour, then for that first pixel we have to store it's value. But, after that, many pixels are the same. So, instead of repeatedly storing the same value for each pixel, we could design the schema so that we can specify a repeat-count (more commonly, in this context, called Run-Length Encoding), which simply says how many more pixels in the predictable sequence, have the same value. That would take much less data to store than otherwise. Yet, (and here's the crucial part), we can perfectly reconstruct the original pixel matrix, to match the uncompressed original, without losing any detail. Thus, lossless, yet still compressed because we used fewer bytes to store the same information. So, lossless is about efficient ways of using fewer bits to represent the same inforamation. Yet, even in lossy compression (eg JFIF (the format spec, which is from JPEG (the organisation/body), if we're sticking with raster graphics as analogues), your characterisation of of it pseudo-randomly discarding arbitrary data, is quite incorrect. There's much research which goes into lossy codecs. Modelling human hearing, both physically (frequency-response, masking (when multiple frequencies play simultaneously, we can't always perceive the difference to when only one is playing; so we effectively can't hear the masked tone; so it can be discarded without us (humans, that is) noticing the difference), and more) as well as how we perceive sound (so, a psychoacoustic model, then). But, sometimes, smaller byte-count is more desirable than perfect fidelity (at least at the DAC output). So, trade-offs. This is why smart music-lovers would have FLACs at home, but Vorbis/Opus on their portable player. They can always (re)generate the lossy from the lossless FLACs. This is also future-proof; when something comes along to replace Opus, then those FLACs can be transcoded to the new format, thus minimising fidelity loss (compared to transcoding from one lossy to another lossy).
On the proviso that this is still about lossy, not lossless encoding/compression. While not wrong, per-se, it's not a good description. Yes, mathematical models, because computers are glorified calculators. But, those models are, as I explained, far from arbitrary (or aiming for highest compression ratio at the cost of everything else; otherwise just discard all the data and you've achieved infinite compression; but your file is empty, which rather defeats the purpose and misses the point; it's about efficiency (reducing byte-count without significantly/noticably reducing fidelity; this is demonstrated in blinded listening tests; almost no-one can tell the difference with a well-encoded output of a modern lossy algorithm (else, the same complaint applies to your digicam outputting JFIF (JPEG) instead of a raw format (containing sensor data which must be post-processed)), not trying to do what is computationally easy, or carelessly discarding lots of data; only that which won't make a perceptible difference. That's the secret sauce. Compare engineering something mechanical. A vehicle, let's say. Extra wegiht means extra fuel-burn. So, you want less weight. But, you need strength too, which generally requires more material than less. So, the trick is to find ways to reduce weight, while preserving strength. Hence why materials science is a whole field, as is the precise shape formed; some shapes are inherently stronger than others, yet use significantly less material. Your characterisation, above, would be like trying to describe any car which didn't weight 50 tons, and dangerously unsafe. Um, just no. Worse; in vehicle design, you don't want the machine being super-strong, because in a crash it'll just transmit the forces (and thus the injury-causing deceleration) to the human occupants. That's bad. You want the car which, in a collision, will disintegrate; because it is absorbing the energy (instead of transmitting it to your body). So, counter-intuitively, the car which looks like a mangled wreck after a collision, is actually much safer than the one which looks mostly intact (for a comparable incident/collision). You're more likely to walk away from (or at least survive) the write-off wreck, than you are the one which look fine. The science, evidence/data, and police statistics, all back this up. What ‘feels right’, and what's actually true, are not necessarily the same. Quantum physics makes little sense to humans, but it accurately describes the small-scale cosmos (or, at least, it's the best model we have, currently; if you can think up a better one, you win a Nobel prize).
Err, no, some do it radically differently. Some don't do it at all (but do something fundamentally different). Video is a clearer example; different codecs can be optimised for different types of images. A tour of an art gallery will demand accurate colour/luminosity representation, but accurately capturing rapid motion doesn't matter (who wants a fast tour of art?). Whereas, for a sporting event, precise colour accuracy matters less, but capturing lots of motion, and/or rapid motion, does matter a lot. Perhaps also small details (think of ball games). Very different applications, requiring different techniques. There are quite a few different audio codecs for human speech (think of digital telephony & VoIP). Those are quite different to the ones for music. Compare how JFIF (JPEG) is good for photographs, but relatively bad for the likes of logos, diagrams, and text. Because (natural) photos don't tend to have sudden harsh changes in contrast/colour, but transitional gradients. For large blocks of uniform colour (logos, diagrams, text, captures of computer GUIs (think tech support); PNG will be much more optimal/efficient (fidelity versus byte-count) than JFIF would. Different applications. Different requirements. Different constraints. Different techniques. Different (sets of) trade-offs. There is no (singularly) ideal compression/algorithm for a wide range of applications. Back to what I said near the start, about general-purposes versus application-specific. Cryptocurrency miners often use ASICs; chips designed to do mining rapidly (and somewhat efficiently; though power-draw is less of a concern than hash-rate), but you wouldn't want it as a CPU (general purpose). Your CPU has sub-processors for some types of operations. Maths is a common one. Same reason why your GPU (and, many years ago, your sound-card) is a separate chipset to the CPU. It also depends on the input. Give a FLAC encoder pure (mathematical) silence, and it'll output very little data; because silence is very easy to represent efficiently. Largely similar with pure (mathematical) sine waves. Speech isn't simple, but compared to music it is. The complexity comes with natural sounds, especially music. Because the waveform is complex. To grossly over simplify, it's the harmonic resulting from the combination of many fundamental frequencies. Hence why a whole lot of (Fast) Fourier Transforms are performed in analysing an audio waveform. Because encoding the components might be more efficient than mindlessly tracking the resultant waveform. Now I'll mention MIDI; which doesn't encode a waveform at all. It encodes which notes were played. Very different. Hence needing a synthesiser to play them. There is no one-size-fits-all. In Video, Xiph is working on Daala, which uses radically different techniques than are used in today's video codecs (which are almost always lossy, due to the sheer amount of information / data; it has to be lossy, for practical reasons; uncompressed HD video is >1Gbps (good luck streaming that).
Technically, yes. But, how much of that is subjective? The real test is blind A/B listening tests (where you don't know which one is which). But, hey, in the end, if you don't like lossy, then use lossless. Lossy is optional for audio. Use FLACs for everything. But, if you're listening to a stream of a radio station, or you're listening via earbuds connected to your phone, then one can easily argue that lossless is entirely wasted. You won't hear the difference, because the limiting factor isn't the codec. So, save the bytes, instead. Think about YouTube, and why well-engineered compression matters for them; because even a 3% reduction is a lot of network traffic, given how many views are happening per minute. Sending lossless would be irresponsible, there.
LOL! Yeah, this is just insulting, really. If you're gonna criticise people (instead of their work), then at least attempt to understand them first. The clueful variety (perhaps you've just never met any real ones, just posers) wouldn't say anything like that. They recognise that there is no perfect (lossy) solution; just a set of trade-offs. Different variables which achieve different (but not inherently better/worse) results. Back to what I said, before, about different applications, and different constraints. It'll depend what's being encoded (music, speech, sine-waves, radio-wave signals (think either communications, or astronomy). The people involved in deep-wizardry at this level, will take great care over various factors. Your characterisation is akin to describing an office-worker as a paper-pusher. Or, a painter as just slapping some emulsion on canvas. Bluntly, for a moment, you're exhibiting the Dunning–Kruger effect. I'll quote Dolly Parton on this point; ‘it took a lot of money to look this cheap’. Is an architect just someone who doodles on a large sheet of paper, or might there be a little bit more to it than that? It's like with great design; just because you don't recognise it, doesn't mean it isn't there. In this unflattering fashion, you could mischaracterise a lock-designer as a common burglar. Anyway, back to those psychoacoustic models (which are described mathematically, because we want computers to use them; same way that you need concepts explained in natural language (as opposed to uploading binary data into your neural-net wet-ware)). Those models are an approximation of the characteristic response behaviour of typical human hearing. They're an attempt to describe (biological) reality (of a particular species). So, by claiming that somehow maths trumps empirical data, is to accuse scientists of being unscientific. Um, OK, if you say so. Care to cite your science credentials? I didn't think so.
Yes. But, is 2+2 even the right question in the first place? That is more akin to the sort of thinking required to design something like a codec. How would you suggest a codec (for a computer to execute) be designed without maths (given that computers are essentially machines to process & store numbers)? By all means, feel free to develop a superior codec. Or, even a better psychoacoustic model of human hearing. Because, clearly, the people who do that for 8+ hours a day, are absolute idiots, right? What have those clowns ever done for us? Forgive my hyperbole, but this is just beyond ridiculous.
I don't even know what that means, or is supposed to mean, at this point. It's not about the maths. Plus, while the fundamentals were worked on by mathematicians (I mentioned a couple of them, earlier, for those paying attention and who are still with me), the development of codecs is more likely done by technologists (in the broadest sense). Perhaps aided by mathematicians. But, hey, if compression is so bad, then you'd best stop using the InterWeb (which relies heavily on it; because network capacity is finite).
Ah, here we go.
An example assumption is that the target auditory system is that of humans. Show me any other species which needs its hearing modelled in order to develop audio reproduction systems. For a long time, humans themselves didn't qualify. You wanna know the kicker here? Your brain is digital. Yup, really. Be careful, now; mind you don't get any of your exploding grey matter all over the freshly cleaned floor. The staff don't like having to clean that up.
Careful consideration. By who? You? Are you serious, given all the misconceptions I've addressed here? I think you might need to check your own assumptions (I wish I had an apt maths joke, here; maybe the classic; your conclusion doesn't compute from your premise (which is also, itself, false); so an invalid and unsound argument, entirely; purely fallacious; you make Socrates('s ghost) cry every time you make a logical fallacy). What do you mean by spatial information? An audio waveform is one-dimensional. My wildly-specuating guesses include:
What do you mean by dynamic information? Dynamic=changing, after all. My best guess would be a misuse of one of the words in dynamic range. A much clearer term, if so, would be loudness/amplitude Whatever do you mean by ‘gets mangled’? I don't see the point of this discussion, at all. It just comes off as misinformed whining about a non-problem. Like some user complaining that they don't like how some software feels. Um, what's a developer supposed to do with that? It tells him nothing useful. Doesn't clearly identify the problem, and certainly doesn't suggest any solutions. Even if this were an attempt at a bug report; well, Fre:AC doesn't design codecs, it doesn't even develop implementations of them. It uses libraries (of other people's code) for those parts. Fre:AC is, in a sense, a collection of components, in a convenient arrangement. Lots of (other) software uses similar libraries, for entirely different use-cases (VoIP software, as one example). Fre:AC just happens (by sound design decisions (no pun intended, but I'll take the credit) of the developer, whom we should be thankful to) to use those existing codecs (via the relevant libraries), so that users may transcode to/from those formats. If you have a constructive bug report, then submit that to the developers of the codec (which isn't enzo1982, or anyone involved in Fre:AC). Else, you could alter your own choices; such as by insisting on lossless over lossy. By asking a friend to help you to conduct a blind listening test on yourself, or various encodings (to really determine if you can actually hear the difference or not; I highly doubt it; even I struggle, and I have more of a clue than most of what I should be listening for; casual listening would be almost laughable). At this point, you're engaging in confirmation bias. You could also choose, in the age of information, to reject ignorance, and actually learn some of the relevant background material. It would make you appear less foolish, since you would know to not pole-vault to outlandish,, baseless conclusions, which are so horribly fallacious. But, hey, easier to criticise others, instead. Outstanding. And normies wonder why engineer-types don't like them. No wonder, with attitudes like this.
Wut? This just sounds like woo (new-age-speak), now. So, digital bad, because you've decided to dislike it, based on misinformation? OK, then. Opinion != fact or truth. Your disparaging description is also true of any audio equipment. Yes, including purely analogue. No microphone is ideal, for starters. Hell, even acoustics; different venues have different acoustics. Which is good/correct, or bad/wrong? Mu (the question is flawed); they're different. I recall, from listening to Howard Shore talk about his experience of composing the score for the Lord of the Rings trilogy (directed by Peter Jackson); in which he gave a precise description of what constitutes the LotR theme music. Part of his description (it's been a long time, so I forget all the details) was that the specific performance he described was in the acoustics of [wherever they recorded it]. Yes; the acoustics of the recording venue were, to the composer, part of what defined the sound which was the score he composed. Different acoustics would mean that it sounded different; enough to (arguably) be different music. I found the implied point rather fascinating. I'm gonna take Shore's word for it; it's hardly a nobody in music. He also knows vastly more than I ever will, about music theory, and music composition. He has the authority to decide on the creative/artistic aspects (he's the composer of the piece, after all). So, in the context of audio engineering, to talk about the ‘evils of digital’ as destroying the acoustics of a venue. Well, no, they don't. At worst, they're not accurately represented. The venue still exists, regardless of how badly you're operating the equipment. Plus, my point here, is that your claim is also true by simply moving to a different venue (as touring performers frequently do). So, simply playing the same piece in a different venue, should also be decried with the same vigor, no? That hurts my brain to even type, with how silly it is. Hell, real instruments are far from perfect, flawless, or stable. Think about an orchestra tuning up just before a performance. Especially instruments made of organic material (eg wood), but also to an extent with metallic instruments, change over time and become out of tune. Worse; natural environmental variations will affect them also; humidity, air-pressure/density, temperature, and probably a few others. What about a violinist having their bow restrung? Should that be forbidden, too? Where does it end? This is akin to characterising a synthesiser as superior to a natural/real instrument. because it sounds more uniform (this is one thing that I can definitely tell the difference of, in recordings; was the instrument a real one, or synthesised; the synthesiser is almost flawless, it sounds ‘too perfect’, whereas a natural instrrument will have imperfections, character, oddities, which makes the sound which it produces much more interesting. You're trying to achieve perfection from an inherently imperfect source. A better example with even more variation, is the same singer's voice. Changes over both the stort-term, and long-term. Yet, nobody's upset over that. No two performances are the same; there's always variation. So, which, exactly, are you trying/wanting to capture so perfectly?
Don't forget what I said, before, about trade-offs. Plus, more bluntly; if they're so horribly flawed, then how is it that they can reproduce music at all? Why use them? But, this line of argument is more deeply flawed. Newtonian physics is, technically, incorrect. It was replaced with Einstein's Relativity. Strictly, we should never use Newtonian physics, ever again. So, that makes for a good trade-off. Simpler, but only 99.999% accurate. Oh, but that's <100%? OK, well, if you're doing something funky (like putting atomic clocks into orbit, which must remain in accurate sync with similar atomic clocks on the planet surface) well yeah, go for broke and bust out the full-on, hard-core Relativity equations, if needed. But, what about in 98% of cases where Relativistic effects are negligible (or, not even measurable), and 99.999% accuracy is more than sufficient? Why should I have to use the much more complex (and computationally demanding) equations of Relativity, when Newton's are simpler (thus less bug-prone to implement), quicker to execute/calculate, and probably use less memory too, yet the accuracy is more than enough for me to determine the ballistics of my projectile? It all goes back to trade-offs (welcome to a fundamental concept in pretty much all engineering, and several other fields). Warning: I'm gonna make your brain explode, again. Your hearing profile changes, both short-term, and long-term. Permanent decline (especially in frequency-response) happens with age. Hence why older folks complain about bass-heavy music; because their top-end is more attenuated, so treble sounds quieter (to them), while the bass (mostly) remains; but their perception is that the bass is much louder than the treble. Even if it's the same waveform, and same loudness (eg same song, on the same equipment, at the same amplification), but played 20+ years apart. I'm very much reminded of You Don't See in 4K. Seriously, go watch Monty's videos; he knows his stuff, and explains well. On top of that, especially the mechanical parts of your playback system will change (not necessarily for the worse, in all cases) with age. So, there are variables all over the place. But, yup, let's just ignore all of them, and focus on the digital recorder. Frankly, I'm reminded of audio-phoolery nonsense, about CPU context-switching causing jitter, about needing magic rubber feet for your vinyl/optical players, about needing massively-overpriced magical exotic-material cables (including the power cable, nevermind the rest of the electrical cabling in the building), about how vacuum-tubes are superior to transistors, about how magnetic tape is superior despite it having a significantly lower SNR (that's Signal-to-Noise Ratio; go look that up, too). Repeated assertion doesn't make any of it true. Like I said, earlier; for non-special speakers, a 5 or 10 metre length of 5 amp, 2-core electrical cabling does just fine, for me. Speaker cable is about carrying current. That's exactly what electrical cable does, and is designed for. But, hey, if you're really serious, then replace all the copper you're using as electrical conductor, with silver. Silver is a better electrical conductor (no, really, it is). Oh, you don't want to because silver is hugely more expensive? Well, there you go; trade-offs. Copper is good enough, for a much lower cost. Need more current (without things melting or catching fire)? Thicker cable, and a shorter run of it. Even in the pro audio world, using TurboSound amps & speakers (really good equipment, made in Britain, company founded during the Brit-rock/pop golden era); you could carry 10 metres of (copper!) cabling to hook the speakers up to the amps, on your shoulder without much difficulty. It wasn't light, but it also wasn't stupidly heavy. If you had 4 stacks of speakers (needing 8 cables), then you might have to make a few trips; or you wheel the relevant flight case over to where you need it, and just place out the cable you wanted.
Why would it? That's inefficient. If it's truly redundant, then what possible use would it have? (By definition; none.)
This is fallacious, as I've already demonstrated. However, your muddling of terms makes it incorrect in a different way. Infomation != data. So, a little information theroy. Information (in this context) would be a piece of music, a photograph, or a document. The thing we're trying to represent with data. However, there are many ways of representing information in/with data. Different methods have different properties. A common trade-off is that simpler schemes require more data (there's more redundancy in it). Think of an uncompressed image. Notice, dear reader, that redundancy in data is different from redundancy in information. While there may, indeed, be no redundancy in the information, there is indeed redundancy in the (simple scheme of representation by) data. Otherwise, compression wouldn't be possible, or exist (even as a concept), at all. Compression is all about removing redundancy in data while, at most, making negligible changes to the information. Play around with an image editor (GIMP, for example). Get any photograph (ideally, which isn't already compressed as JFIF (JPEG)). Load it into GIMP. Tell GIMP that you want to export it as a JFIF (JPEG), and to have it show you the advanced options. You should see a slider for amount of compression, and a preview which responds to your moving the slider. Play with it, interactively. Notice that, when starting at nil compression, that you have to move the slider quite far toward max compression, to start noticing any (obvious) differences at all. It's only at the extreme end (when lots of data has been discarded; recall that JFIF is lossy compression, not lossless) that it's apparent. For all this bashing of lossy audio codecs; the ones used in telephony for voice, are part of why telephony became cheaper than in the past. Because operators could serve more customers for the same investment in infrastructure. It's especially why long(er)-distance calling isn't only for the very rich. Those codecs enabled many more simultaneous calls to be sent over the same strand of optical fibre. A copper loop (twisted pair) is only a single call (as analogue, I'm discounting digital-over-copper). Digital-over-fibre can be dozens of calls (or probably more, now) over a strand. Because of compression. The old school ways of doing it, pre-compression, were much more expensive. I'm reminded of a relevant XKCD #841: Audiophiles.
No; you're so very wrong that you don't even know how wrong. You're conflating different concepts, and drawing nonsensical conclusions.
Except that there is. You've used it. You're probably using it right now; between your Web browser, and the HTTP server, they probably negotiated to use gzip compression (but only to about 60% of the maximum ratio, to keep the compute time short) of the text, in order to reduce how many bytes must be sent over the network (fewer bytes, at a given rate, means the transfer finishes sooner; that's grossly oversimplifying networking, which is a complex field, but that's the gist of the idea). So long as the reduction in transmission time exceeds the compute-time for compression+decompression (former at the server, latter on your computer), then we're gaining (network) performance (hence why the compression ratio is kept modest, lest compute time exceed the reduction in transmission time; which would be a performance loss). Yet, does the text look garbled or otherwise unreadable gibberish on your monitor? I'm gonna guess no, sincec you're able to interact with this site enough to post a discussion. When you post comments (or start discussions), your browser may have been applying lossless compression to the text, and the server decompressing it. It can work both ways. Disagree? Prove me wrong (with facts, credible evidence, and sound reasoning). Ever used a zip file or similar? Lossless compression. But, you've also used lossy compression, too, without even knowing it. So, you're denying your own reality. Well done.
Precisely! Science, it works, bitches! If it didn't, then who would use it? Here's the test, for lossless; if I showed you two datasets (let's say an audio waveform), and told you that one had been through (compressed and then decompressed by) a *lossless codec, while the other was the uncompressed original, and asked you to tell me which was which … Lossy would be different (but not enough that you could reliably tell me which was which), but your conflating lossy with lossless is entirely on you. Learn what you're criticising, before criticising it. Often, what might appear later as foolish design, was actually clever design for a set of constraints that you don't understand. Interlacing (in video) comes to mind. Yes, it's a pain-in-the-arse now, but it made perfect sense at the time, in that context, for that application. It was a clever hack (but, yes, still a hack). If really so concerned about lossy codecs; well, firstly don't use them. But, if you're gonna use them anyway; well, then choose libre ones (Vorbis, Speex, CELT, Opus, for audio, and I think some of the telephony codecs are libre (both copyright on the implementation, and the patents have expired)). That way, you can learn how they were designed & developed. You can also submit (real, legit bugs, not vague whining), and otherwise help to improve them (for everyone). Start with those blinded listening tests, though. And Monty's Xiph tutorial videos. Perhaps also ask questions, instead of making utterly flawed statements which have no basis in fact or reality. Learn first, only then form opinions. Else, opinions out of ignorance can be nothing more than arbitrary, and very likely to be flawed/false.
Music is maths, at core. I heard a compelling argument, once, against using maths as a common language with other civilisations. The speaker's conjecture was that what we call maths is, akin to music being patterns which are pleasing to our brains, something of a self-reinforcing feedback loop in our brains.
Mathematics says otherwise. Lossless codecs which caused differences, aren't actually lossless. When seeking a lossless codec, such algorithms would be rejected. But, OK, let's imagine that you are hearing a difference, and assuming that it's due to a lossless codec. So, we'll start by setting aside the dubious assumption. There are (as enzo1982 suggested) other possible causes for perceived differences. The first is perception, and cognitive bias. Eliminate that with double-blind listening comparisons.
Fallacy; argument from personal truth/belief, else argument from popularity Thinking something, even if many people think the same, doesn't make it true. Many people believe the Earth to be flat. Others believe that lights in the sky simply must be UFOs. I can believe/think that I'm a billionaire all I want; it's not gonna increase my bank account balance. People then engage in confirmation bias. Accepting only that which reinforces their existing beliefs, while rejecting that which calls them into doubt.
Fallacy; argument from authority. Define ‘best’. Being intelligent/competent in one area, doesn't mean that someone is free of cognitive bias. I'm reminded of when James Randi described his being invited to speak at a MENSA (society for high IQ people) event. He arrived, and began his lecture about the nonsense (pseudo-science) of homeopathy. A few minutes in, he was interrupted by some unhappy members of the audience, who objected, because they very much believed in homeopathy. Randi walked out, seemingly in disgust, and never returned. He explained that he was unwilling to endure the cognitive dissonance exhibited by otherwise intelligent people believing in (as he calls it) woo. He walked out, because he deemed it futile. If high IQ people were just as likely to believe in nonsense, then what was the point in him giving a presentation on cognitive bias & cognitive dissonance? None, he concluded. I've seen several of his public lectures. A highlight when he talks about the bunk that is homeopathy, is that he brings an unopened bottle of homeopathic pills. Sleeping pills. 7+ days supply (might be 30 days, actually; I can't recall). On stage, he downs the entire contents, assuring his audience that he'll be absolutely fine, and that they're just sugar pills. He continues giving his lecture, remaining conscious the entire time. His point; if homeopathy worked, then surely downing several days of homeopathic sleeping pills would render him unconscious within the remainder of his time on stage (30+ minutes)? Yet, it doesn't. He appears absolutely fine. What (valid, plausible, matches-with-reality) conclusion is there, other than that homeopathy is bunk, and the bottle contained only sugar pills? If he downed in one go that many real (prescription, medical, sleeping drugs), he may well be dead, let alone merely unconscious. Randi isn't exactly young any more, either. If these so-called engineers fail to understand such basic fundamentals, then I can't take seriously any notion of them being competent techs, let alone the best in the world.
Yes, so small (in lossless codecs), that they're zero. Non-existant.
Because there are many approaches to a non-trivial problem. Different techniques. This is akin to asking; if container ships don't loose cargo while at sea, then why are there so many logistics companies? Multiple attempts is how science works. Why do many mathematicians attempt proofs of the same piece of maths (or maths concept; I'll admit to not knowing the proper terminology; yet I'm receptive to learning)?
You'll have to define that, given your habit of mincing words. Besides, your contention is that there is no such possible thing, inherently, with digital recorders.
Well, see, your reproduction steps are unclear. Do you mean encode the uncompressed original using FLAC, and then encode the output of FLAC with APE? Else, do you mean to make two separate encodings from the same uncompressed source, one with FLAC, one with APE? The two scenarios are (experimentally/procedurally) quite different. Even if the audio should be identical.
Define what you mean by high quality headphones. How're they connected to the host? Because, if you're connecting them directly, without the proper equipment to drive them, then they're hardly achieving their potential. Again, you're focusing on one specific part, while neglecting many other relevant parts.
What difference? Characterise it. Produce a diff between the disparate waveforms. Quantify it, somehow. Please. Otherwise it's just more empty claims. I recall, years ago, being asked to someone's house, due to having computer trouble. The user complained that when a specific process was running (I forget what it was, now, this was from years ago), he could hear a difference in the audio output. This struct me as bizarre, but I was curious. I listened, while he reproduced the problem. I heard no difference. I asked him to describe the difference he hears. Suddenly, he's quiet & evasive, either unable or unwilling to give any description what so ever. I explain the difficulty of even beginning to attempt to diagnose a problem which I can't observe and which he can't characterise or demonstrate to me. He was unhappy about this. That wasn't the only miracle expected of me, that afternoon.
Yet, your claim is that digital recorders inherently distort their input, and don't capture faithfully. You seem to be contradicting yourserf.
Rigorous experiments? Show your methodology. What was your null hypothesis? How did you falsify it?
Encryption is not compression. Compression is not encryption. Please show/cite the mathematical basis on which you claim that lossless codecs are not lossless actually. Please also include experimental data (from rigorous experiments), confirming your proof-of-concept.
Citations? Papers? Even if so, how does any of this relate to lossless codecs being flawed? Seems fallacious. Take Jamie Oliver (celebrity chef); let's assume that he's a perfect chef (if such a thing is possible). Just because he knows food & cooking, doesn't mean that he also knows how to run a lasting, profitable business. Indeed, his chain of restaurants failed, commercially. Yet, he remains a chef. I'm seeing nothing but non-secutors, here. That, and (more) appeals to popularity/authority.
Respond to what, though? So far, I've only seen vague claims/assertions. Nothing to actually seriously consider.
No need; describe them, or present data (with methodology/reproduction-steps). Just as you would present a proof, for demonstrating a mathematical truth (instead of merely claiming/asserting it to be true).
How, exactly, given the above?
Which is an outlier from the consensus. Yes, I'm aware that this could be dismissed as an appeal to popularity fallacy. My point, though, is that without anything more than claims/assertions, while having reached a radically different conclusion to most other recordists and software developers (who are sufficiently familiar with audio encoding), then there's nothing to actually go on, when you can't describe or show the difference. Data would be most effective in persuading people to seriously consider such a wild conjecture. Without it, just seems like dishonesty. Unless, of course, you're a troll. That would explain it, too.
And what's any of that got to do with audio encoding, and more specifically your (so far) unfounded claims?
Yet, I thought that digital recordings were inherently bad? I'm confused, now.
Or, you could just send it electronically, or post it online. No need for sneaker-net.
Kindly cite his academic papers, or other works.
Again, relevance to lossless encoding? There are so many papers, because it's an interesting field merging biology with technology. There are lucrative applications for useful discoveries. The count of papers on the topic is evidence that lossy codecs are of sufficient fidelity, while significantly reducing the data needed to represent that information. Or, at least, the underpinnings needed for them are drawing interest, which suggests interest in lossy codecs. Because they meet a need (as does lossless).
Citation, please.
Perhaps in the data used to represent the information (audio), but not in the information (audio) itself. You're conflating different things, again. Go back to the Run-Length Encoding example, I cited, earlier. Showing how repeat-counts can require much less data to achieve the same output. Compare a set of instructions. Imagine that I find a way to simplify the set, to reduce the count of instructions (perhaps by using features like conditionals, and loops), but which achieves the same outcome/output. No; it's simply made the instructions more efficient, for the same outcome. Take it to absurd levels, to make the point; imagine you're following a set of instructions. Imagine that every other step is to go stand in the corner of the room and turn 360 degrees on the spot. Imagine that all the other instructions are, I dunno, how to construct a box from a flat-pack (so, from a sheet of cardboard, to an empty container). In the end, you'll finish building the box. Yes, I'm assuming that the instructions are bug-free (except for the corner dance routine). Now imagine that I, the author of the instructions, notice this inefficiency of you going to twirl in the corner so often, and revise my instructions to eliminate this, and have every step be about constructing the box. So, I've just halved the instruction set, basically. Will you still end up building the box? Those instructions haven't changed, so I don't see how you would fail, if nothing else had changed. So, you'll still build a box, yet I've reduced the instruction set by 50%. How? By your reasoning, this must somehow mean that you'd only build half a box, or an otherwise incomplete one. Why? How? All I did was eliminate the inefficiency. Same deal with optimising data to more efficiently represent the very same audio information (shape of the waveform). Less data, same representation. You seem to assume that the two are equivalent, or (worse) the same thing. Yet, if so, then it wouldn't be possible to represent a waveform via different methods. However, you said yourself, earlier, that there are many algorithms out in the wild. So, you agree that there are different ways to arrange data, to represent the same information, then. Same with other types of info; different image formats. Different languages, different text-encoding, and so on. Hell, there are a variety of ways of encoding uncompressed audio. No lossless compression what so ever. Else, compare text encoding. For English, UTF-8 uses far fewer bytes than UTF-16 or UTF-32; yet, all of them can equally represent the same English (or other) text. For non-English, things become more complicated. Essentially, UTF-32 is a fixed-length representation (notshell: always 32 bits per Unicode character (until the count of character code-points increases significantly)). UTF-16 and UTF-8 are variable length. ASCII characters will use the minimum length, but any character may be encoded (it'll just require a longer-than-minimum sequence). Still, all of them can represent any sequencec of characters. However, since ASCII characters keep UTF-8 to its minimum bit-length for a character, then UTF-8 (for English text) uses far fewer bytes than UTF-16 and especially UTF-32. Yet, by your reasoning, UTF-8 must somehow be corrupting the text. Well, do explain how you're able to read any of this, given that UTF-8 encoding is pervasive (if not ubiquitous) across the Web). Why wouldn't UTF-32 be preferrable, at 4 times the byte-count (and with much redundancy (most bit-sequences would be zeros).
Oh, the irony. Yeah, funny how the most certain people also tend to be the most oblivious. What, in your estimation, is missing from the maths about sampling, encoding, and compression? Do you reject Shannon's sampling theorum? What about Nyquist's work? Please cite your paper showing a proof which refutes Shannon's sampling theorem. It'll be invaluable to many scientists. Does this same point about uncertainty also apply to your encryption software? Or, are some areas of maths (considered) complete (for a given application)? If encryption is one of them, then what about audio encoding/compression? If not, why not. Explain how.
Ah, that old chestnut. The God of the Gaps argument. Religious propaganda is something I'm quite familiar with. This explains the many fallacies. Nature is all around us, by the way. No need to mis(apply|use) the gawd label. Ultimately; do you care if what you believe is true? Anyway, that's quite enough from me. Class dismissed! |
Beta Was this translation helpful? Give feedback.
-
If you record a live performance in true stereo using a known technique such as a coincident pair of high quality microphones, and feed the outputs directly into a digital recorder without any other processsing, the result will be a pure sound that captures the dynamics of the musical performance as well as the ambience of the auditorium or concert hall. The digital recorder simply slices up what it hears from the microphones and gives each slice a value. On a good replay system, again, without any other processing, the result will be a very good facsimilie of the original performance. There will be no redundant data involved as the digital recorder will not have recorded any.
Now, if we pass this audio signal through a 'lossless' compression system, the codec will identify what appears to it as 'redundant' information and discard it. The problem is that this 'redundancy' is calculated by a mathematical algorithm which makes assumptions regarding what is redundant or otherwise. Each codec does this slightly differently, which is why they all sound different from each other. The codec authors will argue that the mathematics cannot be wrong. After all, 2+2 will always equate as 4. However, while the mathematics may be sound enough to the mathematician, the assumptions upon which they are founded are not.
After careful evaluation, it appears to me that it is the spatial and dynamic information which often gets mangled. The natural ambience of the venue is compromised and the dynamics are curtailed. This is because the codecs involved have made incorrect assumptions. The original digital recorder did not record any redundant information and so, consequently, there is none which may be safely removed. That's right, there is no such thing as truly lossless compression. However, the available codecs nonetheless work well and provide a useful compromise between file size and audio quality.
Beta Was this translation helpful? Give feedback.
All reactions