The Myth of 'High Resolution' audio

[Rothchild]

I must have missed the subtlety in 'Gah.. some opinioned person' that meant I mis-interpreted your post as not endorsing the xiph link.

I'm not sure what you mean when you talk about 'impulse response' what is this and how might it change the conclusions drawn by the paper?

The stuff about imperfect filters is dealt with in the section about oversampling afaict (and is consistent with the second link you've offered).

There is very little doubt that higher sample rates and bit depths are of benefit to recordists (although it has to be said many of the benefits are not purely sonic but relate to ease / quality of processing and provision of headroom), but this is a hi-fi forum so we're talking about (mostly) playback, (and this is what Monty is talking about too, it's about replay not capture) in this regard most of those benefits are pretty much moot.

I find the piece to be well informed, logically structured and apparently based on good research. It starts with the function of the ear and moves back towards the theory and practice of digital audio. Using empirical research to establish an understanding (or opinion in your terms) is fairly standard practice.

There are some links claimed by some to provide rebuttal in this thread: http://forums.stevehoffman.tv/threads/192khz-24bit-vs-96khz-24bit-debate-interesting-revelation.317660/page-3 but others in the thread claim they're not as substantial as a rebuttal but a bit of a spread of understanding, this seems to suggest that Monty's piece may be not be 100% accurate but that it is substantially true.

[Audio Advent]

I neither endorse or err... non-endorse? the link. Yeah, what he says may be correct but it's not the whole story and none of it prooves that higher resolutions don't result in sonically better results. He only opines that it cannot possibly result in better sound because of a) b) c) but ignores d) e) and f).

Even so, where are his references?? He lists none. Google his graph at the beginning of the page for example and it would seem he has added the red graph lines himself - other graphs of the same results are open ended as they didn't test that far up the frequency range, and others which do go almost as far quite clearly use dotted lines to show that they are estimated, extrapolated.

He states in the text below the graph that it includes added results from more modern research which has tested to beyond 15KHz but what results?? Doesn't list any, doesn't say anything about them and how far did they test to? 17KHz? 18KHz? Fudge fudge and more fudge! It looks more cut and paste from around the web to me!

The whole section about comparing it to eyesight as well... OMG! He gets stupid people to realise that they can't see infrared from a remote control and uses that "oh yeah, I can't see infrared from my remote control! He's right! " emotion to reinforce the "truth" of a completely DIFFERENT assertion! It's like the reverse of a straw man argument to proove a positive.

One interesting thing to note about the sight/colour analogy is that in the last 6 or so years it has been shown that our eyes have more than just rods and cones and has other light sensors which allow some blind people to still detect light. People were so convinced that they had the workings of the eye all worked out that evidence that there was more to discover was dismissed as nonsense for about a decade or more until one opthalmologist fought through the arrogance and took it upon himself to proove his hunch despite great ridicule and derision. The guy was a guest on Radio 4's "A Life Scientific" but I can't be bothed to search through the episodes to find his name..

So, not only is the analogy irrelevent and also manipulative, the deeper story of opthamology is that people thought they knew all there was to know about eyesight and were so arrogant and closed minded that they rejected theory to the contrary until they were proven wrong. Why this can't the same for hearing I don't know.. We do have research from the 1950s by R.J Pumphrey which hints that bone conduction in the ear can lead to detecting of sounds into the 50KHz region but I can't find a full paper that's free to read, only brief summaries or mentions..

On the other hand it may be perfectly correct that it's impossible to hear over 20Khz ... but high-res sounding better doesn't necessarily have to do with frequency ranges at all, it can be about other things. But pretty much the article stops there at that, that's his argument.

The rest is spent telling people how digital works and debunking pretty basic mis-conceptions about digital audio. Surely anyone seriously interested in digital audio will know all this already? It does nothing to proove that higher res cannot sound better, only to show that some people's - let's say ignorant to be harsh - arguments are wrong on a basic level.


The stuff about imperfect filters isn't covered - it's mentioned elsewhere on his website but he at no point goes into or acknowledges that you NEED perfect filters (digital and analogue) to get digital to work at its best at CD resolutions and that we don't have perfect filters. There are many who love NOS dacs and hear them to be more realistic and natural sounding precisely because they don't use digital filters and don't have the possible artifacts thereof. When we push into the high resolutions we have no requirement to use these theoretical but difficult to get right Nyquist filters and can instead use easy to implement shallow sloped filters. This deviation from the Nyquist theory will then only create artifacts still far outside of our hearing range whilst leaving the audible section untouched. This is more apparent in the anti-aliasing prefilters on A/D conversion at the recording stage.

If higher resolutions are of benefit to recordists, then they are of benefit to us! Better headroom and better quality of processing is ALL about the sonics! It's about how processing impacts the final sound. If they record in higher resolutions for whatever reason then why would we as consumers then ask to have that served up at a lower rate, what would be the point? Just listen to it as it was recorded. If something is on a mastertape, then I would first want to listen to the mastertape if I could have it.. transfering then to another medium is an unnecesary step unless giving out the mastertape is impractical (of course it is with analogue). With digital, there is no real impracticality to hinder the distribution of the original high-res recordings as they were recorded. If you had the right program and the artist didn't mind, there's no reason why they couldn't also easily distribute the whole multi-track recording as a file and use peer to peer tech to distribute it, like Thom Yorke's last album. "Mastertapes" direct to the fanbase.

If you downsample to some consumer format considered "good enough" for the masses and don't do it properly, you risk adding aliasing artifacts back in to the audio band which you supposedly spent good money on quality A/D converters to avoid in the first place.. so the high-res should be distributed instead.

Impulse response is about the time response of the conversion and is effected by the necessary low pass pre-filtering at the A/D stage. The slower, more gentle these filters, the more accurate the time response. That leading edge of a sound is what tells the brain what to expect next, gives much of the character of the sound, timbre etc. In other words it being more accurate increases the realism of the recording. There's mention of impulse response at the end of this AES paper along with a nice graph of impulse response of the different resolutions: http://tech.juaneda.com/en/articles/dsd.pdf

[Rothchild]

Thanks for the fullsome response, you make some fair points, I agree that arguing by analogy can be a dangerous thing and lead folk to feeling like they've been misdirected.

I've subsequently read the ARA article: https://www.meridian-audio.com/meridian-uploads/ara/coding2.pdf which makes some useful additional observations - although that and this: http://www.ambisonic.net/pdf/hiresaudio.pdf seem to put the jury out on any additional benefits beyond 24/96

Just to pick up on the benefit to recordist thing (as this is closest to my own interests) I do maintain the point that then benefits are not anywhere near in the same order of magnitude upon replay, (digital) headroom seems to me irrelevant in a replay context as the dynamic range is already 'baked in' to the mix.

[Macca]

I've heard a CD of an album pre-mastering - the dynamics practically ripped the speakers apart. You can have a master-tape but it has still been dynamically compressed so that you don't have a huge dynamic range between the loudest and quitest parts of the recording because most equipment - and our hearing - won't cope.

So I can accept that higher sampling rate might be making some sort of difference, albeit subtle, to the sound quality, even if we are not sure why, but we really don't need any more than 16 bits for more than adequate dynamic range.

[Audio Advent]

The irony is that oversampling in the digital domain was a solution to the problems of early digital where the low pass filters into the A/D stages would produce ringing etc. It was a solution to the problem of high sample rates not being possible at the time (they could achieve those rates digitally by adding zeros in between samples).

Now that those high sample rates ARE available to solve those problems... we have a consumer base who argue that those high sample rates are all a con and is just a marketting ploy to get at our money!

Crazy huh?

[Audio Advent]

, but we really don't need any more than 16 bits for more than adequate dynamic range.

Adequate may be (I've been really enjoying a lot of compressed music off the internet recently after getting a Caiman II - so I'm certainly not complaining about say 320kbps max at 16/44.1) but like the Meridian guy says, we need at least 20 bit to fulfill the limits of human hearing with noiseshaping, dithering used at the same time.

If it's easy to implement something much better, why would you go for "good enough"? YOLO! (:lol:)

[Macca]

My experience with Spotify premium is that you can clearly tell - on some recordings, not all - that dynamics are flattened. So I regard recordings as less than red-book standard to be below an acceptable standard for critical listening.

The dissatsfaction with compact disc as a medium for critical listening I attribute to a) over-use of dynamic compression and b) poor implementation i.e the use of low quality active pre-amplification.

The idea of replacing all of the recordings I have in 16/44.1 with the same recording at a higher sampling rate is something I don't regard as 'easy' especially since 1) I don't do downloads and 2) it is often impossible to discover if the download is a genuine 24/96 recording or a 16/44.1 that has been upsampled.

Granted if we were all starting from scratch your argument that we might as well have all recordings issued at 24/96 would hold some water.

And we still haven't got round the issue that early digital and all of the analogue masters will not contain any musical information above 22 KHz (and a lot lower for those made in the 1950s) due to the recording equipment used when they were made.

But - I'm not willing to dismiss out of hand the anecdotal reports from experienced listeners with very good quality equipment - like Gaz - who say 'yes I can hear a difference between the two iterations of the same master' even though it does not tally with the results of my own comparisons. I did make those comparisons on someone else's workmanlike but not outstanding system, though, so perhaps time to revisit that exercise.

[StanleyB]

Adequate may be (I've been really enjoying a lot of compressed music off the internet recently after getting a Caiman II - so I'm certainly not complaining about say 320kbps max at 16/44.1)
There are a lot of things that can be done to an audio signal to "improve it" or get as much out of it as possible. The CMII and its sidekick the Super Firmware are an excellent example of how an audio file can be processed using standard PCM decoding methods.
I am wary of all these companies dreaming up new and allegedly better methods to improve on one or more deficiencies in the PCM format. It sounds like an attempt at licensing revenues to me :rolleyes:.

On the subject of dynamic range: Give me 24 bit over 16 bit where possible. Listening to live music in the pub or local club surely isn't 16 bit. I don't feel that 16 bit does justice to any of the drinks that I got in my drinks cabinet.

[anthonyTD]

Tend to agree!
.

Impulse response is about the time response of the conversion and is effected by the necessary low pass pre-filtering at the A/D stage. The slower, more gentle these filters, the more accurate the time response. That leading edge of a sound is what tells the brain what to expect next, gives much of the character of the sound, timbre etc. In other words it being more accurate increases the realism of the recording. There's mention of impulse response at the end of this AES paper along with a nice graph of impulse response of the different resolutions: http://tech.juaneda.com/en/articles/dsd.pdf

[Rothchild]

Err, careful, I think we're mixing up dynamic range compression/limiting and data compression (as well as having Stanley's pub gig observations mixing in a confusion of overall volume with dynamic range - yes a pub gig is likely not 16bit, odd are it's far less (because the noise floor is so much higher)).

For understanding dynamic range, from a listeners perspective, a good place to start is racking up Bob Katz's 'Honour Roll' tracks as a playlist and noting how you incrementally have to wind the volume knob down to be able to listen to them tolerably.

Macca, I know you don't use computers but it's actually pretty easy to spot if something has been upsampled from a lower res master by using a spectrum analyser (such as the free 'Span' tool from Voxengo) - if it rolls off sharply at 20kHz it's a resample, if there's real content above that frequency then chances are it's the real thing.

My reading of the Meridian guy is that he actually says we need 18.2bits to fulfil the potential of human hearing (in terms of volume range) and this is rounded up to 20bits to allow headroom for mixing and processing, he also suggests that a 58kHz sample rate is enough but argues that, out of convenience to provide simple convergence with currently used rates this could be up-rated to 88.2kHz (2x44.1) or 96 (2x48kHz)

[Audio Advent]

I think Stan has a problem with the overall volume in his glass at the pub gig compared to at home and the drinks cabinet.

[Audio Advent]

The idea of replacing all of the recordings I have in 16/44.1 with the same recording at a higher sampling rate is something I don't regard as 'easy' especially since 1) I don't do downloads and 2) it is often impossible to discover if the download is a genuine 24/96 recording or a 16/44.1 that has been upsampled.

Not sure why you'd feel the need to replace anything. For me, high res is all about future recordings which have actually been recorded at high res with top notch gear. Hopefully labels won't be shy in saying how it was recorded (you can probably look up the studio and see what gear they generally use and make a guess).

The music of the past in your collection can just remain as it is. I guess you could make an argument for pulling out some master tapes and digitising them at highres, but you're certainly not going to hear the original recording in high res, all you will get is simply the digitisation of a tape which is now decades old with all its top end loss, playing on a different machine with different setup and all sorts of imprinting (or whatever it's called.. ghosting?). Whether that's something one wants a copy of is up to the buyer - in some senses it adds it's own character like a decomposing, mouldy cheese. Might as well get the CD if it was made 20 years ago from a much fresher tape (but I'd favour the vinyl anyway probably).

[Audio Advent]

My reading of the Meridian guy is that he actually says we need 18.2bits to fulfil the potential of human hearing (in terms of volume range) and this is rounded up to 20bits to allow headroom for mixing and processing, he also suggests that a 58kHz sample rate is enough but argues that, out of convenience to provide simple convergence with currently used rates this could be up-rated to 88.2kHz (2x44.1) or 96 (2x48kHz)

In my mind, if you want to record something so that you definately end up with a result which fulfils the potential of human hearing in it's replay, then I wouldn't sail close to those limits, I'd go far beyond them to be sure. So for Meridian Man's assertions, I'd take it to 24 bits and 174 or 192 kHz. The only limiting factors to consider would be cost of the same quality converters at the higher rates (192 is pretty standard at the high end now, other than Dan Lavry's products) and data storage costs which are small now.

[Rothchild]

As observed, I think the jury is out on the sonic benefits vs potential disadvantages of 192. (although I'd agree that seeing as we have 24bit and it's way more than we need then stepping 'back' to 20bit is a moot point and just there for academic interest) - I do all my recording at 24bit and make use of the ample headroom (tend to peak between -18 to -12dBFS on individual tracks - also makes interfacing with analogue outboard easier).

One practical benefit to recordists is that, with a suitably powerful computer and well written soundcard driver etc 192 would enable you to have sub-millisecond recording latency through the box (with effects monitoring) which finally means that computer recording is at the point that tape was at the end of the 70s.

[mkrzych]

Hello,
I've been listening to high resolution files for quite some time and after all of these years I must admit that first if the quality of sound has been prime object and if master files are the same, the difference with redbook vs PCM 24/96 is subtle and not always is so simple judgement to pay more for high resolution.

However, if especially big hall orchestra or good studio jazz recordings are considered and are recorded and processed all the way down in 24/96 than you may have hear the difference. This also depends of your audio chain equipment like source and speakers/cans mostly.

Now, where to look for difference. In mu humble opinion first is spaciousness and things like particular instrument timbre and its decay - piano for instance. Second is the instrument placement in the stereo image and potentially feeling of wider soundstage. Those differences may vary as well for each particular recording.

Examples: Try Dvorak cello concertos by Steve Isserlis on Hyperion in CD and 24/96. Or recent Keith Jarred remasters, especially Belonging album. Comparing older CDs with that is nite and day truly! But in this case could be mostly influence of better tape transfer, of course mastering and potentially at the end of high resolution format.

Oh, forgot to mention you must have pretty quite listening room (try during the nite) without ANY major issues like hum, power noise etc. Mine is around 30-35dB of noise. If not the case, listen over decent dynamic headphones with for instance CMII.

[Audio Advent]

As observed, I think the jury is out on the sonic benefits vs potential disadvantages of 192. (although I'd agree that seeing as we have 24bit and it's way more than we need then stepping 'back' to 20bit is a moot point and just there for academic interest) - I do all my recording at 24bit and make use of the ample headroom (tend to peak between -18 to -12dBFS on individual tracks - also makes interfacing with analogue outboard easier).

One practical benefit to recordists is that, with a suitably powerful computer and well written soundcard driver etc 192 would enable you to have sub-millisecond recording latency through the box (with effects monitoring) which finally means that computer recording is at the point that tape was at the end of the 70s.

I'd agree that subjectively the jury is out but that is the same with so many "improvements" in the audio world - only takes one person to say they can't hear a difference and to say everyone else is experiencing confirmation bias and the case is adjurned indefinately. The theoretical arguments for and against often come down to particular practice and then belief - really the matter should be decided subjectively, the basis of empirical science if the theory under scrutiny is about something sounding subjectively better to a human.. so the jury remains out as before.

The tape comment I'm not sure about, not that it really matters as it's off topic I suppose... still, I'll ramble about it anyway :D

Surely that's comparing slightly different things? To actually compare like with like, wouldn't we be monitoring off the tape read heads? So the delay would be down to the distance between record and read heads and the speed of the tape. For there to be 1 millisecond of delay between record head and read head, the distance between them would have to be less than 0.4mm (0.381mm) apart at 15 inches per second (38.1cm/s). In reality they are a good few cms apart.. To compare like with like, you'd have to take the monitoring feed from before the A/D process, tap into the filtering stage before the A/D which is not what happens.

I think what you're thinking about is the use of plugins and monitoring through them otherwise wouldn't you monitor through an analogue desk if you needed to? You need to set the levels in the analogue domain before the A/D conversion so hopefully whatever is setting those levels has a monitor out on it. Same will happen if using any outboard digital effects like a Lexicon PCM90 reverb or if any digital pedals on the guitar, or Line 6 Pod type amp cab emulation is used. Line a few effects up at a time and things will probably experience quite a lot of latency. I guess the trick is for any performer's monitoring to be on a seperate set of channels with seperate analogue reverb for any musicians who feel they need it in their cans..

Even in an analogue system though, roughly each foot a performer is from the microphone introduces a 1 millisecond delay.. And with no system in place at all, a 20 foot distance between two live musicians and they hear each other with a 20 millisecond delay etc etc

[Rothchild]

Yeah, I'm wandering ot with the tape comment ;-) probably should have said analogue as it's not about tape per-se.

Low latency is useful because you can give the talent a decent desk mix back to their cans, with a big old school studio it was much easier to give a monitoring mix back that's close to what's going to tape (without necessarily giving them the tape monitor - which as you observe is delayed by the distance between the record and play head). In practice less that 7ms tends to be acceptable and less that 3ms is pretty good - if we can get the 'straight through' latency down below 1ms that leaves some flex for some processing in the monitor mix (which adds up as you've pointed out) whilst keeping the overall delay down around that magic 3-7ms total.

Interesting though isn't it that were arguing about the relative merits of 50microsecond of timing error (one cycle of 20kHz) and suggesting that we need 10microseconds of accuracy (96kHz) for 'real' hifi and yet we might be happy to expect a musician to play their instrument whilst being fedback the signal from that instrument 100x slower than that. (practically it does depend a bit on what they're playing (drummers tend to be a bit more sensitive to latency in their monitors than other, I'll hold off the bassist jokes ;-))

[Audio Advent]

well... how musicians play and their relative timings is a different thing to capturing a soundfield and all the phase relationships which give our brains 3 dimensional cues. Many musicans are happy listening to other musicians on basic cheap hifi and not bothered about the mp3 quality - the musician's relationships with each other and the notes played are generally captured perfectly by low end recordings. High res and all that, IMO, is all about sound and not music. Good sound of course helps the music to be better received but is not entirely necessary, else I wouldn't get enjoyment from the car stereo..

[Liffy99]

This is where we get into those marginal grey areas which I accept do exist. That's why I mentioned the possibility that frequencies higher than 22KHz, whilst being inaudible to humans, may have an impact on the reproduction of lower frequencies. As you say the additional bits (dynamic range) are not doing anything useful and any case only relate to dynamics and not the 'quality' of the sound reproduction in any case.

The point of my OP was that hi-rez audio is not in any way like hi-rez television. I think it is important that enthusiasts should be aware of that as it is pretty basic and relevant information.

Aliasing may have an effect at supersonic frequecies (the reflection of artifacts above half the nyquist limit back down the frequency spectrum. But anti-aliasing filters sort this out by introducing a gap between that threshold and aliasing components. I think this is why 44.1khz was chosen to cover a desired 20khz upper listening range (2 x 20 = 40, the minimum Nyquist sampling rate needed and the other 4.1 to create the gap).

[Liffy99]

Best summay I have seen . . .

https://youtu.be/nLEhfieoMq8