Wait, no... The content past 20kHz just show that the file is a true 48 or 96kHz sample rate. It does not say anything about bitdepth (16 or 24). But logically, the only 20kHz brickwalled would be derived from 44.1kHz 16bit CD sources.
Now, there are a few exceptions - I know of one: RoxyMusic's Avalon analog tapes where recorded on the wrong kind of tape so, in order to "maintain" them, Sony baked the tapes for one last transfer to digital, analog tapes being destroyed in process. At that time, the only option was 44.1kHz SR, but this recorderd all the individual tracks (actually a little more that 44.1 because an error in settings of the varispeed). An ulterior mix in HD format of those digital files will produce the brickwalled spectrum at 20kHz, but it CAN have a higher bitdepth than 16 bit - when you "add" all 24 channels to obtain 5.1 or stereo mixes, you can get more resolution. This helped the surround remastering process that eventually was released on a SACD.
Last edited:
Right. I ran some of my 96Khz files thru another spectrum analyzer and see a brick wall at 48Khz, not at 22Khz. There is very little (if any) content up that high, but at least I can be pretty sure they aren't just up-sampled 44.1K files.
As for knowing if the file has remained 24 bit (or higher) all thru the process, I don't know how to determine that. Any ideas? Noise floor, maybe?
As for knowing if the file has remained 24 bit (or higher) all thru the process, I don't know how to determine that. Any ideas? Noise floor, maybe?
Noise floor is unlikely to tell you much- presumably, there were mike preamps used in the recording.
I would probably update the software, though, in light of the link Werner posted. The new versions aren't the best possible, but the artifacts are WAY down below where you're likely to hear anything.
I would probably update the software, though, in light of the link Werner posted. The new versions aren't the best possible, but the artifacts are WAY down below where you're likely to hear anything.
I define step function in the mathematical sense.
it looks like some people think that the initial hit of a percussive instrument looks like sudden (as in from sample "n" to sample "n+1") jump to digital max/min i.e. a need for extreme slew-rate.
if such extremely fast transients indeed existed in music, there would be reason to believe that sampling at a lower rate (44.1k) is not enough and that low-pass filtering at ~20kHz is audible.
attached is portion 22.311..22.314 from 3rd pic (tom) zoomed.
Attachments
The cymbals hit with hickory sticks have a very steep attack. Not as steep as the math fuction thou 
Harmonics can go easy past 48kHz... Now, how much of those are needed for a complete audio experience? Well, I doubt that our eardrums can move that fast. I think that 24@88.1 or 24@96 is enough SR for audition.
For mixing purposes, of course more is needed - to cover the eventual digital truncations while all that heavy processing is done (de-essing, notch filters, limiters, compressors, tracks volume adjusting, panning, etc).
Harmonics can go easy past 48kHz... Now, how much of those are needed for a complete audio experience? Well, I doubt that our eardrums can move that fast. I think that 24@88.1 or 24@96 is enough SR for audition.
For mixing purposes, of course more is needed - to cover the eventual digital truncations while all that heavy processing is done (de-essing, notch filters, limiters, compressors, tracks volume adjusting, panning, etc).
Last edited:
One problem with sampling at the Nyquest frequency is that there is the possibility that the samples occur at the zero cross of the waveform.
Output = 0
Granted we don't listen to single frequencies at the sample rate but this points out a weakness in the "Minimum Sample Rate" often proposed.
For this reason it is best to sample at a rate greater than 2X the max frequency. 5X is good. 10 X is better.
Output = 0
Granted we don't listen to single frequencies at the sample rate but this points out a weakness in the "Minimum Sample Rate" often proposed.
For this reason it is best to sample at a rate greater than 2X the max frequency. 5X is good. 10 X is better.
10X sampling might be better if all other things were kept equal - in reality they are not. All datasheets for DAC chips that I've seen have degraded specifications (SINAD for example) at higher sample rates. Go check some out - somebody noticed on this thread (1st post) for example:
Higher Sampling Rate -> Higher THD+N!
Higher Sampling Rate -> Higher THD+N!
Actually, that gives you no extra information, but it does make the filtering a bit easier. If the sample rate is 2.00000001 times the sampled waveform's frequency, the sample waveform is reproduced perfectly. All Nyquist says is that sample rate needs to be larger than 2x.
Do you have any idea of what the anti-aliasing filter does? What it really does?
The closer you are to the limit, the better your filters need to be. Both anti-aliasing and reconstruction filters. Reconstructing from 2.000001 times the max frequency might seem difficult or counter-intuitive but in theory it can be done. In practice you would need very good filters.
I used to puzzle over the problem of 'near the limit' sampling. How can you reconstruct when many of your samples are near zero? The answer is that a bandwidth-limited waveform is constrained, so that to some extent the future is contained in the past. The reconstruction filter knows the past, so it has enough information to reconstruct the present.
Another way of looking at it is to regard sampling as being a bit like AM modulation. Sampling a 21kHz wave at 44kHz sampling rate is a bit like modulating it onto a 22kHz carrier. Of course you get both sidebands, so then the reconstruction filter has to remove the upper sideband (also known as images).
I used to puzzle over the problem of 'near the limit' sampling. How can you reconstruct when many of your samples are near zero? The answer is that a bandwidth-limited waveform is constrained, so that to some extent the future is contained in the past. The reconstruction filter knows the past, so it has enough information to reconstruct the present.
Another way of looking at it is to regard sampling as being a bit like AM modulation. Sampling a 21kHz wave at 44kHz sampling rate is a bit like modulating it onto a 22kHz carrier. Of course you get both sidebands, so then the reconstruction filter has to remove the upper sideband (also known as images).
Filtering does seem to have a noticeable effect on the sound, good or bad. Thanks for the technical view, guys.
FWIW, because of this thread I decided to test whether I could hear a difference between 24/96 and 16/44. So far, I can not. o: Oh sure, I can hear subtle differences when I know which files are which, but blind? Not at all. Embarrassing, really.
I'm so convinced there "has to be" a difference that I'll keep trying. Will let you know if I find it. So far, it's not looking good.
FWIW, because of this thread I decided to test whether I could hear a difference between 24/96 and 16/44. So far, I can not. o: Oh sure, I can hear subtle differences when I know which files are which, but blind? Not at all. Embarrassing, really.
I'm so convinced there "has to be" a difference that I'll keep trying. Will let you know if I find it. So far, it's not looking good.
it depends a lot on the player I guess. some say that very few manufacturers managed to get RedBook playback right, so I'd rule out any test done with cheap players. IMO the only valid test would be to downsample hires material to 44.1/16 not test with what could be different masters.
I have a hires-capable DAC but I don't obsess over it because overall my system is far from the best sound I've ever heard from CD
I have a hires-capable DAC but I don't obsess over it because overall my system is far from the best sound I've ever heard from CD
The big improvement lies in 24bit bit depth and the increased volume dynamic level discrimination gained. There is currently little point in exceeding 96K sample rate.
I hear big improvements up to 24/96k but not higher.
CD 16/44.1k is now obsolete and HDCD 16/88.2 likewise.
24/96k .flac should be industry standard for media, compatible with DVD or Blu-ray.
You can buy a cheap 24/192 DVD player(£80) that will outperform the most expensive CD player(£3000).Of course you need a decent HIFI amplifier etc.
There are plenty of digital and excellent analogue master tapes out there to resample.
Why is it not happening? eventually it will.
I hear big improvements up to 24/96k but not higher.
CD 16/44.1k is now obsolete and HDCD 16/88.2 likewise.
24/96k .flac should be industry standard for media, compatible with DVD or Blu-ray.
You can buy a cheap 24/192 DVD player(£80) that will outperform the most expensive CD player(£3000).Of course you need a decent HIFI amplifier etc.
There are plenty of digital and excellent analogue master tapes out there to resample.
Why is it not happening? eventually it will.
Flawed Article?
I'm new here and could be wrong, but reading the article and posts I think the article makes one gross error. OK. Music is analog. A full symphony playing in a hall live, acoustic has 80 or so instruments of different types with complex waves of over and undertones in addition to the primary tone with different dynamics for each, responding to the players' ministrations, and in turn, the emotions in the music, the technique of the player, the hall itself adding its own character and finally it reaches your ear.
That sound pressure wave is complex. Utterly complex. The author talks about a sort of granularity of the auditory nerves in the cochlea, but we don't just hear with our ears. A tympani up close thumps your chest. A deaf person can "hear" when someone he isn't facing has started to talk by holding a newspaper tight. It isn't just the ears, and I'm not sure even the ears can be reduced to the neat map the author lays out.
So, music is an analog wave or, if you will, waves within waves of great complexity in frequency and amplitude. You can tell if three or thirty violins are playing the same note and it isn't just amplitude. The sound is richer.
Digital tries to recreate this analog world by a series of steps. Completely right angled. The author says, "the signal recovers the exact smooth shape of the original (blue) when converted back to analog." That pinned MY BS meter. There isn't enough information to make the curve. Stuff happens between the steps, and the curve before the step (or steps) is not a perfect guide to what happens next. It was an extrapolation too. Extrapolation is nothing more than an approximation. I'd argue that once you digitize, you can never get back to the analog signal. Approximation is the best you can hope for.
It's like that old sophist argument. Walk from the middle of a room to the wall. Easy. You reached it in a few seconds. Linear process. Now, walk halfway there. Now, halfway again. One half, 3/4. 7/8,15/16...How long will it take for you to reach the wall? Answer: Not in a trillion iterations. You will never reach the wall. Never. Will you get there for all practical purposes? Perhaps. But the theory is sound. In an absolute sense you're never going to reach the wall.
So, digital files can make finer, and finer steps, and the conversion filters do a better and better job of extrapolating between the steps, so higher resolution is better. But that curve is not described by digital sampling of it. It is describable perhaps by calculus, but not integer arithmetic. And the calculus would get unwieldy with even solo flute, let alone brass band complexity.
The 192Khz of the better files does not correspond to wavelengths above human hearing, it corresponds to samples per second, and there we need all we can get. As I understand it, the Khz would correspond to the length of the steps, and the 24 bit the height differences in the steps as the digital file tries to approximate the analog waveform. Not that there are 24 heights, but that the heights are described in a 24 place binary "word". The biggest mistake of the article seems to me to be the confusion between Khz in a digital, computer sense and sound pitch in Khz.
In short, I think 192/24 should be clearer and worth the money. Would that audio could be analog from start to finish, but that ship has sailed. There's plenty of digital only release music that I can't live without, so I'm trying to get back to the analog, back to the garden after biting that apple and it ain't easy.
Now this guy seems very bright and all, and I might be wrong, so comments welcome.
I'm new here and could be wrong, but reading the article and posts I think the article makes one gross error. OK. Music is analog. A full symphony playing in a hall live, acoustic has 80 or so instruments of different types with complex waves of over and undertones in addition to the primary tone with different dynamics for each, responding to the players' ministrations, and in turn, the emotions in the music, the technique of the player, the hall itself adding its own character and finally it reaches your ear.
That sound pressure wave is complex. Utterly complex. The author talks about a sort of granularity of the auditory nerves in the cochlea, but we don't just hear with our ears. A tympani up close thumps your chest. A deaf person can "hear" when someone he isn't facing has started to talk by holding a newspaper tight. It isn't just the ears, and I'm not sure even the ears can be reduced to the neat map the author lays out.
So, music is an analog wave or, if you will, waves within waves of great complexity in frequency and amplitude. You can tell if three or thirty violins are playing the same note and it isn't just amplitude. The sound is richer.
Digital tries to recreate this analog world by a series of steps. Completely right angled. The author says, "the signal recovers the exact smooth shape of the original (blue) when converted back to analog." That pinned MY BS meter. There isn't enough information to make the curve. Stuff happens between the steps, and the curve before the step (or steps) is not a perfect guide to what happens next. It was an extrapolation too. Extrapolation is nothing more than an approximation. I'd argue that once you digitize, you can never get back to the analog signal. Approximation is the best you can hope for.
It's like that old sophist argument. Walk from the middle of a room to the wall. Easy. You reached it in a few seconds. Linear process. Now, walk halfway there. Now, halfway again. One half, 3/4. 7/8,15/16...How long will it take for you to reach the wall? Answer: Not in a trillion iterations. You will never reach the wall. Never. Will you get there for all practical purposes? Perhaps. But the theory is sound. In an absolute sense you're never going to reach the wall.
So, digital files can make finer, and finer steps, and the conversion filters do a better and better job of extrapolating between the steps, so higher resolution is better. But that curve is not described by digital sampling of it. It is describable perhaps by calculus, but not integer arithmetic. And the calculus would get unwieldy with even solo flute, let alone brass band complexity.
The 192Khz of the better files does not correspond to wavelengths above human hearing, it corresponds to samples per second, and there we need all we can get. As I understand it, the Khz would correspond to the length of the steps, and the 24 bit the height differences in the steps as the digital file tries to approximate the analog waveform. Not that there are 24 heights, but that the heights are described in a 24 place binary "word". The biggest mistake of the article seems to me to be the confusion between Khz in a digital, computer sense and sound pitch in Khz.
In short, I think 192/24 should be clearer and worth the money. Would that audio could be analog from start to finish, but that ship has sailed. There's plenty of digital only release music that I can't live without, so I'm trying to get back to the analog, back to the garden after biting that apple and it ain't easy.
Now this guy seems very bright and all, and I might be wrong, so comments welcome.
The same guy has a new video up recently which might help you with your apparent misunderstandings here. Its not extrapolation, rather its fitting a curve with the constraint that the original is band-limited. He explains this well enough for me at any rate.
New Xiph.Org video: "Digital Show & Tell"
And further to that: if you play around with a musical file editor like Audacity, you can verify yourself that software, computing power, DSP can extract "perfect" analogue waves from seemingly far too few bits of information. For example, you can generate a 20kHz sine wave within Audacity at the CD 44.1kHz rate ... and zoom in and see a garbled mess of samples that don't make sense. Yet, you can resample that waveform to higher and higher rates, I've gone up to 5.6Mhz as an exercise, double DSD rate, and, magically, out pops a perfect, very analogue looking, sine wave. No distortion, no nasty spurious noise, it just works ...
Frank
Frank
Your BS meter needs to be sent away for recalibration. Given a band-limited signal, digital really can reproduce it. No stuff happens between the steps. That is precisely why it has to be band-limited.Scott6113 said:
However it arises, the music signal we have is just two channels of a voltage. Converting the complicated vector sound velocity field in the concert hall to just two scalar signals is the recording engineer's art. From then on, it is just science. The art recommences, to some extent, in loudspeaker and room design.
Lots of the confusion and nonsense surrounding digital audio boils down to people confusing:
- bandlimiting
- sampling
- quantisation
Then mix in some Fourier denial (or Fourier ignorance) and you have lots of people pooling their confusion and so creating false memes.
"out pops a perfect, very analogue looking, sine wave"
That seems to me to be something like the difference between peudorandom and random numbers. The result is equivalent. The origin differs. I'm not convinced though. Just because the signal surpasses the resolution of the scope or your eyes does not mean it becomes the exact same wave. You're still squaring the circle. Can you get the billion angles to the point that your eyes can't seem them and the pen and ink into the paper fibers just blend and make the circle? Yes. It didn't come from a Compass though. Then we're arguing whether the difference is perceptible, not whether there is one.
I agree that we are just dealing with two voltages for stereo. Measure the voltages with a digital voltage meter and you will always get two numbers at a given point in time. In reality those numbers are infinite, as are the points in time. Both are in continuous fluid motion. Everyone can agree that a ballerina's grace can't be appreciated from a still photograph. It can't be appreciated by a sequence of one a second. Thirty a second and you can see it all, since our eyes and brains see that as smooth motion. But it isn't. It is a fast series of stills. And yet you say "digital really can reproduce it" when the fact you used: "reproduce" shows that you have altered the original by digitizing it. Why go to an art gallery when there are hi-rez files available?
There is a difference, and the question is, how much of a difference is perceptible. Once again, we need the double-blind test and no one seems willing to undergo such a thing.
That seems to me to be something like the difference between peudorandom and random numbers. The result is equivalent. The origin differs. I'm not convinced though. Just because the signal surpasses the resolution of the scope or your eyes does not mean it becomes the exact same wave. You're still squaring the circle. Can you get the billion angles to the point that your eyes can't seem them and the pen and ink into the paper fibers just blend and make the circle? Yes. It didn't come from a Compass though. Then we're arguing whether the difference is perceptible, not whether there is one.
I agree that we are just dealing with two voltages for stereo. Measure the voltages with a digital voltage meter and you will always get two numbers at a given point in time. In reality those numbers are infinite, as are the points in time. Both are in continuous fluid motion. Everyone can agree that a ballerina's grace can't be appreciated from a still photograph. It can't be appreciated by a sequence of one a second. Thirty a second and you can see it all, since our eyes and brains see that as smooth motion. But it isn't. It is a fast series of stills. And yet you say "digital really can reproduce it" when the fact you used: "reproduce" shows that you have altered the original by digitizing it. Why go to an art gallery when there are hi-rez files available?
There is a difference, and the question is, how much of a difference is perceptible. Once again, we need the double-blind test and no one seems willing to undergo such a thing.
That's been done quite a few times (a few minutes of literature searching will be of great use for you)- and you can even do it yourself quite easily with an hour or two of setup.
To generalize DF96's answer, whenever you don't intuitively understand a fundamental physical or mathematical principle, ask yourself, "What technologies rely on the truth and accuracy of that principle, why do they work, and why haven't tens of thousands of brilliant mathematicians, physicists, and engineers had the same profound insight as I just had? Could the problem be my understanding?" In the case of sampling and reconstruction, airplanes fly by wire, spacecraft successfully insert themselves into orbit around distant planets, MRIs successfully image organs and diagnose disease, industrial process control systems successfully achieve production tolerances, CNC machines accurately produce metal parts to predicted tolerances...
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.