Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Well, subsampling kind of IS a low-pass filter. And that doesn't change the fact that you cannot accurately represent the amplitude of a sine wave by sampling at half its frequency.

Nyquist specifically says below half the sampling frequency, the theorem itself says you can't sample exactly half the sampling frequency.

Not sure what you mean by subsampling and it being a low pass filter.

Who cares what Nyquist says?

Always a fun comment to read in a discussion of digital audio.

Draw a sine wave, and put some dots on it. If you sample the waveform at anything less than around 3x the signal frequency, you get severe beating of the signal.

Beating?

If you sample at exactly 2x, you cannot tell anything about the signal amplitude, except that it is at least a certain value.

You don't sample at exactly 2x, the theorem specifically says that won't work. But for frequencies under that it does capture both the amplitude and frequency. And if you do try graphing it out and drawing the dots you do see that.

If you get to choose where your sample points are, then a minimum of 2x is fine. If you don't, and they are forced to be evenly spaced, you need to use 3x.

Nope. Anything under half the sample rate works out. It's only a problem if there are frequencies at 1/2 the sampling rate or higher that aren't filtered out before sampling. Works with actual recording and works when you graph a sine wav on paper.
 
You're right. I'll correct that to "nobody who actually has a clue about the stock market". At least not based on a three percent stock move, if a stock dropped 10 or 20 or 50 percent in a day then I'd totally agree.



Last three percent drop was September 2. And if you go back and look at the numbers (which is what we should do if we want to compare relative to previous iPhone releases), for virtually every iPhone release there was a day following the release with a similar one day drop. A couple only in the 2.2-2.3% range, but others 4-6%. In one day.

So were all of those releases "relative flops"?



Well informed people certainly don't make an overall judgement about the success of a product release based on the stock market performance for one day (not even release day). Especially when the next day sees the stock bouncing right back.



So assuming the stock market did think that last week based on that one day...you're saying they thought that and then the very next day did a 180 and suddenly went back to thinking it was a good product?


Whoa, whoa, whoa. Where did you get the numbers for virtually every previous phone release? If it's a common thing then I'd have to ratchet back my claims a bit. Everything I've read says this drop was pretty rare.

And I don't know, man. Why did it bounce back? I only know why it dipped.
 
He is probably referencing the apparant "wavy" amplitude pattern when looking at near-Nyquist frequencies in sound editor programs which don't apply the right kind of interpolation on the waveform display. Ref: http://www.windytan.com/2014/01/misleading-representations-of-discrete.html

Even though it looks like the sound would be "beating", it is not if the output DAC is modern (meaning, from the 1990's or later).

Great link, thanks.

So to clarify, a signal that is sampled at just over twice its frequency is reproduced 100% accurately (shocker!), but some audio editing programs that cut corners when displaying the waveform might not make that obvious.
 
You don't sample at exactly 2x, the theorem specifically says that won't work. But for frequencies under that it does capture both the amplitude and frequency. And if you do try graphing it out and drawing the dots you do see that.

OK, two plots, sampling at 2.1 and 3.1 times the wave frequency.

So you're trying to tell me that the first graph, @ 2.1, still accurately portrays the waveform?
 

Attachments

  • 2p1 sample.jpg
    2p1 sample.jpg
    232.9 KB · Views: 117
  • 3p1 sample.jpg
    3p1 sample.jpg
    248 KB · Views: 119
Neither of your red lines accurately portrays the waveform because you just linearly interpolated between the points that were sampled. That's not what a digital-analog converter does. A DAC finds a sine curve that fits the points that you sampled (i.e. the vertices of your red line) and then spits out an analog signal that is 100% identical to your input.

Check the last picture in jont-fu's link for a more accurate representation of how the waveform is reconstructed from individually sampled points.
 
He is probably referencing the apparant "wavy" amplitude pattern when looking at near-Nyquist frequencies in sound editor programs which don't apply the right kind of interpolation on the waveform display. Ref: http://www.windytan.com/2014/01/misleading-representations-of-discrete.html

Even though it looks like the sound would be "beating", it is not if the output DAC is modern (meaning, from the 1990's or later).

So what happens if the true signal IS supposed to be varying in amplitude?

Linear interpolation is the most correct method to use, as anything else assumes you know more about the signal than you actually do.
 
So what happens if the true signal IS supposed to be varying in amplitude?

It's still captured.

This is just off the top of my head, but you can verify this for yourself using Audacity or something. Not sure if I'm listing all the right steps here, but bear with me. Create a new file with a 96 kHz sampling rate, generate a 20 kHz tone, give it a fade-in or fade-out or whatever to vary your amplitude, and save it.

Start a new file using save-as. Resample down to 44.1 kHz -- this is the step that will supposedly lose so much precious information according to the Nyquist-deniers :). Then resample back up to 96 kHz so you can compare directly with the file from the previous paragraph. Invert the signal. Save. Add this waveform with your original file and export this as a new wav file.

Open the final wav file in Audacity and there will be no output whatsoever, because the two signals perfectly canceled each other out, and no information was lost when you downsampled to 44.1 kHz.

Linear interpolation is the most correct method to use, as anything else assumes you know more about the signal than you actually do.

In certain other fields or applications, this might be correct. But it's actually impossible to generate an analog signal that looks like your straight red lines from the sampled points you were given -- unless, of course, you choose to generate that analog signal at a *much* higher frequency than the points that you were given indicate, which is what you actually did. In other words, drawing straight lines between sampled points is actually the method that assumes too much.
 
So what happens if the true signal IS supposed to be varying in amplitude?

Linear interpolation is the most correct method to use, as anything else assumes you know more about the signal than you actually do.

We DO know that the signal does NOT contain frequencies above the Nyquist, and simply linearly interpolating the points (or producing stair-steps) would contain such harmonics. A proper interpolation function will reproduce the signal in the only way it could possibly be reproduced!
 
By the way, this dude does a really great job of demonstrating (not just telling you) how digital/analog conversion works:
https://www.youtube.com/watch?v=cIQ9IXSUzuM

The video is 24 minutes long, but who ever said understanding this stuff was easy? :)

Yeah, this guy is the best.. and he has also the coolest analog equipment to actually show what happens in the DA/AD process! I have watched it at least three times and every time I learn more about sound technology or presentation.
 
In certain other fields or applications, this might be correct. But it's actually impossible to generate an analog signal that looks like your straight red lines from the sampled points you were given -- unless, of course, you choose to generate that analog signal at a *much* higher frequency than the points that you were given indicate, which is what you actually did. In other words, drawing straight lines between sampled points is actually the method that assumes too much.

Drawing the lines was a mistake on my part, I did not intend to suggest that the signal would look exactly like that, only give an indication of how the signal would be degraded. I should have just put in dots.

Despite that, pulling out the high frequency information using a Fourier transform still results in an incorrectly oscillating signal.

Unfortunately I'm not a mathematics major, so I'm not sure how to prove/disprove the theory either way, but I find it very difficult to believe that there is a unique solution to a set of evenly spaced sample points which contains frequencies up to, but not including, half the sample rate.

Unfortunately I don't have 'Audacity'.
 
Unfortunately I'm not a mathematics major, so I'm not sure how to prove/disprove the theory either way, but I find it very difficult to believe that there is a unique solution to a set of evenly spaced sample points which contains frequencies up to, but not including, half the sample rate.

That's exactly what the Shannon sampling theorem postulates:

If a function x(t) contains no frequencies higher than B cps, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart.

You can try to follow the mathematical proof from the article on Wikipedia, although it will be likely difficult to follow without some mathematical background.

I actually appreciate your skepticism, but note that the theorem is very old, is mathematically proven to be correct in theory and is proven to be correct in a lot of practical applications.
 
I was on another forum where they took a 16 Bit sound file and up sampled it to 24/96 and compared it to a 24/96 version and there was at least one person that did pass the ABX test.

Others have mentioned you should be downsampling, not upsampling to conduct this test, and they are correct. That aside, one person does not a sample make; someone has to be lucky occasionally...

It is possible to pass an ABX test. Now, I have a 16 Bit rebook of an album that was subsequently released in 24/96 or 176and there was a HUGE difference. There might have been a difference in the amount of audio compression during the mastering process, but there are definitely sonic differences at least in certain recordings that were originally done on analog tape. So, these 24 bit conversions from analog can many times be a lot better than the 16 Bit Redbook versions originally released. It's hard to tell how much of the difference is based solely on just being higher resolution because they were converted from analog using different equipment at different time periods, etc. But the bottom line is the 24 Bit versions sound a LOT better and it's a LOT more noticeable that a lot of people could easily hear a difference blind folded.

This is a separate issue that has nothing to do with bit depth / sample rate. Yes, there is a tendency for 24/96 re-releases of recordings to be produced from superior masterings than the original releases (although ironically HDTracks has been known to use some terrible masters for their releases, see Red Hot Chili Peppers...), but this is no argument for higher bit depth / sampling rates per se. I often download 24/96 releases downsampled to 16/44.1 for this reason.


To address your $150 HDMI cable statement. here's the scoop in a nutshell

HDMI has video and audio.

With video, you need higher bandwidth over long distances for certain applications. If you have a 4K projector and want the best performance and need a 50 ft run, you are going to have to get the more expensive cables. Especially if you need 18Gbps second bandwidth, the more expensive cables will have at least 10.2Gbps over long cable runs, the cheaper cables generally only go about 15 ft before they lose bandwidth, so in certain applications, you have to get the more expensive cables for video.

In the audio portion of HDMI, the more expensive cables simply have less noise problems which result in cable timing issues which create digital distortion known as jitter. The more expensive cables have less noise creating less jitter resulting in better audio. Now, if you don't have high end equipment and have long cable runs, then it doesn't matter, but for those that are using higher end equipment and have longer cable runs, then the cable is a more important factor. Has this been proven? Yes, it has.

Now, in the audio world, people that download or rip digital audio files to their computer aren't using HDMI to go from the computer to their stereo system to listen to audio. Most computer audio systems are using USB from the computer to the DAC. Is there a difference in USB cables? For some people/equipment there can be audible differences because you have the issue with USB as it has both data and power running along side one another and the power creates noise which can effect the data. Some higher end equipment running high bit and sample rates need a consistent and high bandwidth, otherwise it doesn't work. The cheap USB cables many times won't even work with some of the ultra high end equipment because they demand quality cables, so there are high end DAC mfg that have to have high end USB cables to work. Digital signals are not 1's and 0's, there are electronic pulses and in playing audio, those pulses have to have proper timing, no errors, etc. That's why these high end cable mfg crawl out of the wood work because there are high end equipment mfg and people listening to this equipment can hear subtle differences if they have trained listening abilities.

Just to see what cable has better transfer rate, I took two USB 2.0 cables and ran speed tests and was able to get better results in a speed test with a cheap cable compared to a more expensive cable, so there was one test that showed a difference in USB cables for just data transfer.

Now, if you can't hear the difference, then don't spend the money, but if you can, that's another story.

This is just plain silly. HDMI carries a digital signal. Digital signals are either transmitted in full or they are not; there is no in between. It is true longer distances require higher-spec cables to carry a signal, but determining if you need such cables is a triviality: if the cable isn't up to spec for the distance in question, it simply won't work, if it is, it will. There is absolutely nothing to be gained in investing a single dollar more than the minimum required to get a cable that will produce a signal.


There is a lot of subjectivity in audiophile land, but this is one area where there simply is not (better still this can be demonstrated with math; debunking other areas like amps, analogue cables and DACs requires ABX testing). Anyone who tells you otherwise is selling you snake oil. Indeed, speakers/headphones may be the only area where there even is any subjectivity at all.
 
What, specifically, is wrong?

Your entire argument about LP filters (specifically addressed by the paper I linked), the claim that "24-bit is even more compelling" (again, specifically refuted by the linked paper), the idea that higher-quality equipment enables one to hear the difference between 24/96 and 16/44.1 audio, so, basically your entire post, which should explain my generalized response.
 
Others have mentioned you should be downsampling, not upsampling to conduct this test, and they are correct. That aside, one person does not a sample make; someone has to be lucky occasionally...



This is a separate issue that has nothing to do with bit depth / sample rate. Yes, there is a tendency for 24/96 re-releases of recordings to be produced from superior masterings than the original releases (although ironically HDTracks has been known to use some terrible masters for their releases, see Red Hot Chili Peppers...), but this is no argument for higher bit depth / sampling rates per se. I often download 24/96 releases downsampled to 16/44.1 for this reason.




This is just plain silly. HDMI carries a digital signal. Digital signals are either transmitted in full or they are not; there is no in between. It is true longer distances require higher-spec cables to carry a signal, but determining if you need such cables is a triviality: if the cable isn't up to spec for the distance in question, it simply won't work, if it is, it will. There is absolutely nothing to be gained in investing a single dollar more than the minimum required to get a cable that will produce a signal.


There is a lot of subjectivity in audiophile land, but this is one area where there simply is not (better still this can be demonstrated with math; debunking other areas like amps, analogue cables and DACs requires ABX testing). Anyone who tells you otherwise is selling you snake oil. Indeed, speakers/headphones may be the only area where there even is any subjectivity at all.

HDTracks isn't doing the mastering. The Record labels send the masters to various studios who do the actual mastering and conversion, HDTracks only sells what the record labels give them.

Yes, in some cases all they are getting is 24/44.1 versions and there is not much difference to the original 16 Bit. Some of their content is actually still in 16 Bit form. What I see HDTracks has in SOME cases is the record label has the original analog tapes either reconverted to 24/96 or higher and they simply remove a lot of the compression they used in the Redbook CD version and they sometimes do other things in addition. What exactly, I don't know, but I've compared the 16 Bit rebook to the 24/96+ and there is a HUGE difference.

Unfortunately, there is no set standard on what's being done. The first issue is how was the recording originally done, Analog or Digital. Then there is a matter of do they upsample or downsample or if they use less/more compression, etc.

Some of their recordings are simply converted from DSD (SACD masters) to PCM, that's a fairly simple process, but i don't know how many of the preexisting albums were archived in DSD to begin with.

Some of the newer digital recordings were originally done at 24/96 or higher and it makes so much sense to offer those recordings in both 16 Bit MP3/ACC for the mobile crowd AND 24 Bit versions that are left alone for the home audio enthusiasts crowd.

As far as what some of these non-audiophiles say, sometimes they have a valid point and sometimes they don't. The problem in the audio world is for there to be set standards for being able to do more quality of sound measurements that are valid and repeatable and used throughout the audio industry.

A lot of measurements these companies are making don't exactly analyze quality of sound. Looking a 1kHz sine wave doesn't tell you anything because music is NOT a consistent one frequency sine wave. Sine waves only exist on a test bench, not with music. With music, you have rise time, sustain, decay, and harmonic structure, etc.

Now, in terms of HDMI, most people get hung up on just the video portion. If you look at 4K video, you have different bit depths and fps, the higher bit depths and fps require higher bandwidth and in the case of certain installations, you need longer distances and these longer distances and higher bandwidth of 18Gbps require higher quality cable and in the longer runs, they need the more expensive silver plated copper or solid silver, with gas injected foam insulation, etc. And the terminations have to be well made and lots of good shielding. The cheap garden variety HDMI cables simply won't go long distances at 18Gbps, they are only rated at 10.2Gbps and for shorter distances. Then there is a thing called CL rated so it be installed in wall, not all cheap HDMI cables can be installed in wall. In terms of what works and what doesn't. I've seen tests where the cheap cables simply don't perform well when connected to lots of different devices, even for shorter distances, some have too much sparkles. Now, if you don't have high end equipment, then it's not as big of a deal because if it works, it works, but if it doesn't, then it doesn't. But when it comes to AUDIO portion of HDMI, these higher end installations are using up to 32 Channels, whereas the lower entry level systems are 5.1 systems. There is jitter in the audio portion of HDMI, PERIOD. The more expensive cables will simply have less jitter and that is measurable and that is audible in the higher end systems. The majority of consumers, I agree, will probably do fine with a cheap cable, but some will have to at least go to Monster's highest end cable, which is less than $100 (which is completely reasonable) and then there are those that are installing more expensive equipment in a much bigger installation and they will hear/see a benefit with the more expensive WireWorld or some other brand of higher end cable. The higher end crowd is using far more expensive calibration equipment for video and they are running their projectors at high color depths and fps vs the low to mid range consumer, in those cases they need the higher bandwidth.

I agree there is a lot of subjectivity in audio, you can't simply measure equipment and just use technical specs to choose a product, it only helps narrow down products. You simply have to hear the equipment in your setting with the music you play because you are the judge of the sound quality and it varies from person to person as to what they like and dislike.

I think there is simply a long ways to do before they can really determine quality of sound with JUST measurements or a spec. Again, making a generalization about 16 Bit vs 24 Bit is kind of crazy since there are varying quality of AD/DA converters with vastly different specs. There are some DACs that can play 16 Bit Redbook VERY closely to a high end turntable where you get the same emotional connection with the music. Why that is, I couldn't tell you. CDs for the longest time were very flat sounding, etc. and that may be partly due to the way the CD was mastered or the DAC or both. But they are getting much better with these DAC designs over what they had 30 years ago.
 
As far as what some of these non-audiophiles say, sometimes they have a valid point and sometimes they don't. The problem in the audio world is for there to be set standards for being able to do more quality of sound measurements that are valid and repeatable and used throughout the audio industry.

If you mean "quality" overall in the production of music, I agree, but it's important to determine which factors play a role and which factors are transparent.

In the context of bit depth, sampling and "hd audio" in general, the goal is fidelity: the ability to accurately reproduce audio. The tests are pretty clear about fidelity: 16/44 is enough and increasing bit depth/sampling rate does not provide a discernible improvement of fidelity at realistic sound volumes (it gets discernible at unrealistically high sound volumes).

This basically means that once you get your final master at whatever high bit depth/sampling rate you want, correctly downsampling it to 16/44 is transparent to the human hearing. Of course if you start your downsample from something which sounds bad, you'll end up with something which will sound exactly as bad, and if you start from something which sounds very good you will end up with somethingh which sounds exactly as good.

A lot of measurements these companies are making don't exactly analyze quality of sound. Looking a 1kHz sine wave doesn't tell you anything because music is NOT a consistent one frequency sine wave. Sine waves only exist on a test bench, not with music. With music, you have rise time, sustain, decay, and harmonic structure, etc.

From the point of view of the representation of sound, music is the sum of many different sine waves so mesuring a single sine wave is maybe not enough, but a good starting point.
 
If you mean "quality" overall in the production of music, I agree, but it's important to determine which factors play a role and which factors are transparent.

In the context of bit depth, sampling and "hd audio" in general, the goal is fidelity: the ability to accurately reproduce audio. The tests are pretty clear about fidelity: 16/44 is enough and increasing bit depth/sampling rate does not provide a discernible improvement of fidelity at realistic sound volumes (it gets discernible at unrealistically high sound volumes).

This basically means that once you get your final master at whatever high bit depth/sampling rate you want, correctly downsampling it to 16/44 is transparent to the human hearing. Of course if you start your downsample from something which sounds bad, you'll end up with something which will sound exactly as bad, and if you start from something which sounds very good you will end up with somethingh which sounds exactly as good.



From the point of view of the representation of sound, music is the sum of many different sine waves so mesuring a single sine wave is maybe not enough, but a good starting point.

Sine waves measurements are good for checking phase measurements, but to determine sound quality? NOPE. What they SHOULD be measuring are musical notes played by various musical instruments over the complete range of the instrument to see how well the audio component you are working with will accurate preserve that note. The harmonic structure contains harmonics, in harmonics, overtones and at different amplitudes, and they differ from one instrument to another. Since all audio equipment acts like a filter, the harmonic structure will alter and it's trying to preserve that harmonic structure along with the rise time, sustain and decay time of each note being played.

Jitter plays a factor in digital audio, so that will effect sound quality. But different DACs have different amounts of jitter.

The problem with looking at theories, is they are just that, a theory. Heck, most people in the audio world use the THEORY that the speed of sound is 1,122 feet per second, but in reality, it isn't. It changes due to the environment. Air temperature can effect the speed of sound. So please don't go by THEORY, because the reality is what and how an actual physical device performs.

I've read that 16/44.1 was good enough that digital audio experts thoughts and then they changed their stance on it and then later came out with 24/96 was good enough, now there are others that are arguing that DSD 2x is even better.

I think in the context of the average consumer, 16 Bit is probably good enough since the average consumer doesn't have high quality playback equipment and "trained" ears that can discern subtle differences in audio quality, not to mention good soundproofed and treated listening rooms, but some people do.

When we get to recordings, they vary too much. I've heard great 16 Bit recordings, but I've awful ones as well, same goes for 24 Bit. DACs aren't perfect, but the theories assume that they are. That's that problem with going by theories. Theory assumes everything is perfect, when in reality, it isn't.

They are still making improvements in DACs and how they can improve the sound quality. The best sounding DACs are unaffordable by the average consumer, unfortunately and to get a decent sounding DAC costs vastly more than the average consumer is going to spend. There's one mfg that makes a DAC that takes PCM signals and converts it to DSD, it's supposed to sound better than just keeping it in PCM. I haven't heard it in person, but the product just hit the market and is being touted as one of the best DACs made by the reviewers that compare it to other high end DACs. Plus we are dealing with both older recordings that were originally done in analog and those tapes are old and might have some degree of degradation to them, plus the majority of recordings were originally done in 16 Bit. There are ways to upsample it and run it through filters, whether or not it sounds better is up in the air. But it's generally a safer bet to record in high bit depths and sample rate and down sample than it is to upsample. Since the market is pretty much 16 Bit since most computers only have 16 Bit DACs, putting out 24 Bit won't matter to the average consumer, but as consumers are buying more and more equipment with 24 Bit DACs, there is a small niche group of people that want 24 Bit recordings and they want the old analog tapes reconverted using better equipment and to not use audio compression during the mastering process. Unfortunately, the content isn't coming out fast enough at the higher resolution.
 
Last edited:
Looking a 1kHz sine wave doesn't tell you anything because music is NOT a consistent one frequency sine wave. Sine waves only exist on a test bench, not with music. With music, you have rise time, sustain, decay, and harmonic structure, etc.

So we've been having a discussion lately about whether a sine wave at just under half the sampling frequency can be captured perfectly. The answer is yes. This means that all frequencies below that frequency are also perfectly captured. Any frequencies higher than that frequency is not captured, because they were bandlimited out before sampling, and that's fine, because you can't hear them.

The point is, music absolutely is a set of sine waves superimposed upon one another. Rise time/sustain/decay/harmonic structure are all perfectly captured by the right superposition of sine waves at varying frequencies. And if some characteristic of a musical instrument is causing the air to oscillate higher than 22 kHz, you can't hear it. So we cut those frequencies off, sample the rest at 44.1 kHz, and nobody can ever tell the difference.

The harmonic structure contains harmonics, in harmonics, overtones and at different amplitudes, and they differ from one instrument to another. Since all audio equipment acts like a filter, the harmonic structure will alter and it's trying to preserve that harmonic structure along with the rise time, sustain and decay time of each note being played.

So, according to you, how could these harmonics be measured? If only there were some kind of oscillating mathematical function that could precisely describe the movement of sound at differing frequencies through the air... ;)

So please don't go by THEORY, because the reality is what and how an actual physical device performs.

Understanding how an actual physical device performs requires scientific measurement. If there's a weakness in the Nyquist-Shannon sampling theorem, or if you believe that your ears can detect frequencies higher than 22 kHz, or if you believe that a harmonic at 64 kHz actually changes the characteristics of a 22 kHz sine wave in a way that can't be captured by a system that samples at 44.1 kHz... there's nothing preventing you from expressing any of those points in objective, scientific terms.

Vague statements about capturing the full range of emotion or whatever are completely subjective. The frequency range your ears are physically capable of detecting, and our ability to reproduce and measure that frequency range, is not subjective.

I've read that 16/44.1 was good enough that digital audio experts thoughts and then they changed their stance on it

Could you please share a link? I'd like to see who these experts are and to try to understand the evidence they presented.

When we get to recordings, they vary too much. I've heard great 16 Bit recordings, but I've awful ones as well, same goes for 24 Bit.

I agree completely on this point. There's a never-ending supply of terrible-sounding music reproduced at 16/44.1. I'd bet that today, the average 24/96 commercial release "sounds better" than the average 16/44.1 commercial release. But that has nothing to do with the science of the bit depth or frequency range and everything to do with marketing to people who fancy themselves audiophiles.

I feel like this has been said too many times in this thread already, but bandlimiting and downsampling any 24/96 recording to 16/44.1 will result in zero discernible change to any audiophile sitting in a soundproofed room. Anyone can test this for themselves using a wide range of ABX-testing software, such as Lacinato ABX or foobar2000 with the ABX plugin.
 
So we've been having a discussion lately about whether a sine wave at just under half the sampling frequency can be captured perfectly. The answer is yes. This means that all frequencies below that frequency are also perfectly captured. Any frequencies higher than that frequency is not captured, because they were bandlimited out before sampling, and that's fine, because you can't hear them.

The point is, music absolutely is a set of sine waves superimposed upon one another. Rise time/sustain/decay/harmonic structure are all perfectly captured by the right superposition of sine waves at varying frequencies. And if some characteristic of a musical instrument is causing the air to oscillate higher than 22 kHz, you can't hear it. So we cut those frequencies off, sample the rest at 44.1 kHz, and nobody can ever tell the difference.


So, according to you, how could these harmonics be measured? If only there were some kind of oscillating mathematical function that could precisely describe the movement of sound at differing frequencies through the air... ;)



Understanding how an actual physical device performs requires scientific measurement. If there's a weakness in the Nyquist-Shannon sampling theorem, or if you believe that your ears can detect frequencies higher than 22 kHz, or if you believe that a harmonic at 64 kHz actually changes the characteristics of a 22 kHz sine wave in a way that can't be captured by a system that samples at 44.1 kHz... there's nothing preventing you from expressing any of those points in objective, scientific terms.

Vague statements about capturing the full range of emotion or whatever are completely subjective. The frequency range your ears are physically capable of detecting, and our ability to reproduce and measure that frequency range, is not subjective.



Could you please share a link? I'd like to see who these experts are and to try to understand the evidence they presented.



I agree completely on this point. There's a never-ending supply of terrible-sounding music reproduced at 16/44.1. I'd bet that today, the average 24/96 commercial release "sounds better" than the average 16/44.1 commercial release. But that has nothing to do with the science of the bit depth or frequency range and everything to do with marketing to people who fancy themselves audiophiles.

I feel like this has been said too many times in this thread already, but bandlimiting and downsampling any 24/96 recording to 16/44.1 will result in zero discernible change to any audiophile sitting in a soundproofed room. Anyone can test this for themselves using a wide range of ABX-testing software, such as Lacinato ABX or foobar2000 with the ABX plugin.

The reason to down sample is easy. If a recording was originally recorded in let's say 24/192, you'll at least capture it at a level that SHOULD give you as good of audio quality as you can get and you would down sample it to provide the content in a smaller file size for those will limited storage or in a CD which most people can play, but you could still offer the original 24/192 as a digital file for those that have more storage available and a better system to hear the difference between 24/192 (lossless) and 16/44.1 (lossy) which is MP3/AAC which is what people are buying for mobile devices.

Some record labels are recording at DSD 2x and then if you don't have DSD playback, you can convert to PCM at what ever level you want. But DSD 2x are HUGE file sizes. Can they hear the difference? Well, according to the people involved in listening tests say they can. Is it is huge difference? For the average person, they might not hear that difference, but to those that have better trained hearing with better equipment might be able to hear that subtle difference.

When comparing 16 bit vs 24 Bit, they aren't listening to the upper frequencies, they are listening for better accuracy and spaciousness in the the frequencies they can hear. Have you heard speakers that can go beyond 20kHz and compared them to speakers that can't?

Here's a little test you can do fairly easily at any high end audio dealership that carries Sennheiser HD800 headphones. Go listen to a pair of garden variety headphones that have a frequency response of 20Hz to 20kHz (which is the normal hearing range accepted by the majority of audiologists) and then listen to a pair of headphones that have a frequency response of 14 to 44.1kHz (Sennheiser HD800) and then tell me you can't hear a difference. Odds are you will hear a difference and it's not a placebo effect.

As far as Foobar is concerned, I don't use it because I don't use Windows. I wish they would get off their ass and come out with a OS X version.
 
I should probably clarify something... when I said the average 24/96 release "sounds better" than the average 16/44.1 release, that's not solely because of marketing but because, on balance, the average 24/96 release will likely be produced by better recording/mixing/mastering techniques. These techniques can be, and often are, applied to producing a 16/44.1 release that'll sound identical. Where the marketing comes in is in convincing people to believe blanket statements like "24/96 sounds better [than 16/44.1]."
 
Great post from garrettj, to add to what he said...

What I see HDTracks has in SOME cases is the record label has the original analog tapes either reconverted to 24/96 or higher and they simply remove a lot of the compression they used in the Redbook CD version and they sometimes do other things in addition. What exactly, I don't know, but I've compared the 16 Bit rebook to the 24/96+ and there is a HUGE difference.

Like you said, they process it differently knowing it is going to this different release format. So any difference you are hearing is because of the content actually being different, not the extra resolution. As has been said over and over, if you want to hear the difference that comes from the extra data, downsample the higher resolution file and ABX.

Recordings are remastered all the time, they do things they think sound better (listeners may or may not agree) and release again, whether it's a higher resolution or not.

Some of the newer digital recordings were originally done at 24/96 or higher and it makes so much sense to offer those recordings in both 16 Bit MP3/ACC for the mobile crowd AND 24 Bit versions that are left alone for the home audio enthusiasts crowd.

For the sake of marketing and making more money, sure. For the sake of people actually hearing a difference, not really.

The problem in the audio world is for there to be set standards for being able to do more quality of sound measurements that are valid and repeatable and used throughout the audio industry.

I don't see the problem. There are set standards for being able to do more quality of sound measurements that are valid and repeatable and used throughout the audio industry.

A lot of measurements these companies are making don't exactly analyze quality of sound.

You're trying to insert magic pixie dust into the process where it doesn't exist.

There are some DACs that can play 16 Bit Redbook VERY closely to a high end turntable where you get the same emotional connection with the music.

No question some DAC sound better than others, there is great 16 bit and probably not so great 24 bit. But "close to a high end turntable" is not really a good place to set the bar. People love how records sound, if you are one of those people you should just listen to records. Same with trying to bring in "emotional connection". If you can't hear the difference in an ABX test, you're not going to have a different emotional reaction.


CDs for the longest time were very flat sounding, etc. and that may be partly due to the way the CD was mastered or the DAC or both. But they are getting much better with these DAC designs over what they had 30 years ago.

Both, probably more of the latter. Technology improves over time and it definitely has in the case of digital recording, even without increasing the data rate.

Sine waves measurements are good for checking phase measurements, but to determine sound quality? NOPE. What they SHOULD be measuring are musical notes played by various musical instruments over the complete range of the instrument to see how well the audio component you are working with will accurate preserve that note.

In the case of testing equipment they absolutely do use musical material. We talk about sine waves because that's the easiest way to explain the process and the math.

The harmonic structure contains harmonics, in harmonics, overtones and at different amplitudes, and they differ from one instrument to another.

Yep, harmonics are just additional frequencies along with the fundamental. If they're within the range recorded by a piece of gear, they are recorded, otherwise they're filtered out. In the case of digital it was designed to record audible frequencies and leave out inaudible ones.

The problem with looking at theories, is they are just that, a theory. Heck, most people in the audio world use the THEORY that the speed of sound is 1,122 feet per second, but in reality, it isn't. It changes due to the environment. Air temperature can effect the speed of sound. So please don't go by THEORY, because the reality is what and how an actual physical device performs.

The speed of sound isn't a theory, it's a universally agreed upon fact. And it does take into account air temperature, nobody would ever claim that it's always exactly 1122 under any conditions.

And to be perfectly clear in the case of Nyquist "theory/theorem" is a misnomer. At this point there's nothing theoretical about it, probably should be called the Nyquist Law to avoid confusion, not sure why it isn't.

I've read that 16/44.1 was good enough that digital audio experts thoughts and then they changed their stance on it and then later came out with 24/96 was good enough, now there are others that are arguing that DSD 2x is even better.

Depends who your "experts" are. Obviously if a company is selling gear they're going to insist that what they are selling is better.

I think in the context of the average consumer, 16 Bit is probably good enough since the average consumer doesn't have high quality playback equipment and "trained" ears that can discern subtle differences in audio quality, not to mention good soundproofed and treated listening rooms, but some people do.

You're saying your home listening environment has a noise floor of 0dB SPL? Sorry, that's a crock.

You haven't even told us that you down sampled a 24/96 recording and consistently passed an ABX comparison.
 
you could still offer the original 24/192 as a digital file for those that have more storage available and a better system to hear the difference between 24/192 (lossless) and 16/44.1 (lossy) which is MP3/AAC which is what people are buying for mobile devices.

Whoa, lossless vs. lossy is a completely different discussion. We've been discussing the difference between 24/192 and 16/44.1 only. That is, both are lossless for the purposes of this discussion.

Some record labels are recording at DSD 2x and then if you don't have DSD playback, you can convert to PCM at what ever level you want. But DSD 2x are HUGE file sizes. Can they hear the difference? Well, according to the people involved in listening tests say they can.

What listening tests? My understanding is that DSD 2x samples at hundreds of kHz. I would be extremely interested to see the results of a double-blind study that demonstrate some people have superhuman hearing reaching into the hundreds or thousands of kHz. That would be mind-blowing if true. My guess is that the listening tests you're talking about were either not double-blind, or were somehow produced by people with a financial interest in selling DSD.

When comparing 16 bit vs 24 Bit, they aren't listening to the upper frequencies, they are listening for better accuracy and spaciousness in the the frequencies they can hear. Have you heard speakers that can go beyond 20kHz and compared them to speakers that can't?

Yeah, but we're not talking about comparing equipment, we're talking about comparing recording techniques. If you can hear a difference between those two speakers, it's down below 20 kHz.

Here's a little test you can do fairly easily at any high end audio dealership that carries Sennheiser HD800 headphones. Go listen to a pair of garden variety headphones that have a frequency response of 20Hz to 20kHz (which is the normal hearing range accepted by the majority of audiologists) and then listen to a pair of headphones that have a frequency response of 14 to 44.1kHz (Sennheiser HD800) and then tell me you can't hear a difference. Odds are you will hear a difference and it's not a placebo effect

Again... you're right! But that's not because you can hear frequencies above 20 kHz, it's because different headphones color the sound differently in the audible range, i.e. it does a better job at reproducing frequencies between 20 Hz and 20 kHz than the garden variety headphones. Just because I agree that the HD800 is a wonderful pair of headphones and sounds better, doesn't mean that I, or you, or anyone, can hear above 20 kHz.

The quality of sound equipment is all over the place. But converting analog audio to digital and vice-versa can be measured very precisely. We're talking about the latter.

I'll reiterate my point again, rephrased slightly: Bandlimit and downsample any 24/96 recording to 16/44.1. Test your trained, golden ears with ABX software that will compare the 24/96 and the 16/44.1 in a soundproofed room with your very best pair of speakers and your very best pair of headphones. Every objective scientific test we've done so far, every theorem, says you won't be able to tell the difference. Therefore, 16/44.1 is sufficient for listening and anything beyond that is unnecessary overkill. What part of this don't you agree with and why?

----------

Sorry for the pile-on. :)
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.