So, the noise of that signal is -5dB SPL. 0dB SPL was set to be the lowest possible perceivable level of a single sound in an aechoic chamber. And that's not even considering other sounds in the recording, or ambient noise levels in a typical living room, etc.
In your example, moving to 24 bit would been a long way from having any effect (other than a 50% increase in file size). And if you use, say, an 8 bit signal as an example, then things are even less noisy. Note that the noise is the only consideration here: any fidelity loss is represented in that figure.
The audio engineers of yore who (among other things) decided that 16 bits was more than enough for final mixdown were much more competent than they get credit for (many were downright amazing at what they did, in fact). They thought of stuff like this.
If the results are no better than chance, then 24-bit doesn't matter, regardless of how sound the underlying argument is.
EDIT: The experiment would also be extremely difficult to design. For example, you'd need to run this test with music, not simple sounds. So the question is, which music? I think whatever is most popular at the time would be a good candidate, because if people are listening to music they hate, they won't care about the fine details of the audio. But that introduces an element of uncertainty and noise into the results which is hard to control for.
Some people might deliver accurate results with https://www.youtube.com/watch?v=2zNSgSzhBfM but not with https://www.youtube.com/watch?v=4Tr0otuiQuU whereas for others it's the opposite.
Or, it could be the exact opposite: Maybe you can only detect whether a sound is 24-bit when it's a simple tone, and not music.
Age is also a factor. My hearing is worse than a decade ago.
The headphones used by the test are another factor. If you feed 24-bit input to headphones, there's no guarantee that the speakers are performing with 24-bit resolution. In fact, this may be the source of most of the confusion in the debate. I'm not sure how you'd even check whether speakers are physically moving back and forth "at 24-bit resolution" rather than a 16-bit resolution.
A quick summary would be that most "popular" music has been mastered with the following goal: the song should be recognizable and listenable on a FM-radio with only a limited bandwidth midrange-speaker. One of the many things they do to achieve this is by eliminated almost all dynamic range through a process called "compression" (dynamic compression, not digital-compression).
They also limit the spectral range to not have "unheard" sounds cause distortion when played through limited bandwidth amplifiers and speakers.
This means that the kinds of musical pieces which could benefit from the increased dynamic range of 24-bit would be thoroughly excluded from the test.
And then you'd probably get the "expected" result, but only because you now test whether music mastered specifically not to have dynamic range benefits from having increased dynamic range. For which the answer is given.
Note: I'm not claiming 24-bit end-user audio has merits, of which I have little opinion. I'm just pointing out the flaw in the proposed experiment.
If you feed 24-bit input to headphones, there's no guarantee that the speakers are performing with 24-bit resolution.
Not sure if you're just imprecise in your language here or if you're genuinely confusing things. Speaker-elements, as found in both speakers and headphones are analogue. They operate according to the law of physics, and respond to changes in magnetic fields, for which there is practically no lower limit.
They have no digital resolution. A quick example: Take your 16-bit music, halve the volume and voila! You are now operating at "17-bit resolution". Halve it again. 18-bit resolution. Etc.
There's probably some minimum levels of accuracy, yes, but it just doesn't make sense to measure it in bits.
If you're aware of this and were just trying to adjust the language to the problem at hand, I'm sorry for being patronizing, but I just wanted to make sure we keep things factual here.
I disagree. I think all the factors you are concerned about can be eliminated with a large enough sample size, like in the thousands (or maybe 10s of thousands).
You allow each person to select the genre of music they like, and you play a few clips from a few songs of each bitrate. Then they guess which is 24-bit and which is 16-bit.
I'm not paying to set it up. But it could all be done online without too much grief. It would be good to track the other statistics (age, headphone brand, etc.) as well, and see if something falls out of that.
More realistically, the participant might choose music for which no 24-bit recording exists.
It's very important to control for every variable. It's actually not possible to gather info about what headphones the listener is wearing. Even if it was, it wouldn't be possible to know whether they're doing the experiment in a quiet room, or whether there's a traffic jam just outside their apartment window, or whether their dog is barking during the test. Stuff like that.
Crowdsourcing this is an incredibly cool idea, but it'd just be so easy to believe you've performed a reliable test even though some variable undermined it.
I forgot another variable: Whether the music was recorded at 16-bit resolution. Most musicians use 24-bit, but it's easy to imagine that some of their samples might've been quantized to 16-bit without them realizing it.
It's not, actually. Say you have 10,000 listeners and you randomly assign each one to 16-bit vs 24-bit listening. You have enough listeners that any differences between the groups are due to chance and will very close to even out. Now, if you find people are unable to distinguish between 16-bit and 24-bit you might want to try the test again with more control over the environment, but if you find a substantial difference in a large blind randomized test that's a real finding.
Well, obviously we'd need to have a limited set of music selections for which we have 24-bit recordings.
As you suggest, I expect the biggest impact on playback fidelity is going to be other factors like the noise in the system (likely a PC) and such.
But the flip side of it is that's also a good real world test. If the only time you can tell a difference is to be in an acoustically dead room with top end equipment, then the higher sample rate really isn't worth it.
Hey, that's a great point! Hadn't considered that.
Proving "most people can't tell the difference between 24-bit and 16-bit in real-world settings" is less compelling than proving "no one can ever tell the difference," but it's still very relevant.
So you'd have to decide in advance what difference is meaningful and choose your sample size to ensure you can detect it.
If I pay for music, I want to truly own it, including the possibility of someday making a mashup, a music video, a hip-hop beat, et cetera. A 24-bit source gives casual creatives the same quality material as the original masters, for a relatively paltry 1.5x increase in file size.
Chris Randall of Sister Machine Gun (at least used to?) use a low-frequency generator at live shows to produce a sound that the audience could feel but not hear in order to make the music more intense. I suspect that you'd gain some of that effect with a larger bit size.
...or I could be completely wrong. Whatever.
What you mention, though, points directly to what /will/ improve the quality of sound reproduction: speakers. It gets harder and harder to move that much air with precision as you get lower and lower in frequency. It's a definite technical limitation, but it's to do with very high-power amps and giant speakers, not the recording format.
We have (to the degree that humans can prove that they can perceive), perfect reproduction from digital recordings, perfect amplifiers for reasonable prices (at lower-than-concert-power-levels at least), but we haven't yet developed good enough speakers to cover the whole perceptible range of frequencies to anywhere near the same degree.
Audiophiles love to try and improve the whole chain, but really the only place it matters is at the very end.
What you feel is parts of your body resonating (because the low frequency sound is exciting modes of your body). This is unlikely to happen at high frequencies, partially because it would necessarily be much smaller parts of your body (see [0] for a diagram of typical body resonance frequencies) which we probably don't feel and because attenuation of sound greatly increases at higher frequencies (for example, see [0] for air), making it likely impractical to excite any such modes. My guess is that you might cause some tissue damage if you had significant ultrasonic excitation in your body (see [2] for something that may or may not be true...).
[0] http://physics.stackexchange.com/questions/37543/does-the-hu... [1] http://www.kayelaby.npl.co.uk/general_physics/2_4/2_4_1.html [2] http://www.tovatech.com/blog/4376/ultrasonic-cleaner/ultraso...
In fact, you should be very careful about to think about quantization in digital sound reproduction, because it can easily lead you astray. Think of bit depth as a measure of dynamic range. Do look at that digital music primer for geeks posted elsewhere in this thread, it makes for awesome reading.
24 bits is great for recording, but the end distribution medium only needs to be 16 bits, after you normalize the thing you tracked at 24 bits.
You can get libsndfile and FFTW and do the tests yourself.
Edit: While it would be possible to represent each of those individual sinusoids in 6 and 10 bits individually, the signal you describe has the high frequency signal "riding on " the low frequency signal. you need 16 bits to represent the amplitude of that signal at e.g. t = pi/4.
Sound attenuates proportional to f^2 in air. So, a 10-fold increase in frequency causes a 100-fold increase in attenuation. So, a 100KHz signal has 10,000 times the attenuation for a 1KHz signal over the same range.
In addition, attenuation due to water vapor is particularly bad above about 40KHz. Much of that study cannot be replicated, and people have tried: http://en.wikipedia.org/wiki/Hypersonic_effect
In addition, the effect went away when using headphones.
However, I can certainly believe that if you can pump enough energy to rattle things at ultrasonic frequencies you are going to get a result. Especially since ultrasonic frequencies rattle in water particularly effectively.
As an example, if I pump enough energy into an ultraviolet or infrared signal at your eye I will eventually get a detection result in your brain. However, pain and a burned retina are not what we think about when we consider a brain response.
I find the intermodulation argument a convincing one - it's hard to not have it affect the test, and if they haven't taken specific steps to avoid it then it would be easy to join a large amount of tests that have fallen prey to it.
I note, however, that Wikipedia doesn't mention any specifically brain-scan-based studies that counter it. If you know of any more I'd be very interested in hearing about it.
On one hand, higher frequencies above 20KHz can't be heard at all, so there's no point having them! You can't hear them!
Then on the other hand, higher frequencies above 20kHz affect the audible region of the sound (intermodulation distortion), so you can hear them, so make sure they aren't there!
What if the presence of the higher frequencies in a spectrum that shares a harmonic relation to the audible region causes intermodulation distortion that is pleasing and musical to the ear. What if the complete absence of this high frequency information, or alternatively a non-harmonic higher frequency signal (say some kind of switching or power supply noise) causes the audible region to be perceived in a less pleasant manner?
Certainly some people find certain kinds of distortion pleasant, but the people arguing for 192kHz claim increased fidelity, not pleasing distortion - when it is just the opposite for any stereo that introduces these artefacts.
In typical PCM recordings, like CDs, mid-range frequencies (e.g. 1kHz to 4kHz) are recorded with lower amplitudes because our ears are more sensitive to them.
Sampling theory is correct and 16-bits can reproduce any waveform with ~100dB of range, however, in a complex waveform consisting of low, mid and high frequencies, the mid- and hi-range frequencies quite simply get shortchanged.
Imagine a recording of a bass sinusoid and a mid-range sinusoid of equal volume. It might use e.g. 10 bits to store the bass and only 6 to store the high frequencies. (2^10sin(200wt)+2^6sin(4000wt)). That means the resolution of the high frequencies is less than the lower frequencies. When the volume of those frequencies changes dynamically, the high frequencies' amplitudes are more quantized. That is quite simply why 16-bits are not enough.
This is similar to the problem with storing waveforms unprecompensated on vinyl. The precompensation makes up for the non-uniformity of the medium. It could be done with 16-bit digital as well. Or alternatively, larger sample sizes like 24 can be used.
I haven't A/B tested this. The A/B test in the article compares CD with SACD. SACD isn't PCM, so its artifacts are going to be totally different from 24-bit PCM.