13 bits would already be pretty much best case vinyl during a quiet passage, so no, you wouldn't need more.
Sure, with 16 bits you're going to be on the safe side.
Even as you throw more bits at the problem you approach the original signal+noise but never quite captured it exactly. So whether 16 bits is sufficient or not is very much subjective.
De-noising filters generally operate based on assumptions of known statistics of the noise. Often they don't perfectly remove just the noise. There is likelihood that some minute amount of what we consider a sound signal is removed in the process as well. The question of this balance is, again, subjective to the listener.
That's fine in theory, but in practice, digital noise filters are extremely useful. Just another tool that can be used well or poorly depending on the person using it.
Some noise filters use aspects that are beyond the range of the music itself to determine what is noise and what is signal... for instance impulse info or super audible frequency information.
Gregorio - thank you for this incredibly compelling and helpful discussion. I'm not a sound engineer nor do I have any technical expertise - with that, two questions:
1. Why are companies migrating to high resolution audio (see Sony, for example) when the real problem in consumer audio reproduction is the mass sale/distribution of compressed audio. I own a bunch of CDs - but it is a pain to have to go buy CDs and then rip those into lossless audio files. The alternative is to go with HD Tracks - but no one is selling CD quality lossless for consumer sales outside of physical CDs in any volume. Seems crazy to me that the non-compression market distributors aren't saying - hey, you really only need a lossless version of the CD. Just very frustrating. Also, I take your point that higher resolution doesn't actually "resolve" anything that a listener can actually hear - and that the quality of the sound recording is the real determinant factor for playback. With that - why not market the quality of the sound recording in a lossless format at 16 bit - and not fool consumers into buying "high resolution" snake oil (which also consumes a lot more memory and processing power).
2. Re 16 bit v 24 bit, is there something that the music industry can do to improve audio reproduction now that we all own 3+ terabit tank hard-drives and enough processing power to resolve higher resolution files. In other words, what IS a good idea in terms of the development of audio reproduction (now that I know that added bit depth is useless).
Thanks - this is a great discussion.
1) People have been told that higher resolution sounds better due to the format, so they ask for that format. Companies are just serving what people are asking for.
No company wants to be stuck at some "good enough" point. For example there is not much Ultra HD (4k) material out yet and companies are already working on 8k. They always try to sell you more expensive stuff which is supposedly better.
2) "Improving" reproduction will not change the limits of human hearing.
Sampling rate: An infamous analogy is trying to blind yourself with an IR remote. Do you need monitors that reproduce colors outside the visible band?
Bit depth: Do you need someone whispering a few meters away followed by a jackhammer next to your ear?
The most important improvements are still happening in headphones themselves. With research being done by Harman, NRC .. there's finally an attempt of making headphones that have a more accurate frequency response, lower distortion.
Oh and of course audio processing technologies ranging from simple crossfeeds to more complex room simulation with head tracking and your personal HRTFs.
Thanks Xnor - so what about up sampling, over sampling and the other "fancy" DAC processing tricks that companies advertise - is that all snake oil as well. Given Moore's Law - did audiophile processing power hit the wall - with NO appreciable improvement with continuing advances in computer processing power?
Some techniques are needed to make high quality reconstruction possible and less expensive. For example oversampling in a ΣΔ converter is used so that you don't need expensive high-precision analog anti-aliasing filters.
Sample rate conversion, no matter if done in software or hardware, doesn't hurt if done right. It can only add redundant information but there may be some gains in performance in certain configurations. Whether those differences are even close to being audible is another story.
I don't think companies will stop at 96 kHz and 24 bit. Not long ago HDMI 2.0 was announced with:
That sample rate may "only" be 32 channels with 48 kHz each (32*48=1536) but even so .. 32 is quite a big number of channels. Maybe it also means that audiophiles can resample their good old 2.0 stereo files to 768 kHz for that extra placebo .. ehh audio fidelity?
A sampling rate of 48khz allows you to accurately capture signals up to about 24Khz (fs/2)
Right fs/2. After that - it's cats and mice.
The behavioral audiograms of two cats were determined in order to establish the upper and lower hearing limits for the cat. The hearing range of the cat for sounds of 70 dB SPL extends from 48 Hz to 85 kHz, giving it one of the broadest hearing ranges among mammals. Analysis suggests that cats evolved extended high-frequency hearing without sacrifice of low-frequency hearing.
Well, turning Fs/2 around: you need more than 2 samples per cycle to reconstruct the signal.
44.1/22.05 = 2 samples per cycle for a 22.05 kHz tone sampled at 44.1 kHz
In practice, most DACs roll-off between about 20 kHz to 21 kHz with a 44.1 kHz sampling rate, so around 2.2 to 2.1 samples per cycle.
Hearing high frequencies really is very problematic for humans. First there are less and less hair cells that are tuned for high frequencies. As a result, the hearing limit is rising to a point where you need to play the tone at extremely dangerous levels to just barely hear it:
That's why 20 kHz is the generally accepted upper limit of hearing.
Secondly, that limit degrades with age and there are many people who cannot even hear a clean 18 kHz tone anymore. Elders usually have troubles hearing stuff above ~16 kHz.
Thirdly, if you look at the spectrum of real music you will see a downward slope. The higher the frequency, the less energy. I don't know of any natural musical instrument that produces mainly treble with lots of energy.
Even if you could hear a clean 20 kHz tone at 100 dB SPL you would have to turn up the music so loud, that the spectrum of the music would be pushed to far above 150 dB SPL at 100 Hz. (Having heard a jet engine at ~150 m away I cannot recommend anyone hearing it at a distance of only 30 m, which according to Wikipedia would be roughly 150 dB SPL.)
Interesting to see this thread is still alive and well 4 years after I started it! Some of the questions and responses are also interesting, as is the fact that some posters can't seem to get their head's around the concepts in my OP. Generally this appears to stem from the fact that they have an entrenched concept which they are unwilling or unable to to question, namely that more resolution = more detail. One can't entirely blame them for this entrenched position as it has been created and reinforced by the marketing of all those companies who sell higher bit depth files or equipment. The very term "Hi Rez" itself is a purely marketing term rather than an accurate description because if we are talking about resolution in terms of detail (of the sound which comes out of your DAC), 24bit has no more resolution than 16bit or indeed than 1bit! I saw one truly bizarre set of posts in this thread from someone who thought it was simple common sense both in theory and in practice that 24bit had more detail and sounds better than 16bit and that anyone suggesting otherwise must be essentially insane. This poster then held up SACD as the highest audio quality but of course SACD uses just 1bit and therefore, according to his common sense argument, SACD should sound many times inferior to the humble 16bit CD.
Some of the points raised in this thread (and others on Head-Fi) could do in my opinion with a little more explanation, so here goes:
To help those still struggling with the concept that resolution does not equal detail, try turning your logic around and think of it this way: Exchange the word resolution with the words "less error". More bits = more accuracy and of course more accuracy = less error, hence higher resolution (more bits) therefore means less error. The next leap in understanding this question of resolution is understanding the fact that error in digital audio manifests as noise when it is converted back into an analogue signal. More resolution therefore results in less noise compared to lower resolution, the detail (fidelity) is always there and always the same (at any bit depth) but the more bits you use the less noise will accompany that detail. By the time we get to 16bit, the noise (due to error) is already many times below the noise which is already present in the recording due to other unrelated music production factors. Using noise shaped dither further reduces the audibility of that noise to well below the noise produced by even the very finest DAC, amp and headphone or speaker system. In other words you cannot hear this dither noise! So, if you already cannot hear the errors (noise) introduced by 16bit what is the point of increasing the resolution (to 24bit) and reducing the noise even further. In other words, what do you possibly imagine could be gained by taking inaudible noise and making it quieter? If this is true, why does 24bit even exist, what point is there? This brings me on to recording:
When recording in 16bit, we want to create the cleanest recording possible and that means making sure that the noise caused by our 16bit errors are below the noise of our microphones, mic pre-amps and the noise floor of our recording studio. This means setting our microphone pre-amps so that the loudest parts of our recording are somewhere near the maximum (0dBFS) level of our system. The problem with this is that we don't know what the loudest part of our recording is actually going to be until after we've recorded it. Sure, we can (and do) do a sound check but even the very finest artists cannot reproduce a performance exactly, so a sound check is only a very rough guide. Performers can create peaks 12dB higher or even more during recording compared to the sound check. So our problem is how to record loud enough to ensure our digital noise (errors) does not encroach on our recording but not too loud so that the loudest parts of the performance exceeds 0dBFS. In some recording situations this can be a quite small window of opportunity to hit. Recording in 24bit though ensures that our digital error noise is so far below the noise of our recording equipment and environment that we don't need to think about it. With 24bit we can afford to set our mic pre-amps so that the loudest part of our recorded signal is nowhere near 0dB, even -20dB is fine in 24bit, so it doesn't really matter how loud the performer performs, we're not going to ruin that brilliant and unrepeatable performance because we've hit the limit (0dBFS). We don't use 24bit for recording because it provides more detail, better quality or higher fidelity, a 24bit recording will have the same fidelity as a 16bit recording made within that window of opportunity, it just makes that window of opportunity far easier to hit! In other words, the advantages of 24bit for recording are all about workflow and nothing to do with quality or fidelity. Of course, when we mix/master and create the distribution mix, we already have the recorded material, already know where it is going to peak and already know what that peak value is, so we don't need that spare dynamic range, can go right up to -0.1dBFS without ruining our mix and are going to be so far away from the 16bit digital noise (errors) that it's guaranteed to be many times below the threshold of audibility. This brings up the question of mixing:
When we mix music (or a film or TV program) we could be dealing with anything from a few channels of sound up to over 1,000 channels. On each of these channels we may have anywhere from zero processors, up to as many as 8-10, (EQ, compression, reverb and a wealth of others). Each of these processors process the audio, which in digital audio means runs a series of mathematical algorithms (calculation processes) and each of these algorithms is likely to introduce an error in the least significant bit (LSB). The LSB can only hold a 1 or a 0 but our algorithm (calculation) could easily result in say 0.6, so we set the LSB to "1" but we've introduced an error (noise). You're not going to hear this error in 16bit and obviously you're not going to hear it in 24bit but what happens when we've got say 80 channels of sound, each with say 2 processors and therefore a total of say 200 algorithms, the results of which are all being summed together? You will certainly hear the result of these combined errors at 16bit and even probably at 24bit. This problem is overcome by operating the processors and mixing at far greater bit depths, the errors in the LSB are now are so minuscule that even summing thousands of them together does not introduce anything remotely audible. It's common for many years for processors to be operating at 48bit, with mixing at 56bit and becoming even more common today to run everything (processors and mixing) at 64bit float. Once the mix has been made it can be brought back down to 16bit which is more than enough for every playback situation. I hear some audiophiles screaming: "I want the recording at it's original resolution". Well, you can have it! We record at 24bit (for reasons explained in the paragraph above) and we mix at various different bit depths, which generally we cannot print. The 56bit and 64bit float mixing environments common today are just for internal processing, we cannot actually write (record) files at these bit depths because they are useless, nothing can play them back and even if something could play them back you wouldn't be able to hear what was in the least significant 50bits or so (of a 64bit file) anyway. So when you see companies advertising "24/96 or 24/192 hi rez as it was created by the studio", that's a double lie! 24/96 (or 24/192) is not hi rez and is not as it was created by the studio. Which brings us on to mastering:
Mastering is the process of taking a mix created in a recording/production studio and processing it so that it sounds good on the target audience's playback equipment, rather than sounding good only in the recording/production studio where it was made. This raises a number of important questions such as;what genre is the music, who is the target audience, what is their playback equipment and related to this, what format are we distributing? This obviously requires making assumptions and generalisations but if for example we are distributing jazz or classical on SACD we are most probably looking at an older target audience, who most probably take their music listening seriously (otherwise they wouldn't have bought an SACD player) and who are therefore the most likely demographic to have high or very high quality playback system/environment. So, we are far more likely to record, mix and master to a high standard and with a wide dynamic range. Instead of burning SACDs from this high quality master we could just convert it to 16/44.1, put it on a CD and this CD will sound absolutely identical to the SACD. For the record company there are two problems with this though: Firstly, this wide dynamic range, high quality recording might not be suitable for many playback systems/environments and secondly, how do you justify a significantly higher price for a "hi-rez" SACD which contains a recording indistinguishable from the much cheaper CD version? Of course there's an easy solution, you make a different 16/44.1 version which is distinguishable from the "hi rez" version! This brings us on to:
For practical purposes, dynamic range can be defined as difference between the highest energy in a signal (recording) and the lowest. As explained earlier in this thread, 16bit is capable of containing more dynamic range than you can safely listen to and even the best and most dynamic of SACDs have a dynamic range of no more than about 60dB and most recordings have a dynamic range of less than 40dB. To put this into perspective, 16bit is capable of roughly 1,000 times more dynamic range than even the most dynamic of SACDs. I can't understand this attitude from some audiophiles of wanting even more dynamic range than 16bit provide, enough dynamic range in fact to kill them if it were actually possible to use 24bits of dynamic range. Not only is this desire for 24bits (144dB) of dynamic range literally suicidal, it makes no sense in many cases to increase dynamic range of even some of the crushed music which many audiophile abhor. If you are listening to a recording in a quiet environment with very low environmental noise then yes, a decent dynamic range is a good thing and will allow the recording to sound breath and sound more alive but listen to that same recording say in a car while you're driving along the interstate and the high ambient noise will mean that you won't be able to hear half of recording without turning the volume up and nearly deafening yourself when a loud piece of the music comes along. I saw a thread on Head-Fi earlier where someone seemed desperate to get 24/192 playback from a galaxy smartphone. A quick look at the specs shows that at the headphone outputs this phone has a dynamic range of 92dB, well above the dynamic range of any commercial recording but well below the dynamic range possible with 16bit, why then is he wanting 24/192 playback? Even if his phone can handle 24bit format files it can't actually play more than about 15bits of them and for all intents and purposes ignores the other 9bits.
Coming back to mastering, why is it that many audiophiles seem to spend so much time, effort and money obsessing about aspects of their equipment which cannot possibly be heard but relatively little time and effort understanding and appreciating good mastering? I've heard the response to this question, which is "well the mastering is fixed on the recording and there's nothing we can do about it but we can do something about the equipment we use". Sorry, but this is horsesh*t, for two reasons: 1. If, as some fanatical audiophiles contend, power cables, digital inter-connects and ridiculously expensive speaker cables do actually make an audible difference, then every choice they make changes the mastering! For example, if an extremely expensive cable makes their system sound brighter, the mastering engineer almost certainly used relatively cheap copper cables and made the master exactly as bright as he/she intended, the expensive cable is changing the master! and 2. If these audiophiles spent more time on something which is easily audible instead of on what is patently not audible, they would soon learn what good mastering actually is and if they only purchased well mastered music, the record companies would soon take note and make sure their mixes/masters were of a higher standard to cater for this market! Let's not forget that mastering engineers are human beings, you can have an album mastered for $40 a song through the internet from a young newbie with little knowledge, experience or facilities or you can have an album mastered in one of the top mastering facilities by one of the world's great mastering engineers for $20k. Which album would you think is likely to sound better and which would you choose to buy if the price were the same or nearly the same?
Can you hear a difference between 16bit recordings and 24bit ones? Yes, most definitely you can, I certainly have! There is however only 3 possible explanations 1. You are imagining the difference or 2. Some serious mistake has been made during mastering or the most likely 3. You are actually listening to different recordings/masters. 16bit, 24bit, 1bit (SACD) are just containers, what you put in those containers defines the quality of what you are hearing, not any inherent quality of the container itself. It's like trying a drink from a square bottle, liking it more than a previous drink you had from a round bottle and therefore deciding that you'll only drink from square bottles in future. In reality of course it's the drink which is in the bottle which makes you like the taste or not, not the shape of the bottle it's in.
One final thought: We've already covered that 16/44.1 is more than will ever be needed but going the other way, is there any potential problems with listening to the same recording at 24/96 or 24/192 or even 32/384? As far as bit depth is concerned, the answer is "no" there's nothing to be gained from higher bit depths but there's also nothing to loose, except storage space. This isn't necessarily true of the higher sample rates however! Most amps and speakers are not designed to reproduce any frequencies above about 20kHz, feeding them any significant amounts of frequency outside this range can cause them to create inter-modulation distortion (IMD), which is spurious tones or sounds within the range of human hearing. It's possible that some audiophiles might actually like this distortion or feel it is in some way "better" but for most sane people unexpected and unpredictable distortion is something to be avoided! Furthermore, once we get to 192k sampling rates and beyond it is impossible to correctly filter them according to the demands of the Nyquist Theorem. While it's extremely unlikely that this will result in any audible problems, it is in theory at least, lower digital fidelity. So don't get sucked into the marketing hype that 24/192k (or even worse 24/284) is somehow higher quality or higher fidelity than say 96k or 44.1k!
Well, bit depth/resolution with PCM can be understood as detail since higher bit depth means less quantization noise and lower noise floor which potentially masked low-level details before, BUT people don't understand that the noise floor is quite low with 16 bit to begin with! And yeah, they seem to think that higher resolution will also improve loud parts of the signal.