Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: More misinformation (Read 111873 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

More misinformation

https://www.youtube.com/watch?v=bjTxEwlypA0

Can you believe these guys? For the flac/MP3 comparison, that is exactly the type of music where you will not hear much difference at all, because that is all digitally synthesized music.

Then, the bit where they say sample rate doesn't matter... and also that bit depth doesn't matter... ::)  As part of my Studio Sound Engineering and Music Production classes, the VERY first thing they tell you is that yes, sample rate and bit depth do matter, and that you do indeed hear the difference (We even did a proper A/B comparison in class) provided you have capable hardware for the play back, and files that were created at those settings.

  You see, while a sample rate of 44.1Khz will represent audio up to 22.05khz, it will be doing so only twice per cycle. the result is, a sine wave at 22.05Khz sampled at 44.1khz, will come out as a square wave, 90 degrees out of phase (Assuming your samples happen at peak and trough)now, the representation of waves does get more accurate as the frequency goes down, but imagine for a moment a sample rate of 44.1Khz on a sine wave of 11.025Khz. this wave will only be represented at 4 points each cycle. This again leads to a far less than accurate representation, at a frequency where you will definitely hear the difference.

  Bit depth on the other hand refers to the number of different steps at which the amplitude of the wave can be measured. At 8Bit, there are only 256 different possible levels. At 16Bit, there are 65536 different levels. at 24Bit will yield 16 777 215 different levels. Not only do larger bit depths more accurately represent the slopes of sound waves, they also help with dynamic range. 16Bit audio will have a dynamic range of approx 96db, while 24Bit will result in a dynamic range of approx 144db, a difference you will definitely hear in music.

Now i realize this might be a bit in depth and yes it does suffice to say that 44.1Khz at 16Bit is sufficient for the average person on cheap speakers with no capable hardware, But if you have a capable DAC, decent amp, and speakers that can actually represent audio accurately, there is a massive difference in going to 48Khz at 24Bit, and again a big difference going to 96Khz 24Bit, and again if you push it all the way to 192Khz at 32Bit.

But again, if you set your playback up at 192Khz 32Bit, and have decent hardware capable of this, but you play a file that is 44.1 at 16Bit, it wont sound any different to if it was setup at 44.1 16Bit. So it is important to know (If you want to compare, and get the most out of your system) what sample rate and bit depth were used during recording, editing, exporting, and so on, in order to ensure the file is never down-sampled.

Re: More misinformation

Reply #1 –
Quote from: Mikhale on

  You see, while a sample rate of 44.1Khz will represent audio up to 22.05khz, it will be doing so only twice per cycle. the result is, a sine wave at 22.05Khz sampled at 44.1khz, will come out as a square wave, 90 degrees out of phase.
 

You need to go back to school and get a better grasp on how sampling and reconstruction works.
Creature of habit.

Re: More misinformation

Reply #2 –
You are new to Hydrogenaudio, first welcome, and be open to learn.

It looks like, even though you are supposed to have a technological knowledge of how ditigal audio works, in reality, you don't.

I haven't listened to the whole video, but skipping through it, it seems that what they are saying is correct. Probably they talk too much and don't show why, but other videos show it.

I recommend you to view these two videos from xiph.org, the creators of (ogg) Vorbis and Opus:  http://xiph.org/video/ , especially the second video, where it shows why your explanation is wrong.

There are some other threads here that talk about these things (bitrate and samplerate), but as a quick startpoint, this is what you should know:

1) Sample rate and bit rate are important on the recording/mixing/post-processing/mastering stage.
You want to have the best SNR ratio on the beginning, or else you could be adding and adding more distortions and the final stage could sound worse than what it should.
Sample rate is not as much important on recording (once you have an adequate margin), but generally DSP plugins work better on some specific sample rates than others. Also, it could help on plugins like noise reduction effects, pitch scale and these type of DSPs.

2) Sample rate and bit rate is not so much important on distributing/playback.
We know quite well how the ear works, and so what are their limits. As an example, it doesn't matter if you are shown a micrometer or a nanometer of something, you will not see any difference (well, you will probably not see it at all) without a microscope.

3) Your undestanding of how digital audio (the samples of a sound) are converted back to analog is wrong. This is explained very well in that second video i mention. Audio is reconstructed. In a DAC there is something called a reconstruction filter. You should think of samples as the amplitude of a sound at the infinitesimal moment that it represents. To be more specific, the sample does NOT have the same value during the time that passes from the previous sample to the next sample. Watch the video.

4) There are not massive reproduction differences between listening to the same piece sampled at 44Khz 16bits to a 48Khz 24bits. There can be massive differences when using bad quality resamplers  on doing so (to convert from one to another), but nowadays it is not usual.  Many times, what is "massive" is the trick that our mind do to fool ourselves into thinking that a difference exists where there isn't one.

Said that, yes, 24bit versus 16bit might be ABXable, but I'm not sure right now about a test showing it. As you said, 16bit represents 96dB of SNR. (using dithering, that can be increased to more than 110dB).
An anechoic chamber (not sure if i wrote that correctly) has a 20dB SPL. A silent room is more like 30dB SPL. Usual noise on cities is around 60dBs to 70dB.  Music on music clubs goes from 90dB SPL to 110db SPL. (may be a bit higher too, but not too much).
So... how do you expect to hear the difference given the SNR of 16bits?

Re: More misinformation

Reply #3 –
Quote from: Mikhale on
Can you believe these guys? For the flac/MP3 comparison, that is exactly the type of music where you will not hear much difference at all, because that is all digitally synthesized music.
Most of the killer samples for this sort of test, like "fatboy", "eig", etc. (these refer to specific samples which have been heavily tested around here) are all electronic. They break codecs in strangely audible ways. That's not to say analogue music doesn't do the same. I'm very sensitive to loss in the "castanets" sample. This might seem counterintuitive, and well-trained analogue musicians hear loss in their preferred music more easily (I'm thinking of guruboolez and his love of harpsichord here). I'm a (ex?) DJ, so my preferences run that way.

Re: More misinformation

Reply #4 –
In addition to the xiph.org video, I suggest the multitude of fun tests and downloadable test files at http://www.audiocheck.net/

To add to the good posts above, some skepticism is warranted about the "proper" A/B comparison you did in class. It was almost certainly subject to expectation bias, if not also poor level matching (louder universally sounds "better"), or the use of sounds not derived from the same source (e.g. comparing different masterings). Even when they come from the same source, audible differences can occur if they were converted using a poor resampler. And if they were using a computer for playback, what steps did they take to assure the data fed to the DAC wasn't resampled by the OS or audio drivers along the way?

Also there is the fact that people will think they hear differences where there are none. For example, tell people you're doing A/B but then just play A every time, and you will find people swearing they hear bass differences and whatnot. A coin toss will predict the results just as well. That's why we like the ABX test around here. You have to consistently show you hear a difference. If both samples are the same, or they are different but you are unable to tell, then your score will be around 50%, a roll of the dice... if you really do hear a difference (fatigue notwithstanding), you will score 95%+.

Re: More misinformation

Reply #5 –
5% chance you can do the same or better by flipping a fair coin or 95% correct?  These two things are different and the difference is very important.

Quote from: Mikhale on
We even did a proper A/B comparison in class
Like mjb, I very much doubt this as well.

Re: More misinformation

Reply #6 –
I knew "score" would get me in trouble. Of course I meant 95% statistical confidence that you're doing better (same or better?) than a coin flip. Actual # of correct answers to get that confidence in 12 trials would be 10, but for 1200 trials it would only need to be 635, though it seems weird that you could be wrong 47% of the time and still be proving that you reliably hear a difference.

Re: More misinformation

Reply #7 –
"Reliably" isn't really the right word.  The 95% confidence level means only that there is only a 5% chance that you got your results by luck (aka coincidence) rather than by detection of a real difference.  You may only be to detect a difference in one test out of a hundred, but if you can keep that edge consistently over enough tests it can be considered a "reliable" reading (albeit a rare or highly marginal detection).

Put another way, if 20 monkeys randomly press buttons, one of them will achieve a 95% confidence level :)  If enough people perform a test and you cherry-pick the results you will be able to "demonstrate" that they detected almost anything you want to prove.  They say "lies, damned lies, and statistics" for a reason.

Re: More misinformation

Reply #8 –
If you are struggling to understand sampling, then the wiki page has a pretty good explanation:

https://en.m.wikipedia.org/wiki/Nyquist–Shannon_sampling_theorem

Probably best to read that before asking a bunch of engineers to explain it to you.

Re: More misinformation

Reply #9 –
Quote from: mjb2006 on
Actual # of correct answers to get that confidence in 12 trials would be 10, but for 1200 trials it would only need to be 635, though it seems weird that you could be wrong 47% of the time and still be proving that you reliably hear a difference.
Well, yeah, as the number of trials goes up the percentage of right trials required for statistical significance decreases. The p-value will not tell you the size of the effect, just the probability of observing _any_ effect by chance.
"I hear it when I see it."

Re: More misinformation

Reply #10 –
Your post is misinformation OP.
FLAC -> JDS Labs ODAC/O2 -> Sennheiser HD 650 (equalized)

Re: More misinformation

Reply #11 –
It's going to take me a few days to recover from the OP. You sound so sure of yourself but your understanding of digital audio is severely flawed. I also recommend you read the links and watch the videos suggested earlier.

Re: More misinformation

Reply #13 –
Seems like OP needs to unlearn Schiit before he can come up to speed.
AVI ADM 9RS | HD 650 | HD 598 | Asus Essence One Muses Edition DAC & HA | Garage1217.com Project Ember HA | Asus STX DAC & HA | Fiio E12 | iDevices

To get added to my Whatsapp Audiophilia group, PM me!

Re: More misinformation

Reply #14 –
Quote from: Mikhale on
You see, while a sample rate of 44.1Khz will represent audio up to 22.05khz, it will be doing so only twice per cycle. the result is, a sine wave at 22.05Khz sampled at 44.1khz, will come out as a square wave, 90 degrees out of phase
That's not how audio DACs convert digital signals to analogue ones. They typically come much closer to the ideal reconstruction than this "simple stairstep approach". The ideal reconstruction would be to treat the samples as a sequence of scaled dirac pulses and filter those with an appropriate lowpass filter. That's the theory. In practice, you deal with an "oversampling DAC" which involves
  • digital band-limited interpolation to a higher sampling rate -- I'd say this is where most of the magic happens
  • a "stairstep" intermediate analogue signal but with much much smaller steps due to 1.
  • an analogue lowpass filter that basically gets rid of the steps only.

Quote from: Mikhale on
the representation of waves does get more accurate as the frequency goes down,
That's not true. The sampled representation is equally good for all frequencies between 0 (inclusive) and the Nyquist frequency (exclusive).

Quote from: Mikhale on
but imagine for a moment a sample rate of 44.1Khz on a sine wave of 11.025Khz. this wave will only be represented at 4 points each cycle. This again leads to a far less than accurate representation, at a frequency where you will definitely hear the difference.
The representation is perfectly fine. You think it's not probably because you don't (yet) know/understand how a proper reconstruction works. But there is no shame in that. It's not exactly trivial. Really understanding this might require you to know what the Fourier transform of an infinite train of dirac pulses looks like and how multiplication and convolution are connected via the Fourier transform. A decent audio DAC has no trouble reconstructing the 11 kHz sine wave in this setting. It won't give you a triangle wave.

Quote from: Mikhale on
Bit depth on the other hand refers to the number of different steps at which the amplitude of the wave can be measured. At 8Bit, there are only 256 different possible levels. At 16Bit, there are 65536 different levels. at 24Bit will yield 16 777 215 different levels. Not only do larger bit depths more accurately represent the slopes of sound waves, [...]
You don't seem to be familiar with dithering. With proper dithering one does not have to worry about representing slopes. Properly dithered quantization is only adding uncorrelated noise, nothing else.

Quote from: Mikhale on
But if you have a capable DAC, decent amp, and speakers that can actually represent audio accurately, there is a massive difference in going to 48Khz at 24Bit, and again a big difference going to 96Khz 24Bit, and again if you push it all the way to 192Khz at 32Bit.
Not really. There is a scientific method for determining whether this is true. It is designed to rule out the placebo effect. You should try it.

Quote from: Mikhale on
But again, if you set your playback up at 192Khz 32Bit, and have decent hardware capable of this, but you play a file that is 44.1 at 16Bit, it wont sound any different to if it was setup at 44.1 16Bit. So it is important to know (If you want to compare, and get the most out of your system) what sample rate and bit depth were used during recording, editing, exporting, and so on, in order to ensure the file is never down-sampled.
As an intermediate format during production a higher sampling rate and bit depth have benefits, sure. It gives you more headroom, allows lower latencies and allows you to do many processing steps without increasing the noise too much. But as an end product for you to comsume, a sampling rate beyond 48 kHz does not give you anything.

Re: More misinformation

Reply #15 –
SebastianG is right.

16/44.1 FLAC gets 98-99 per cent analog audio information available and it is not an error to use this for HiFi.
24/96 FLAC gets 99,9999 percent.

Anything more when digitizing is overkill and can make things worse.

Other things is the reproduction where some soundcards/DACs work better at 48/96 kHz than at 44.1 due to having one clock only (majority of Realtek or SoundBlasters are an example of this). And for sure they work better at 24 bit, because of volume control and possible resampling in OS (Windows). But playing 16 bit source at 24 bit card mode is completely transparent so there is no reason not to do it.

44.1/48/96 kHz is more tough. It would be probably ideal if a new standard for common audio releases would be introduced, at 96 or 48 kHz, to prevent resampling standard CDs when playing on 96/48 KhZ kHz mode (and not utilizing WASAPI). This was actually DVD-Audio with 24/96 record. It would have been great if the recording companies now regularly release the recordings in 24/96 or 24/48 format instead of plain CD, the debates about 16/44.1 and/or dithering would then be to the vast extent obsolete and the customers would get the highest reasonable quality record. 

Re: More misinformation

Reply #16 –
Quote from: jumpingjackflash5 on
SebastianG is right.

16/44.1 FLAC gets 98-99 per cent analog audio information available and it is not an error to use this for HiFi.
24/96 FLAC gets 99,9999 percent.

Anything more when digitizing is overkill and can make things worse.

Other things is the reproduction where some soundcards/DACs work better at 48/96 kHz than at 44.1 due to having one clock only (majority of Realtek or SoundBlasters are an example of this). And for sure they work better at 24 bit, because of volume control and possible resampling in OS (Windows). But playing 16 bit source at 24 bit card mode is completely transparent so there is no reason not to do it.


96k is totally pointless.

24 bit to 16 bit is not resampling, and how Windows treats audio has nothing to do with 24 bit.

Virtually all DACs have one clock. That doesn't mean they can't do multiple sampling rates.

Re: More misinformation

Reply #18 –
But Windows does resample in shared mode to the setting set at the sound card properties. And resampling at 16 bit resolution -  16/44.1 to e.g. 16/48 or 16/96 could be theoretically of slightly worse results than resampling at 24 bit resolution to 24/48 or 24/96.

Also digital volume control can be more precise when the output is 24 bit, since Windows do all calculations 32 bit internally and then utilize the output resolution of the device available.

I agree that difference between 48 and 96 is minimal, if any. But if there is a new common format introduced like CD is (was), there should be a debate if 48 or 96 kHz is suitable as a common standard.

Re: More misinformation

Reply #21 –
Well, I think that some of those debatable claims could have disappeared from the public by introducing new standard for distributing audio, as I wrote at 24/48 or 24/96. This way the "near limit" issues of CD (16/44.1, dithering, filtering etc) would be eliminated and on the other hand the unjustified dreaming about "superhifi" like 192 kHz or so would disappear. People would get the highest reasonable quality available for playing on todays devices. And this was the core of my posts, not personal offendings.

Re: More misinformation

Reply #22 –
Quote from: jumpingjackflash5 on
But Windows does resample in shared mode to the setting set at the sound card properties. And resampling at 16 bit resolution -  16/44.1 to e.g. 16/48 or 16/96 could be theoretically of slightly worse results than resampling at 24 bit resolution to 24/48 or 24/96.

Also digital volume control can be more precise when the output is 24 bit, since Windows do all calculations 32 bit internally and then utilize the output resolution of the device available.

I agree that difference between 48 and 96 is minimal, if any. But if there is a new common format introduced like CD is (was), there should be a debate if 48 or 96 kHz is suitable as a common standard.



The input and output formats have nothing to do with resampling.

Digital volume control is rarely used on pcs. Even 1 dollar parts have analog volume these days.

96k is pointless.

Re: More misinformation

Reply #23 –
Digital volume control is used on a lot of Windows workstations, since it is sometimes more convenient to set volume with mouse that on the receiver (loudspeaker).

Windows does resampling e.g. when not utilizing WASAPI and the shared sample rate does not match the rate of FLAC played, e.g. in Windows media player. And then it (slightly matters if it resamples to 16 bit or 24 bit "space", and does all the calculations 32 bit internally.

Thus, it can be useful to set the sound card shared mode to 24 bit and "native" kHz mode, typically 48 kHz or 96 kHz for cheaper (onboard or common Soundblasters) cards. Or, if internal DAC does 44.1 sampling very well, to 24/44.1, but some cards work at 44.1 worse than on 48 kHz.

If there is general agreement on new 24/48 audio standard for album releases, then 96k could become obsolete for common discussions.

Re: More misinformation

Reply #24 –
If you really don't have analog volume control I would upgrade to something better.

I dont know what you think resampling does but its not right.