[gyrodec]

While I do think, once you are at state-of-the-art ADC levels the difference between 44.1 and high-res can be surprisingly small, I wonder why such demanding people as David Chesky have moved to high res (176.4/24 in his case). I trush his ears more than mine, and to be honest more than any other person who I don't actually know. So why did he, and the many like him, move to high-res? I agree than you need nothing more than 16 bit for consumer delivery, but I don't think anybody would argue the extra 8 bits are very, very useful for recording.

 

[gregorio]

Gyrodec - If you think about it, what people are hearing makes complete sense. If you've decided to create an DAC that upsamples everything to 96k, why would you bother implementing good filters at 44.1k? I'm not saying that people are lying or fooling themselves, they may indeed be hearing a difference. It's my ascertion that the difference they are hearing is in fact the reconstruction filters in their DAC, not an inherent improvement because of a higher sample frequency. Interestingly, the kind of artefacts I would expect from a weaker filter with phase (jitter) issues is exactly whats been described by those who feel 96k sounds better than 44.1k: A narrowing of the stereo soundfield, a loss of definition and weaknesses in the mid low frequencies. The point is, that providing a high quality ADC has been used in the first place, these artefacts are eliminated at 44.1. My point is that under these conditions and providing the listener has a DAC with a high quality 44.1k reconstruction filter, these same people would not now be able to discern any difference or if there is a difference it is more likely to be in favour of the 44.1k rather than 96k. The effects people are hearing are not due to any inherent weakness with the 44.1k format.

I would guess the reason David C has moved to 176.4 is that he has heard a quality improvement at this sample frequency with his particular ADC or it's entirely possible that he has fallen into the consumer trap of thinking that more data means higher quality.

24bit is really essential for recording digital audio but I dispute that it makes any difference whatsoever for the consumer.

 

[Budgie]

This thread makes my head hurt.



I want plug and play.

 

[gregorio]

Hi Budgie,

Tell me about it! The problem is that people are getting confused because digital audio theory uses some confusing terms and unless you have a good basic understanding of the theory then it appears to fly in the face of common sense.

For example, we talk about bit depth in terms of resolution. This is misleading as higher resolution in an image means better quality, it doesn't mean the same thing in digital audio. 24bit audio does not provide any more resolution than 16bit audio, the resolution of the audio is just as perfect at 16bit as it is at 24bit or indeed at any number of bits. The higher bit depths just define a lower noise floor.

I know I'm going to get some people arguing over my last paragraph and that's because it doesn't sound logical but it is logical if you understand the basic principle of a dithering quantizer.

How's your head now? lol

 

[CyberTheo]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif it's entirely possible that he has fallen into the consumer trap of thinking that more data means higher quality.

I don't think so...

 

[gyrodec]

gergorio

I hope I never gave the impression that I though a higher sampling rate will record any more information that is useful to humans, as opposed to dogs and bats. You are right, it doesn't. We can only hear to 20K (on a very good day) so 44 is more than enough to perfectly encode a 20k signal. And yes, it is the filters that people are hearing differences in, that was always my point.

I've been reading GearSlutz, pro recording version of this site, and many comments there support your point to a certain degree. For example, amny people think their mid-level ADC sounds much better at 96 than 44.1, but the prism and Lavry Gold croud say they see much less difference. But, the majority of them still say there is an audible difference, while your position seems to be than there is basically none. Just adifference of degree really.

 

[gyrodec]

gergorio

I just read a few pages from the end of the 16/24 bit thread, and I just wanted you to know I agree with almost every word of yours on that tread. Crowbar has some very interesting ideas and has though much of it through quite well, but it only feels right as a though experiment and I think your interpritation of the "real world" is likely spot-on.

I'm still doing to disagree slightly with you on this thread though.

 

[b0dhi]

Quote:

Originally Posted by gregorio /img/forum/go_quote.gif For example, we talk about bit depth in terms of resolution. This is misleading as higher resolution in an image means better quality, it doesn't mean the same thing in digital audio. 24bit audio does not provide any more resolution than 16bit audio, the resolution of the audio is just as perfect at 16bit as it is at 24bit or indeed at any number of bits. The higher bit depths just define a lower noise floor. I know I'm going to get some people arguing over my last paragraph and that's because it doesn't sound logical but it is logical if you understand the basic principle of a dithering quantizer.

Supply a credible reference (that means not yourself or a recording engineer) supporting your statement that "the resolution of the audio is just as perfect [equal] at 16bit as it is at 24bit or indeed at any number of bits".

 

[gregorio]

b0dhi - "Supply a credible reference (that means not yourself or a recording engineer) supporting your statement that "the resolution of the audio is just as perfect [equal] at 16bit as it is at 24bit or indeed at any number of bits"."

References, that's easy, Harry Nyquist, Shannon and all the other people who have contributed to the digital audio theorem. The whole point of a dithering quantizer is that the measurement (quantisation) errors present at any bit depth are converted into noise. The result is a perfectly linear system with noise. Doubling the quantisation points (adding another bit of data) results in half the number of quantisation errors and therefore half the noise. So each additional bit used to encode the audio results in the noise floor being reduced by about 6dB. So in 16bit audio the noise floor of the system is about 96dB below peak level, while in a 24bit system it is at 144dB. Of course in reality no piece of equipment can match the theoretical noise floor of a 24bit system. By the time we take the noise floors of all the pieces of equipment in a recording and playback chain, plus the physical limitations of amps, speakers and the noise floor of the recording venue, then even the noise floor of a 16bit (96dB) system has easily more dynamic range than required. The average high quality recording of a large scale symphony is likely to have a dynamic range of about 50dB, way below our 96dB limit in 16bit.

If we think of the quantisation points being like the rungs of a ladder: The common misconception is that adding more rungs to the ladder is going to make the rungs closer together and therefore improve the resolution. But that is not how digital audio works, the rungs of the ladder are fixed and always the same distance apart, adding more rungs just makes the ladder longer. To make our analogy with digital audio more accurate, even the short ladder would reach well beyond the top of the highest building in the world. 24bit audio has a specific application when recording, for the audience listening to a finished product there are no advantages of 24bit over 16bit whatsoever.

It's really rather simple and forms the basis of all digital recording. Obviously the mathematical implimentation of dither is quite complex, especially as for the last 10 years or so most professional dithering processors noise shape the dither, maintaining the statistical conversion of errors into uncorrelated white noise but then moving some of that noise into the high frequency where it will be less obvious.

SACD uses DSD technology which is based on one bit digital audio. Does SACD sound that much worse than 16bit CDs which uses 15bits more?

If you doubt any of this just go and do a little research into dithering and quantisation. This really is digital audio 101.

 

[gregorio]

Quote:

Originally Posted by gyrodec /img/forum/go_quote.gif the prism and Lavry Gold croud say they see much less difference. But, the majority of them still say there is an audible difference, while your position seems to be than there is basically none.

The theory is that there shouldn't be a difference. In practice there is a difference but it would be unnoticable. To record using Prism converters is going to cost about $12,000 per 8 channels of input. I've worked with Prisms, they're lovely. But if you're spending this amount on world class converters you're going to have a monitoring environment to match and have highly trained ears. Under these conditions you can sometimes tell a very slight difference in the filters. In a sitting room, even with audiophile equipment, you're not going to be able to detect any difference at all. Don't forget, that the discussion you're referring to is probably discussing recording at 96k vs recording at 44.1k. While this discussion is about 44.1k material being upsampled to 96k. When upsampling I would be extremely suprised if anyone can hear a difference using Prisms. It's more likely that the 96k upsampled audio will sound worse on a Prism than the 44.1k original.

 

[b0dhi]

Clearly you do not understand what the word "reference" means. But that's what I thought.

 

[gregorio]

b0hdi - Your are right, being a university lecturer I wouldn't have any idea what Harvard referencing is. I didn't havard reference my message because frankly I couldn't be bothered. If you are incapable of understanding my layman's explanation then go and read digital audio theory for yourself, AES publications are good in this regard. If you want an explanation in someone else's layman's terms then go and look up some articles in Sound on Sound or trawl through their web forum. If you don't want to do this, then don't blame me for your lack of understanding. If you want to believe the earth is flat you go for it but don't blame me for your lack of understanding because I haven't accurately harvard referenced Copernicus!

 

[b0dhi]

Hahahahahahahahahahaahaha.

Sorry that is just hilarious.

I'm well capable of understanding your "explanation", I just happen to know it's incorrect, which is why I'm asking you to supply a reference, because I know you won't be able to.

Please - since you're a university lecturer who is aware of what the Harvard referencing system is (lol), you should have no problem finding some lecture notes stating that "the resolution of the audio is just as perfect [equal] at 16bit as it is at 24bit or indeed at any number of bits". Although I'm not sure how you will do that due to the fact that time-domain resolution depends directly on bit-depth (given enough distance from Fs/2), and you might also have some trouble convincing me that a 1-bit dithered stream has the same "resolution" (that word, by the way, you are confusing because it relates more to sampling rate than bit-depth, but I will assume you're talking about quantisation resolution) as a 16-bit dithered stream.

 

[nick_charles]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif Hahahahahahahahahahaahaha. Sorry that is just hilarious. I'm well capable of understanding your "explanation", I just happen to know it's incorrect, which is why I'm asking you to supply a reference, because I know you won't be able to.

This is interesting. Could you do the reverse and provide the proof that your case is correct, I would be genuinely interested to see it, no irony here. I will have to ask you to limit your proof to peer reviewed journal or quality conference papers though. Generally speaking the onus is on rejecting the null hypothesis rather, in this case that a 24 bit reconstruction is the same as a 16 bit reconstruction.

Cheers

Nick

 

[gregorio]

"In an analog system, the signal is continuous, but in a PCM digital system, the amplitude of the signal out of the digital system is limited to one of a set of fixed values or numbers. This process is called quantization. Each coded value is a discrete step... if a signal is quantized without using dither, there will be quantization distortion related to the original input signal... In order to prevent this, the signal is "dithered", a process that mathematically removes the harmonics or other highly undesirable distortions entirely, and that replaces it with a constant, fixed noise level."

S.P. Lip****z, J. Vanderkooy and R.A. Wannamaker. 1991. Minimally-audible noise shaping. J. Audio Eng. Soc. 39, 836-852. and J. Vanderkooy and S.P. Lip****z. 1987. Dither in digital audio. J. Audio Eng. Soc. 35, 966-975

Vanderkooy - Present Research Activities: A study of the linearizing effect of dither in quantized data systems, especially digital audio, has resulted in a number of papers which show that properly dithered systems are totally free of distortion and noise modulation at all signal levels.

Try this link too: Phys. Rev. E 61 (2000): Robert A. Wannamaker, Stanley P. Lip****z, and John Vanderkooy - Stochastic resonance as dithering

So you know dithering theory is incorrect do you, well why don't you write your own thesis and disprove the theory of dithering which has been a routine proceedure in digital audio for more than 15 years.

Ignorance is bad enough but then to argue and broadcast your ignorance just makes you an idiot! Now why don't you supply a reference to a paper in as reputable a publication as the AES journal which proves the dithering theory incorrect.