24bit, SACD and DVD-A recordings. (first update)

[wnmnkh]

One of my favorite online stores is Linn Records store.

They really provide excellent recordings and great songs which make my phones truly rocks. They really have some excellent catalogs for Jazz and Classical music.

They give you some options; you can get CD, SACD and/or vinyl (not common option, but good for analog people if can be ordered) or you can download with various quality as low as 320kb mp3 to as good as 24bit FLAC. Of course, the most expensive one is 24bit FLAC (still far cheaper than buying SACDs and shipped to US from UK, and I only have stereo setup and really no intend to go for multi-setup.)

I only buy 24bit recordings. Well, I buy these because they are the BEST quality ones available on the market. It is technically superior to lossy formats, CD and probably vinyl for sound quality, while practically superior to SACD and DVD-A for convenience (come on, I want to rip all of them and put into my server. Never want to mess with compact discs) So I want the best, so I buy the 24bit FLAC files.

Now, notice I use 'technically' for sound quality. The question is, of course, that whether 24bit is 'practically' superior to mp3, CD and vinyl for sound quality.


Well.... it was pretty easy; buy a favorite song from Linn Records, with different quality... One for typical CD quality (16bit) and SACD/DVD-A quality (24bit.) Now, I checked my Foobar's setting. ASIO checked. 24bit playback checked. select the files and launch the ABX comparator......

So, as a first test, I decided to compare mp3 ripped one from 16bit source and original 24bit one. I clicked the 'play A' button, and a while later 'play B' button for some practice before I go for clicking 'play X' and 'play Y' buttons.




All of sudden, I could foresee the result of this.

I realized that source A and B are identical for me. Of course, the result was not so surprising as well.




Second part.


Have you ever tried this? You can downsampling your music files using the foobar2000 to mere 8bit. You will hear a lot of noise due to higher noise floor, yet overall detail of the music remains intact. For recording, the highest bit is always better since you can reduce a lot of noise. But we are end-users, and we really do not need this higher than 16bit (For SNR concern: if properly mastered, the maximum possible effective SNR of CD is whooping 111, I wonder how many people really use/hear with this extreme SNR.)

In fact, I did similar tests in past ; I used Beresford DAC via optical out, and I also used my Denon DVD/SACD player (1920) to compare SACD ones and normal CD (it was not ABX,) and the result was always same; I just could not hear any difference.


A while ago I have acquired Integra C-501XD, a really decent, old CD player, probably better than Denon 1920. I connected this to my beta22, put some old CD (made in 1980s), and was deeply surprised how can sound be this good. This taught me something; it is all about mastering, not the format.

I suspect, what we really need are some GOOD CD masters, which use all of the capabilities of CD. To be honest, I estimate most of today's CD masters, other than from classical records, are just far from being reasonable, let alone 'close to perfect.'


**wait for the last part of this....**

 

[Arainach]

Quote:

So, as a first test, I decided to compare mp3 ripped one from 16bit source and original 24bit one. I clicked the 'play A' button, and a while later 'play B' button for some practice before I go for clicking 'play X' and 'play Y' buttons.

That doesn't prove 24-bit isn't worth it, it only shows that any differences between 16-bit and 24-bit aren't noticable once the sound is compressed to MP3.

 

[ClieOS]

Pouring either 1 liter or 5 liters of water into a 100ml cup will still give you a 100ml cup full of water at the end.

 

[Pio2001]

As I understand it, wnmnkh did not compare two mp3, but directly the worst (mp3) to the best (24 bits wav, sample rate unknown).

I did the same, following the example of a french forumer here : homecinema-fr.com • Afficher le sujet - Quel intérêt du 24/96 en hifi ?

He is comparing 192 kHz 24 bits vs 44.1 kHz 16 bits.
Since my soundard can't play 192 kHz, I compared 96 kHz 24 bits to 44.1 kHz 16 bits.

For him, the difference is obvious. He is currently running the ABX. For me, no difference through Sennheiser HD600 headphones.

However, I then used a different method : instead of listening to music, I listened to the silence before it, at very high volume (about 90 dB(A) SPL RMS, with peaks at 113 dB, if I was listening to the music), and I used closed headphones (Superex Pro-B VI) in order to get the best possible signal-to-noise ratio.
I succeeded the ABX : 10/10.

I chose this method after having undergone an online blind test about quantization noise, from 12 bits to 16 bits. So I was already used to the sound of low definition digital : hiss, no more, no less. Quantization noise is just background noise, modulated by the signal.
So the best way to hear it is during quiet parts.

My result confirms the findings of the AES : the difference between 16 bits and 24 bits is audible at very high listening levels. And I can add that the difference is just background noise. It is stronger in 16 bits than in 24 bits, nothing more.

I also tried the Mustang challenge of ff123 (samples available here : Samples for Testing Audio Codecs ). The highest lowpass that I could hear was 13 kHz, while at that time I could hear 15 kHz sines (now my audition have dropped to 14 kHz). In blind tests, the highest lowpass that we can hear on music is generally a couple of kHz below our threshold of hearing for high frequencies.
It means that for me, a digital format of 28 kHz / 16 bits would be enough for transparency.

Nota : the HD600 have a strong treble roll-off (-10 dB in high frequencies). Results could have been different with a Beyer DT990, for example, that outputs the same treble at +5 dB.

 

[ClieOS]

I guess that bring up the question of whether wnmnkh's computer setup is capable of playing those 24bits music?

 

[Pio2001]

With Foobar2000 in 24 bits playback, ASIO drivers and an E-MU soundcard, 24 bits playback is OK for him.

But hearing the 16 bits quantization noise have required a max playback volume of 113 dB (insane), and the use of closed headphones in a parfectly calm environment. The noise is 96 dB below the full scale, which, for the file I was listening to, meant a noise level of 18 dB ! A mosquito makes more noise than that !

Actually, in the recording that I was listening to, the signal level were, by order of amplitude
-The 16 bits quantization noise
-Stronger than that, the recording studio background noise + microphone hiss. I'm not sure if the 16 bits noise can be heard behind the microphone noise... I was listening to the initial digital fade-in during which all the background noises come forth.
-Much louder, the movements and breathing of the musicians, just before playing.
-Deafening : the first violin note.

 

[Budgie]

Excellent information! I have alway found that cd's can sound fantastic at redbook rates, and the bad sounding recordings end up sounding bad due to what happens to them before they get burned to cd. The format is not what's at fault, as much as the process. Bad engineering etc, is a bigger problem. Improving the format, while worth doing, won't help if the recording process is at fault. Garbage in...........................

 

[wnmnkh]

To clarify, I compared 320kb mp3 ripped from 16bit source and ORIGINAL 24bit source (NOT ripped to mp3) So Pio2001 is correct about this (comparing mp3 vs original 24bit flac) The sample rate of the test music files ranges 88 to 96.


My setup for testing was Foobar2000 -> ASIO -> E-MU balanced out->balanced beta22 -> HD650. I doubt I would hear difference using my other phones.

And yes, you can distinguish 16bit and 24bit using noise floor, but in such case the volume is simply insane.

 

[HeadLover]

amm, doesn't 24 bit mean there is much more place for the recording eng to make it better after all? With less noise and so on ?

 

[BigTony]

The flaw in the comparison between SACD and redbook CD is that SACD DOESN'T use PCM for conversion, and most of the sonic problems regarding CD's is in the conversion from Digital-to-Analogue.
SACD avoids this problem by not using PCM, and thus the main failings of CD's are removed. Add to this a greater SNR, and there is a big improvement in sound with SACD over CD, easpecially in comparison at the same price point (a $200 SACD will blow a $200 CD player out of the water).
However, many SACD actually use PCM for the conversion to analogue... yes we all wonder why, but you can easily and cheaply find one that doesn't.

 

[LnxPrgr3]

Quote:

Originally Posted by BigTony /img/forum/go_quote.gif The flaw in the comparison between SACD and redbook CD is that SACD DOESN'T use PCM for conversion

What does DSD have over high resolution PCM?

 

[nick_charles]

Quote:

Originally Posted by LnxPrgr3 /img/forum/go_quote.gif What does DSD have over high resolution PCM?

There is a (close to) religious debate over the merits of PCM vs DSD. There are a stack of AES papers on this subject - the key protagonists in this debates being SONY, Philips, Meridian and the ARA. Both sides point to fatal technological flaws in the other's design.

What does it mean in practical terms , well Blech and Yang (2004) found that in their tests that only 4/110 subjects could actually detect a difference between DSD and 24/172 PCM[1]. These four tested it using single instrument samples through headphones and all were Tonnmeister students with good hearing. The other 106 failed to detect a difference, through speakers nobody could detect a difference at all.

1. B and Y also tentatively suggested that it was possible that these 4 could hear not a difference but heard a difference in the switchover noise which was subtly different between PCM and DSD samples.

 

[Pio2001]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif What does it mean in practical terms , well Blech and Yang (2004) found that in their tests that only 4/110 subjects could actually detect a difference between DSD and 24/172 PCM[1].

I disagree with this conclusion. There were 145 individual results, because some listeners tried the test under different conditions, four of which fell above the 5% probability threshold of guessing.

Since the majority of the results are close to chance, in average, we can expect, by definition, to get 5% of them above the 5% threshold. So, with no one hearing a difference, we should have in average 145 x 5 / 100 = 7 positive results.
The study did not even match this random distribution, let alone a significative distribution.

...exept the one who scored 20/20 ! Even among 145 tests, 20/20 remains a significative result : one chance in a million of guessing. Divided by 145, it still remains one chance out of 7000.

So there was in fact one positive result among 145, not 4.

 

[nick_charles]

Quote:

Originally Posted by Pio2001 /img/forum/go_quote.gif I disagree with this conclusion. There were 145 individual results, because some listeners tried the test under different conditions, four of which fell above the 5% probability threshold of guessing. Since the majority of the results are close to chance, in average, we can expect, by definition, to get 5% of them above the 5% threshold. So, with no one hearing a difference, we should have in average 145 x 5 / 100 = 7 positive results. The study did not even match this random distribution, let alone a significative distribution. ...exept the one who scored 20/20 ! Even among 145 tests, 20/20 remains a significative result : one chance in a million of guessing. Divided by 145, it still remains one chance out of 7000. So there was in fact one positive result among 145, not 4.

100 sets of trials were with speakers and 45 were with headphones.

So for speaker trials the 75%+ criteria was achieved by 0/100 and for headphones it was managed by 4/45.

Now if you treat headphones and speakers as separate conditions then you coud argue that 4/45 is beginning to look interesting.

However I would not base my choice of medium on even a 10% chance of hearing a difference with simple samples and well trained and young ears, possibly if I were 20 years younger

No, not really...

 

[wavoman]

The statistical reasoning presented by some posters here, while correct, is based on classic significance testing, which begins with the assumption (the "null hypothesis") that there is no difference between the two systms and will not move from that hypothesis unless there is major-league evidence in the other direction. It also assumes a homogeneous population vis-a-vis the ability to detect a difference.

Not appropriate here.

Think like this: suppose there is in fact a small difference, not always obvious, but some of the time A sounds better than B to some individuals (and never the other way).

With this hypothesis, people who really can hear a difference some of the time, and would like therefore to own A instead of B, still post results that would seem to only be chance.

The way out of the bind is to isolate the subjects who show a preference for A over B (not significant, but in the right direction), and re-test them.

If over-and-over they prefer A to B even by a small majority, but the direction is consistent -- guess what? There is a difference.

This is about the "size of the effect" and "power analysis". The number of trials etc. can all be worked out.

Moreover the published tests are not well done, they don't simulate real listening, they ask for difficult (A/B/X) instead of realistic (A>B) comparisons, etc. etc. ... all discussed in other threads.