24bit vs 16bit, the myth exploded!

[Kaffeemann]

 
For fun I reripped it as MP3 v0
 

 
very similar but.... on the loud bits
 

 
Not clipping according to Audacity but right on the limit more often than with either Cd copy
 

 
Keep in mind that there might be intersample clipping. Audacity does not display the real waveform so you can't see it.
This is how extreme intersample clipping looks like:

 
 
Back to topic. Benchmark Media Systems published this article on the bit depth stuff:
http://benchmarkmedia.com/blogs/news/15121729-audio-myth-24-bit-audio-has-more-resolution-than-16-bit-audio

 

[audiosampling]

Just a little intrusion here, to tell that the 16-bit v/s 8-bit blind test page that has been often cited in this thread, has been updated. Now using Neil Young's own music. I am such a bad guy indeed...
 
http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

 

[Hungryhoss]

Just a little intrusion here, to tell that the 16-bit v/s 8-bit blind test page that has been often cited in this thread, has been updated. Now using Neil Young's own music. I am such a bad guy indeed...


 


http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

 


Genius. I am glad you are going there. I look forward to the result.

 

[Yourt]

I've started reading and advanced well through pages 1 tot 20. then I started skipping some parts of the thread, ending up reading the last set of pages (boy, what a job, let alone the fact that some clicking to referenced pages need sto be done too).
 
What I already knew got confirmed now in a scientific way. One can indeed show and proof technical / theoretical (show meaning either audible or (mostly) in statistics) differences between different formats, both in dynamic range (bit depth) as in bitrate (the sampling).
 
Fact is that for real life use there are very little musical works that could benefit from any of these differences, because of a bunch of limitations in the chain (production, reproduction, sensitory system and psyche of the listener).
 
My concerns were at the stated "subharmonics and overtones, 30K cymbal stuff and some more of those together with the influence of the sampling into the lower frequencies (teh last one been shown to be able to sometimes even worsen the final result).
 
Considering phono... there's more to it than just LP DR figures: RIAA curves and preamp rumble filters with low frequency rolloff have an impact on the final sound shaping. Anyhow, a good player attached to a nice set of preamp-amp-LS has very little hissing or noise (same goes as above: crank up the volume and hear the noise, and next when the music kicks in hear a beep... ).
 
As for commercial reasons: I got a recording from the library last week, SA-CD. So I was figuring: let's give it a go for some hi-res testing. No need to, because after a simple check on my laptop (Audacity) it turned out to be 16/44 material.
 
So I'll leave it at looking for good quality material (starting at good recording and mastering and decent pressing/burning/ripping) and enjoy music at 16/44, or eventually - but not necessarily - higher.

 

[azteca x]

As for commercial reasons: I got a recording from the library last week, SA-CD. So I was figuring: let's give it a go for some hi-res testing. No need to, because after a simple check on my laptop (Audacity) it turned out to be 16/44 material.

Your computer was reading the CD layer. SACDs (which are encoded as DSD) can only be played in authorized SACD players and ripped by early PS3s. So yes, of course the files you got were 16/44.1.

 

[Dark_wizzie]

  Just a little intrusion here, to tell that the 16-bit v/s 8-bit blind test page that has been often cited in this thread, has been updated. Now using Neil Young's own music. I am such a bad guy indeed...
 
http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

It's a convenient test but somebody criticized the test because it ain't high DR content. (Or something.) But that obviously didn't get the guy to conduct his own Foobar abx test.

 

[RRod]

  It's a convenient test but somebody criticized the test because it ain't high DR content. (Or something.) But that obviously didn't get the guy to conduct his own Foobar abx test.

 
The page itself says that you can readily hear the effect of 8-bit, but that the effect lessens with much popular music, exactly because it often isn't high DR content.  It would be interesting to have an example of, say, the ending of Mahler's 6th in 16-bit-noise-shaped vs. 24bit.

 

[Hudson]

  Just a little intrusion here, to tell that the 16-bit v/s 8-bit blind test page that has been often cited in this thread, has been updated. Now using Neil Young's own music. I am such a bad guy indeed...
 
http://www.audiocheck.net/blindtests_16vs8bit_NeilYoung.php

 
Is the 8-bit sample dithered and noise shaped?

 

[audiosampling]

No... because noise shaping would add distinctive noise in this case (during the fade out at the end of the sample... one would fade into noise)... and make the blind test too easy to succeed

 

[jcx]

that seems backwards - truncation or rounding adds distinctive spitting, hissy amplitude jumps during fades, good dither gives smooth music signal fades below the lsb - but can add audible but constant noise that we can easily "listen through"
 
 http://audio.rightmark.org/lukin/dither/dither.htm
 
noise shaping is applied with dither, changes the dither noise spectrum to be lower where our hearing is most sensitive, reducing dither noise audibility
 
 
I bet few would ever pass a 8/192 shaped dither comparisons with music playing at reasonable levels (if their setup didn't distort, create audible IMD difference products with the ultrasonic dither)
 
http://www.meridian-audio.com/w_paper/Coding2.PDF 

 

[Hudson]

  No... because noise shaping would add distinctive noise in this case (during the fade out at the end of the sample... one would fade into noise)... and make the blind test too easy to succeed

 
Are you sure? Is it your website? So no dithering either? I just want to clarify as I would have expected noise shaping to reduce the amount of audible noise during the fadeout, not increase it. Cheers.
 

  that seems backwards - truncation or rounding adds distinctive spitting, hissy amplitude jumps during fades, good dither gives smooth music signal fades below the lsb - but can add audible but constant noise that we can easily "listen through"
 
 http://audio.rightmark.org/lukin/dither/dither.htm
 
noise shaping is applied with dither, changes the dither noise spectrum to be lower where our hearing is most sensitive, reducing dither noise audibility
 
 
I bet few would ever pass a 8/192 shaped dither comparisons with music playing at reasonable levels (if their setup didn't distort, create audible IMD difference products with the ultrasonic dither)
 
http://www.meridian-audio.com/w_paper/Coding2.PDF 

 
This makes more sense to me.

 

[stv014]

No dither is not always necessarily a problem, especially compared to simple non-shaped dither. While it obviously makes a major difference with artificial test signals, complex music that is never very quiet (e.g. during a long fade-out to silence at the end) may have enough entropy in the least significant bits that it essentially becomes "self-dithered". Basic dithering always adds a few dB of extra noise floor (which can make an audible difference at 8-bit resolution), but it is "safer" because it guarantees that the quantization error is uncorrelated white (if not shaped) noise. By the way, the Meyer&Moran Red Book vs. SACD tests did not use dithering either.
 
You can test dithering to various resolutions vs. no dithering with the samples from this older thread. The original 24-bit sample has been quantized from 16 to 8 bits with a very simple "shaped" triangular dither, and the 10-bit version is also available with white noise TPDF dither, uniform dither, and no dither. Try listening to the files, or extract the residual with an audio editor. Ideally, it should sound like pure noise, with no recognizable bits of the original signal.

 

[Hudson]

  No dither is not always necessarily a problem, especially compared to simple non-shaped dither. While it obviously makes a major difference with artificial test signals, complex music that is never very quiet (e.g. during a long fade-out to silence at the end) may have enough entropy in the least significant bits that it essentially becomes "self-dithered". Basic dithering always adds a few dB of extra noise floor (which can make an audible difference at 8-bit resolution), but it is "safer" because it guarantees that the quantization error is uncorrelated white (if not shaped) noise. By the way, the Meyer&Moran Red Book vs. SACD tests did not use dithering either.
 
You can test dithering to various resolutions vs. no dithering with the samples from this older thread. The original 24-bit sample has been quantized from 16 to 8 bits with a very simple "shaped" triangular dither, and the 10-bit version is also available with white noise TPDF dither, uniform dither, and no dither. Try listening to the files, or extract the residual with an audio editor. Ideally, it should sound like pure noise, with no recognizable bits of the original signal.

 
I see. I understand the theory but I'm just curious to see how low a resolution I need to go to before I'm able to perceive an audible difference.
 
I will take a look at those files and abx when I get a chance. For no real reason I always expected quantization noise on a non-dithered low res sample to be more noticeable, but I guess not. I can't tell the difference between those files on the audiocheck website.
 
Cheers

 

[castleofargh]

I guess at some point it ends up being a question of listening volume levels and how quiet your room is. I know that with 16bit what I usually end up hearing first is noise from the amp section, and going 24bit doesn't help a bit

.

 

[audiosampling]

   
Are you sure? Is it your website? So no dithering either? I just want to clarify as I would have expected noise shaping to reduce the amount of audible noise during the fadeout, not increase it. Cheers.
 
 
This makes more sense to me.

Yes, this is my website. And yes, I am 100% sure: there is no dithering applied. Actually, I did apply dithering first, but the fade out (at the end of the extract) would then fade out into audible noise (with or without shaping). So, it would have been very easy for people to find out which was the 8bit version, by simply listening to the fade out. That's what I wanted to avoid and why I removed the dithering.
 
I am very aware of what dithering does and how it works. BTW, I wrote the following page, providing audio examples and audible tests:
 
http://www.audiocheck.net/audiotests_dithering.php
 
It shows how dithering, applied to an 8-bit file, increases the dynamic range by 18dB!