[Krutsch]

   
There's nothing special about 16bit; it just happens to be the bit-depth that gets the necessary dynamic range for the broad range of music out there. 

 
I have to believe that the adoption of 16-bit was driven by the processors of the era; and 8 bit was obviously too small and 24-bit would take way too much space. Everything else, sampling frequency, playing time, and disc diameter were choices.

 

[RRod]

   
I have to believe that the adoption of 16-bit was driven by the processors of the era; and 8 bit was obviously too small and 24-bit would take way too much space. Everything else, sampling frequency, playing time, and disc diameter were choices.

 
Probably the way to put it is that the ability of processors to handle 16bits meant digital was finally viable, as the requisite dynamic range was finally possible (though I know 14-bits was favored by one company giving input into the mix).

 

[limpidglitch]

   
There's nothing special about 16bit; it just happens to be the bit-depth that gets the necessary dynamic range for the broad range of music out there. Sometimes for fun (just to show you how weird some of us are), I find the minimal PCM bit-depth and sample rate at which I cannot pass an ABX test of a track vs. the original Redbook version. I have plenty of classical tracks that work fine at 14bit, lots of non-classical tracks that can go down to 12bits, and the occasional brick-wall special that is indistinguishable at 8bits. Since I'm older than some of the folks on here (not bigshot obviously ), I can usually also down-sample to 36kHz or so without affecting quality either.
 
You can see then, perhaps, why some of us get miffed at the suggestion that people need MOAR bits and samples. If 12/36 in indistinguishable to me from 16/44.1, why on earth would I want 24/192? And if someone else says their ears are much more super than mine, then I am naturally skeptical if they haven't done the same type of experimentation I have with their own hearing. People are all happy to put in the work of buying gear and hi-res material and writing subjective reviews, but they scoff at those of us who go a step further and put in the work of *testing* these things against the limits of our own organism.

 
I've been playing around with lower bit-rates as well, more specifically 8/44.1, to see for myself the effects of dither and noise shaping. (linky)
I must say the results impress me. Although claiming transparency would be to take it a bit too far, seeing how low you can get the noise when having only 256 levels available makes it pretty obvious how when having 65 536 levels available we can make the noise completely disappear.

 

[lamode]

  I'd like to add that I compared the size of some lossless 16 bit files to lossless 24 bit files, and it seemed the size difference was about 100%, rather than the 50% it is for uncompressed. I guess it is easier to compress 16 bit files. This raises the stakes beyond what I was expecting. 
 
The main thing I was trying to say by bringing up lossy is that 16 bit is an arbitrary standard. It seems to me that people in this thread treat it like an ideal value, where everything above 16 bit lossless is a pointless waste of space. What is to stop someone who listens from high quality lossy files from saying that it is the sweet spot and 16 bit lossless is a pointless waste of space? Or that 24 bit is the sweet spot, and DSD is a waste of space? Is there anything particularly special about 16 bit lossless?

 
I already gave you an example which demonstrated quite clearly, I thought, why 16 bit is enough for real life situations. I'm almost certain that you have never heard 16-bit quantization noise in your life and therefore it's just not an issue.

 

[RRod]

   
I've been playing around with lower bit-rates as well, more specifically 8/44.1, to see for myself the effects of dither and noise shaping. (linky)
I must say the results impress me. Although claiming transparency would be to take it a bit too far, seeing how low you can get the noise when having only 256 levels available makes it pretty obvious how when having 65 536 levels available we can make the noise completely disappear.

 
Fixed format PCM is a bit hammer and nail in this regard. There are certainly sections of many tracks that can be successfully captured with fewer than 16 bits or with frequencies lower than 22kHz, but with a fixed format you just go with the flow.

 

[stv014]

  But also, I think room noise is usually constant and can be tuned out due to habituation.

 
Dithered quantization noise is also constant, and uncorrelated to the input signal. It is possible to hear a tone at less than -100 dBFS level even after 16-bit quantization noise is added (about -95.8 dBFS A-weighted for simple white noise TPDF dither).
 

  I'd like to add that I compared the size of some lossless 16 bit files to lossless 24 bit files, and it seemed the size difference was about 100%, rather than the 50% it is for uncompressed. I guess it is easier to compress 16 bit files.

 
Of course it is, because the extra 8 bits added by increasing the sample size to 24 bits usually contain mostly noise that cannot be compressed. It is not uncommon for recordings to have too high analog noise floor even for 16-bit PCM.

 

[Lespectraal]

 
Glad to see that this person I've been following for years has some sense in him.

 

[sonitus mirus]

  Glad to see that this person I've been following for years has some sense in him.

 
This reviewer has a great presentation style to go along with a high quality video production.  I subscribed.  Thanks for sharing.

 

[Krutsch]

   
This reviewer has a great presentation style to go along with a high quality video production.  I subscribed.  Thanks for sharing.

Hmm... this one I won't be sharing with others. Fun to listen to, but isn't in the same league as the earlier example we were discussing, which endeavored to explain *why* you don't need HD audio to a layperson audience.

 

[limpidglitch]

 
Hmm... this one I won't be sharing with others. Fun to listen to, but isn't in the same league as the earlier example we were discussing, which endeavored to explain *why* you don't need HD audio to a layperson audience.

 
Yeah, a bit condescending, without really bringing any information to the discussion. Not much better than his peers on 'the other side'.
(And stop it with the arms!)

 

[prot]

High end math degrees can't hear better than my ears, or your own ears. If you only have measurements of 1 type of hearing (the frequency range) then you only have math for that 1 domain.

Timing and spatial recognition in the simulated stereo field is where the major change is, and it's where all the 16bit holdouts like to avoid. The MP3 people outright dismiss all of it. How convenient to throw it away since you don't have math for it.

Vinyl people and recording engineers and musicians and music lovers hear it because it's "presence".

16bit has really impressed me through the ponoplayer, it is a decent format. But the 24bit files just have extra room, extra goodness, extra presense that you will never get a measurement for. 

The "myth" is believing we have math equivalent to the human auditory system. Not even close. 

I am prolly quite late to this 'party' but I wanna comment on this thing. It is simply just another one of those "there is voodoo in the audio and only us the initiates can hear it"
No there isnt. Not since about 1890 when things like waves and electromagnetism were fully explained by solid math and physics. The same math and physics that are behind any audio device ever built. If there was anything missing or wrong with those, it will be missing or wrong in your playback device too. And you wont hear it even if that audio-vodoo existed. So no, there is no audible voodoo anywhere. And btw, that's called logic. Much better than ears .

Prolly John Dunlavy put it even better when he said 20+ years ago that if anyone discovers and unmeasured effect in audio, he'll get a nobel prize. Didnt happen afaik.
The only myth here is that you have golden ears. Is our math perfect? Nope, not even close. Just better than ears. Mine, yours, anyones.
And it's posts like this that actually keep me from buying 24bit music. Normally I don mind buying and using 24bit devices and music. Just dont put that 'it sounds better' bull into it. Noone ever heard that. Anywhere. And many people did test. And test. And nothing.

 

[prot]

And many thanks to user @gregorio for starting this controversial thread and for a very informative first post. As controversial as they are, those things need to be said and discussed and tested. Preferably without preconceptions and without golden ears.

 

[Greenears]

There's nothing special about 16bit; it just happens to be the bit-depth that gets the necessary dynamic range for the broad range of music out there. Sometimes for fun (just to show you how weird some of us are), I find the minimal PCM bit-depth and sample rate at which I cannot pass an ABX test of a track vs. the original Redbook version. I have plenty of classical tracks that work fine at 14bit, lots of non-classical tracks that can go down to 12bits, and the occasional brick-wall special that is indistinguishable at 8bits. Since I'm older than some of the folks on here (not bigshot obviously ), I can usually also down-sample to 36kHz or so without affecting quality either.

You can see then, perhaps, why some of us get miffed at the suggestion that people need MOAR bits and samples. If 12/36 in indistinguishable to me from 16/44.1, why on earth would I want 24/192? And if someone else says their ears are much more super than mine, then I am naturally skeptical if they haven't done the same type of experimentation I have with their own hearing. People are all happy to put in the work of buying gear and hi-res material and writing subjective reviews, but they scoff at those of us who go a step further and put in the work of *testing* these things against the limits of our own organism.

What tool do you use to make 14 bit files and what volume level did you listen to when passing the abx? I'm interested to try this test.

 

[RRod]

What tool do you use to make 14 bit files and what volume level did you listen to when passing the abx? I'm interested to try this test.

 
Doing a non-standard bit-depth is a pain (meaning it takes more than one script, so far), since sox (the tool I normally use) only handles the common values (8, 16, 24, 32). I have to go into something like Matlab or R to do the dither + truncation.
 
p.s. If anyone knows a gnu or posix command line way of doing this, that would be great!

 

[limpidglitch]

It's easy to do in Audition, but then Audition isn't strictly free.
 
When you truncate, do you just use a script to remove the last few bits of each word? How do you make sure the file is still playable afterwards?