24bit vs 16bit, the myth exploded!

[priest]

And I thought the Headphones forum made me feel stupid. Anyone know any good particle physics or cosmology forums?

 

[Sandman65]

Quote:

I like to mail letters in refrigerator boxes just to make sure they don't get any creases in the mail.

 
 
Thanks for Vernor's through my nose!
 
Once again another highly informative thread illustrates how evil marketing types are trying to get you to reinvest in your music library.
 
 
listening to Rued Langgaard's "Music of the Spheres"

 

[spyrusthevirus]

you know 50+ pages is a bit much for tonight, so I wont read it all.
but here is what I am thinking.
Let's suppose we want to sample a sine wave at 15khz. no problem, we can hear tha and mics can hear it too. so we will use 44.1khz sample rate.
a 15khz sine is repeated 15k times in a second and we will take 44.1k samples in one second. which means that we will have 2.94 samples per period. lets make that 3.
now if you have seen a sine wave, how is that remotely accurate?
if we try that at 8khz, with the thought that no instrument makes a sound with a main frequency above that, it leaves us with 5.5 samples per period. It does not seem that accurate either.
Just because the rule of thumb says that you use double the sampling rate of the maximum frequency you want to record, it does not mean the 96k and 192k are completely useless for say 15-20khz. if my basics in this are correct, you do capture more detail with a higher sample rate, the ideal being an infinite one.

 

[xnor]

Quote:

but here is what I am thinking.
Let's suppose we want to sample a sine wave at 15khz. no problem, we can hear tha and mics can hear it too. so we will use 44.1khz sample rate.
a 15khz sine is repeated 15k times in a second and we will take 44.1k samples in one second. which means that we will have 2.94 samples per period. lets make that 3.
now if you have seen a sine wave, how is that remotely accurate?
if we try that at 8khz, with the thought that no instrument makes a sound with a main frequency above that, it leaves us with 5.5 samples per period. It does not seem that accurate either.
Just because the rule of thumb says that you use double the sampling rate of the maximum frequency you want to record, it does not mean the 96k and 192k are completely useless for say 15-20khz. if my basics in this are correct, you do capture more detail with a higher sample rate, the ideal being an infinite one.

The sine can - in theory - be perfectly reconstructed as long as there are > 2 samples. Since the lowpass filter in a DAC needs some room to work we cannot go up to 22.05 kHz, but somewhere around up to 21 kHz.
Thinking in straight lines between sample points is a common fallacy, that's not how it works.
 
No, your basics are wrong and a higher sample rate will not capture more detail in a given range of frequencies (say 0 to 21 kHz).

 

[spyrusthevirus]

Quote:

No, your basics are wrong and a higher sample rate will not capture more detail in a given range of frequencies (say 0 to 21 kHz).

why does that happen? because a lower sample rate can already catch enough to reproduce those frequencies that interest us? of because of the way digitalising audio works?

 

[xnor]

Because > 2 samples are enough due to the sampling theorem.
 
Clean 21 kHz sine wave (red dots are the samples) sampled at 44.1 kHz will be reconstructed to the blue line:

 
This is NOT how it works:

 
 
 
edit: better images

 

[ctoth666]

This thread inspired me to perform some very impromptu sonic experiments to determine whether or not I could tell a difference between bit depths. I have previously determined that I can hear the difference between 192khz and 44.1khz sample rates. Now were these blind experiments? No they were not, but that does not change the fact there IS an audible difference. My equipment includes a MacBook Pro 2011, an Echo AudioFire2 firewire interface, and modded D2000 cans. Also 21 year old ears.
 
Here's what very unscientifically went down:
 
-I set the playback sample rate on my interface to 96khz.
-I recorded single C4 oscillating tone with Sylenth VST.
-I exported three separate normalized .wav files at 16, 24, and 32 bit. All were sampled at 192khz.
-I played back each of these recording in iTunes.
 
I did the same experiment with the sample rate as the independent variable, and absolutely there is a clear difference at higher sample rates. I have played around with the playback sample rate on my interface with various file formats, and can discern a difference between 44.1khz and 96khz, and .wav and mp3, etc. The difference is not night and day, but still apparent on my equipment especially when I use fx processing in Ableton. Now, as for the different bit rates, I again can pick up subtle differences between 16 and 24 AND 24 and 32 bit. Yes, 32 bit does sound different on my equipment, and I stress "different" because it's not necessarily better and I don't what to attribute those differences to. Is it enough of a difference that I care? No it's not, except that now I know I can hear a difference so I'll have to get over that fact somehwere down the line.

 

[bigshot]

Try AAC 256 VBR

 

[xnor]

ctoth666, how did you export the files, with or without dither? Some sort of noise shaping?
 
Why did you set your interface to 96 kHz but generated 192 kHz files? Why didn't you do an ABX test?
 
What interface are you using and what audio API (ASIO, WASAPI, DirectSound ..)?
 
What effects are you talking about? They might cause aliasing at lower sample rates, but applying the effects at higher sample rates/bit depth and doing a final conversion to 44.1/16 shouldn't sound any different. (The effect should do this upsampling automatically, else I'd say it's broken.)
 
Why didn't you use a piece of music for the test?

 

[stv014]

Quote:

Now were these blind experiments? No they were not,

 
That alone prevents the results from being particularly useful. However, if you uploaded the files somewhere, and posted a link, it could be found out if there are any differences that are not the result of different sample rates and/or resolutions.

 

[ctoth666]

Quote:

ctoth666, how did you export the files, with or without dither? Some sort of noise shaping?
 
Why did you set your interface to 96 kHz but generated 192 kHz files? Why didn't you do an ABX test?
 
What interface are you using and what audio API (ASIO, WASAPI, DirectSound ..)?
 
What effects are you talking about? They might cause aliasing at lower sample rates, but applying the effects at higher sample rates/bit depth and doing a final conversion to 44.1/16 shouldn't sound any different. (The effect should do this upsampling automatically, else I'd say it's broken.)
 
Why didn't you use a piece of music for the test?

My knowledge of these things is very limited, but the only variable that changed was the bit rate. I exported the files all directly from Ableton Live with Triangular dither and did NOT do a final conversion to 44.1/16 if that's what you're asking. I set my interface to 96khz sampling but exported 192khz from Ableton because I figured it would contain the most data, but again all three files were 192khz. I do not know what an ABX is nor do I know how to perform one. The API was Core Audio and the audio interface was an Echo Audiofire2. The effects that I noted were very minor but not aliasing I don't think. Since my technical knowledge is so limited, I'll try to explain as best I can.
 
-From 16 bit to 24 bit, the sound is brighter and there is more "oscillation" in the synth sound, like the tone has a different rhythm. On my headphones, higher frequencies are more discernible.
-From 24 bit to 32 bit, there is one particular frequency sin wave that just sounds different if not more pronounced. When playing back the sample, I can hone in and "track" it with my ears. There is also a higher frequency oscillating tone that stands out more.
 
That's about as well as I can describe the differences. And I didn't use a recording because I don't really have any recording equipment, and I didn't use any pre-sampled or pre-recorded music because I figured that it would defeat the purpose. 

 

[bigshot]

From 16 bit to 24 bit, the sound is brighter and there is more "oscillation" in the synth sound, like the tone has a different rhythm.

I've run into that with weird down sampling rate mismatches. I'd bet your software has a setting that would do it correctly. Play around with it and see.

 

[xnor]

Afaik, CoreAudio resamples just like DirectSound to whatever is configured. ctoth666, if you want you can send me your test signal(s) in the highest possible format (192/32) and I'll convert them all to 96 kHz but one file will only contain frequencies up to 22.05 kHz (44.1 kHz sample rate), another one up to 24 kHz (48 kHz sample rate) and another one up to 48 kHz (96 kHz sample rate). If you don't have a spectrum analyzer enabled you can then try to pick out which is which "blind".

 

[ctoth666]

Quote:

Afaik, CoreAudio resamples just like DirectSound to whatever is configured. ctoth666, if you want you can send me your test signal(s) in the highest possible format (192/32) and I'll convert them all to 96 kHz but one file will only contain frequencies up to 22.05 kHz (44.1 kHz sample rate), another one up to 24 kHz (48 kHz sample rate) and another one up to 48 kHz (96 kHz sample rate). If you don't have a spectrum analyzer enabled you can then try to pick out which is which "blind".

 
Well it was a very straightforward test. If you have the free Synth1 plugin then you can reproduce the tone that I exported from Ableton, but I suppose any plugin would do. I would like to learn something about audio sampling and bit rates from this. I don't quite understand: if I send you the test signal at 192/32, is your plan to downsample them and then upsample them to 96 kHz? I played back all of the recordings @ 96 khz through my headphones, but the actually files were all exported at the different bit rates and thus were different sizes. When I played them back, they were all resampled to 96 kHz, correct? All that I have demonstrated to myself is simply this: that there are sonic differences in the files that I exported. Again, I don't necessarily know why, but there are differences. I'm not saying that there is definitely an audible difference between 16 bit and 24 bit audio, but rather that I can hear a difference between the files that I exported.

 

[xnor]

Quote:

Originally Posted by ctoth666 /img/forum/go_quote.gif
 
Well it was a very straightforward test. If you have the free Synth1 plugin then you can reproduce the tone that I exported from Ableton, but I suppose any plugin would do. I would like to learn something about audio sampling and bit rates from this. I don't quite understand: if I send you the test signal at 192/32, is your plan to downsample them and then upsample them to 96 kHz? I played back all of the recordings @ 96 khz through my headphones, but the actually files were all exported at the different bit rates and thus were different sizes. When I played them back, they were all resampled to 96 kHz, correct?

Yes, I'd make sure the files look the same in terms of bitrate but actually only contain 44.1/16 content. That way you could play all the files without resampling from CoreAudio and wouldn't know which is which.
 
 
Quote:

All that I have demonstrated to myself is simply this: that there are sonic differences in the files that I exported. Again, I don't necessarily know why, but there are differences. I'm not saying that there is definitely an audible difference between 16 bit and 24 bit audio, but rather that I can hear a difference between the files that I exported.

I would be more careful with the conclusion. Do a proper ABX test (there should be apps for this available on OS X) and post the log here. Also, generate 96 kHz files if you set CoreAudio to 96 kHz. You have to eliminate all the extra variables...
 
Btw, you also said you can hear a difference between 44.1 and 192 kHz. What was the test file for that? I'd be very surprised if you can hear above 21 or even 20 kHz.