Results of the 24-bit challenge

Ethan Winer

Ethan Winer

Acoustics Expert
Folks,

A week ago I invited folks here to compare a series of audio files, to see how audible are the differences between dithering from 24 bits to 16 versus truncating. I also included files that were truncated to (approximately) 13, 11, and 9 bits, which correspond to recording at peak levels of -20, -30, and -40. There are five files altogether, as described on my web page at www.ethanwiner.com/bitstest.html. Here are results:

I received 23 replies altogether, and only one person identified all five files correctly. Not to minimize the one person who got them all right, but since the two worst files were fairly obvious, one person in 23 getting the order of the remaining three files correct likely falls within the boundaries of normal probability. Then again, based on his comments ("#1 was very clean throughout ... the guitar string buzz on #4 was more grainy") shows that perhaps he really did hear the changes in quality!

The short answer is that most folks were not able to distinguish any of the files. This comment is typical: "To tell the truth I couldn't hear any difference." In fact, one person thought the dithered file (the best one) was the worst, and a few picked the worst one (9 bits) as the best! So it seems that grit - whether caused by analog tape or digital artifacts - can make a recording sound "better" even if it is in fact worse in terms of pure accuracy. Note that the totals below do not add up to 23 because most people got some right but missed others.

Number of people who admitted they heard no difference in any of the files: 4

Number of people who thought the best file sounded worst: 1

Number of people who thought the worst file sounded best: 4

Number of people who correctly identified the two worst (11- and 9-bit) files: 9

Everyone who sent me their guesses has been sent the list of files in order of degradation. I am not posting the order here, so that others who want to try the test can still do so in the future. After listening to the files, email me which you think is which and I'll send the results by return email.

Failings of this test: Several people pointed out that the real advantage of 24-bit recording is when an entire project is recorded and mixed at 24 bits, because the increased resolution helps minimize errors that accumulate from gain changes and plug-in processing. At some point I hope to test that too. And though nobody mentioned this, I'll add that another possible advantage of 24-bit recording is the smoother decay of instrument and reverb tails. Neither of these were tested here, so all that this test can claim to show is how important 24 bits and dithering are when used with full-level audio. However, this experiment also shows that recording at an average level of -20 to avoid distorting - another claimed advantage of using 24 bits - is probably not as harmful to audio fidelity as many would think.

Thanks to all who participated!

--Ethan
 
"I'll add that another possible advantage of 24-bit recording is the smoother decay of instrument and reverb tails."

You know, I've heard a lot of people say this too. I haven't had a chance to test plug-ins at 24 bit yet because the band is on hiatis till June and I can't record at 24 bit here at home with the SB Live I have in my box. One of my complaints about computer recording in general is the lack of good effects plug-ins. I seldom use them because they're no match for even the most inexpensive outboard gear. Now everyone claims that going to 24 bit will make those DirectX effects sound much better (but then they also claim that going to 24 bit will making recording in general sound better and I know it does not), but my problem with this claim is that my outboard effects record beautifully at 16 bits and sound head and shoulders above any of my DirectX plug-ins. Now some of my outboard "digital" gear may have greater than 16 bit converters (my favorite reverb unit has 21-bit/44.1 AF Method input and 20-bit/44.1 output) but I'm still recording them into the card at 16/48. You'd think that would degrade the sound to a point that they wouldn't be any better than my 16 bit directX plug-ins but that's clearly not the case. So I question whether or not going to 24 bit will help plug-ins to sound better too. Awaiting a test at that....
 
W,

> One of my complaints about computer recording in general is the lack of good effects plug-ins. <

What are you using? I have no complaints about my UltraFunk plug-ins. Yes, reverb plug-ins often are grainy compared to more expensive outboard units, but that's because the designers have to balance reverb density with CPU loading. But there's no reason a plug-in EQ or compressor or whatever can't be every bit as good as an outboard box.

> Now everyone claims that going to 24 bit will make those DirectX effects sound much better <

"Everyone" claims a lot of things. :)

--Ethan
 
This exercise was indeed informative. You pretty much put the 16 vs 24 bit issue to bed in my mind, although people will argue otherwise for years to come. Your thread will be referenced everytime the issue arises.

thanks again
 
Ethan, I cannot agree that the Ultrafunk plug in sounds great. It has a hard sound and is very brittle i.e. either too little or too much. No "just right".

This could be be a function of the Audio converters as mine are on and old DMAN PCI. Just curious as to your card for these tests.
 
Ethan Winer said:
And though nobody mentioned this, I'll add that another possible advantage of 24-bit recording is the smoother decay of instrument and reverb tails.
Click to expand...
I've given a very clear example of the advantage of higher bit resolution with respect to processing tails in this thread https://homerecording.com/bbs/showthread.php?threadid=46381&pagenumber=2

One may argue this is not significant to the 24bit -> 16bit case since the effect is 48dB lower and may or may not be audible to most listeners. Still, the mechanism is far from mysterious and the fundamental advantage is undeniable.

barefoot
 
Whatever it's worth I've noticed a huge increase in quality with my Rolands' built in 24bit effects over the effects processors I used a few years back. Especially with Reverb tails.
 
Originally posted by Ethan Winer
Folks,



A week ago I invited folks here to compare a series of audio files, to see how audible are the differences between dithering from 24 bits to 16 versus truncating. I also included files that were truncated to (approximately) 13, 11, and 9 bits, which correspond to recording at peak levels of -20, -30, and -40.
Click to expand...

I don't think you should say that the files were "truncated to (approximately) 13, 11, and 9 bits", because when you lowered the volume they were in fact dithered. So what the test really says is that the dithering algorith in the software that you use is in fact so good that not much detail is lost at lower volumes. This can be clearly heard on the lowest resolution file where the dithering produces a significant amount of noise. Had the files actually been truncated the lower resolution files would have been completely noise free.


ps. Pleeeease promise NEVER to tell anyone about my guesses! :D
 
Middleman,

> I cannot agree that the Ultrafunk plug in sounds great. It has a hard sound and is very brittle i.e. either too little or too much. No "just right". <

I'm not there so I can't argue what you're hearing. But maybe the problem is not related to the plug-ins at all, and is a function of the original recording? The only way to know for sure is to create two otherwise identical versions - one EQ'd with the UltraFunk plugs and the other EQ'd some other way - and have someone else give you a blind test.

> This could be be a function of the Audio converters as mine are on and old DMAN PCI. Just curious as to your card for these tests. <

The sound card is out of the picture at that point. I have a Delta 66 and also an Audigy, and the UltraFunk stuff sounds great regardless of which card I play through.

--Ethan
 
Re: Re: Results of the 24-bit challenge

barefoot,

> One may argue this is not significant to the 24bit -> 16bit case since the effect is 48dB lower and may or may not be audible to most listeners. Still, the mechanism is far from mysterious and the fundamental advantage is undeniable. <

Sure, but a theoretical "fundamental advantage" is not the same as a practical advantage. If you have to crank the gain up by 40 dB. to hear it, then who cares?

--Ethan
 
Tex,

> Whatever it's worth I've noticed a huge increase in quality with my Rolands' built in 24bit effects over the effects processors I used a few years back. Especially with Reverb tails. <

I can't imagine how many variables are in play there! :)

--Ethan
 
Re: Re: Results of the 24-bit challenge

BasPer,

> when you lowered the volume they were in fact dithered. <

Why do you think that?

> Had the files actually been truncated the lower resolution files would have been completely noise free. <

That simply is not true at all!

--Ethan
 
Re: Re: Re: Results of the 24-bit challenge

Ethan Winer said:
BasPer,

> when you lowered the volume they were in fact dithered. <

Why do you think that?

> Had the files actually been truncated the lower resolution files would have been completely noise free. <

That simply is not true at all!

--Ethan
Click to expand...
Ethan,

This is actually true. Truncation "noise" (actually distortion) is only associated with signals from the original file which fall at or above the new LSB. Lower level signals are quite since they are set to zero during truncation. When the file is dithered, however, low level noise is continuously present.

barefoot
 
Re: Re: Re: Re: Results of the 24-bit challenge

barefoot said:

This is actually true. Truncation "noise" (actually distortion) is only associated with signals from the original file which fall at or above the new LSB.
Click to expand...

Indeed, the most quiet part of this sample, at 11.65 seconds (the fade-in and fade-out don't count), is 6000 levels peak-to-peak. Well above the truncation noise, that is 100 levels for file 5.
Therefore the truncation noise is well audible (if not masked by a possible dither noise).
 
Err Umm, I don't think that's what I said.

Take a 16bit or 24bit file which has pauses in the music with only very faint background noise, then truncate it down to 11bit or 9bit. The truncated file will be dead quite during the pauses. A dithered file will have hiss during the pauses.

Since Ethan's files don't have any quiet pauses, I think it's difficult to say whether the noise is dithering noise or truncation distortion - especially while listening through my shitty sound card at work.:)

Hope this clarifies,

barefoot
 
Had it been truncation distortion it would have been varying with the signal, and that's not the way I remember it.
The reason I think it is dithered when the volume is lowered, is that I would be very disappointed with the program if it didn't.
 
According to Bob Katz advices (http://www.digido.com), dither should be applied only once, after all processes are done. Therefore it shouldn't be applied each time you apply a process.
The noise shape of dither is computed so as to be inaudible. If you dither twice, it becomes audible.
 
You must log in or register to reply here.
Back
Top