24 vs 16 bit not audible?
Old 18th January 2009
  #1
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
24 vs 16 bit not audible? Test yourself!

There have been several claims on this forum that 16 bits is mostly 'good enough' because of the limitations of human hearing etc, and even that we in fact can't hear the difference anyway in typical or normal conditions. Decided to do a simple test that can be repeated by anybody interested.

Here's what I did. Took a 24-bit (unprocessed) stereo recording and made a 16-bit version of it using high pass triangular dither. Downloaded the free foobar2000 software, and did a full installation which includes the ABX test. Loaded and marked both of the files, right-clicked for the 'Util' -> 'ABX Two Tracks...' option and took the test.

The sound files can be downloaded here:

http://www.nordicaudiolabs.com/samples/24.wav
http://www.nordicaudiolabs.com/samples/16-hpt.wav

EDIT: Another 16 bit version made with Ditherbox (Airwindows Audio Unit Plugins) uploaded here for easy access:

http://www.nordicaudiolabs.com/samples/16-avd.wav

For playback I used the built in sound card of my small Sony Vaio laptop and a pair of Beyerdynamic DT250 headphones. The phones are the 250 Ohm version so the resulting sound was not loud at all. Spent some time trying to focus on where the difference was. With this primitive setup, it's much harder than with a high resolution studio monitor setup.

Here's the result:







As you can see, I made one mistake out of ten attempts. The big issue to me is, what does the bit reduction actually do to the music? Looking forward to your comments.

Martin
Old 18th January 2009
  #2
Moderator
 
narcoman's Avatar
 

raised noise floor against dB(FS).

Less ability to get into decent headroom ranges - eg AES -18 - without falling foul of noise.

That's the main thing. At AES -18 in 16bit you pretty much get down to 11 and 12bit sound for average forte parts... And shitty sound for quiet passages.

24bit for me EASILY. And hearably!
Old 18th January 2009
  #3
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Yes, good points.

The interesting thing with this specific case is that noise is not an issue when we're listening to 16-bits at normal or low levels. What I heard was something different, noise was way below my hearing threshold.

Martin
Quote
1
Old 18th January 2009
  #4
Lives for gear
 
FeatheredSerpent's Avatar
 

I'm new enough to this game to never have recorded at 16 bit, always 24, interested to hear what differences you heard, Martin?

Also narcoman, can you just clarify for me what you mean when you say AES -18?
Old 18th January 2009
  #5
Lives for gear
 
travisbrown's Avatar
I thought the bigger argument was between 24 and 32 bit.
Old 18th January 2009
  #6
Lives for gear
 
FeatheredSerpent's Avatar
 

Why? You can't hear 32 bit.
Old 18th January 2009
  #7
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by FeatheredSerpent View Post
I'm new enough to this game to never have recorded at 16 bit, always 24, interested to hear what differences you heard, Martin?
Sure, I'd be happy to talk about it, but let's hear from some other people first. Does anybody (not) hear a difference between the two files? Are you brave enough to take the test and share the result?

Martin
Old 18th January 2009
  #8
Lives for gear
 
travisbrown's Avatar
Quote:
Originally Posted by FeatheredSerpent View Post
Why? You can't hear 32 bit.
That's the argument. Some claim they can. Same with 96/192.
Old 18th January 2009
  #9
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by travisbrown View Post
I thought the bigger argument was between 24 and 32 bit.
My whole CD collection is 16 bits and there are no 32-bit converters really. There is research claiming that 16 bits is practically speaking transparent.

Martin
Old 18th January 2009
  #10
Lives for gear
 
The Beatsmith's Avatar
 

i think he probably means that recording @ 32 is the real 'pointless issue' - recording and mixing at 16 vs 24 has obvious and audible differences
Old 18th January 2009
  #11
Lives for gear
 
taturana's Avatar
there's an obvious detail difference..24 being the obvious better choice.


but there's a difference in a 32bit mix file...not in tracking in 32 bit.... you can bring a 32 bit file un-dithered to mastering... so if you are making cd's or 24 bit material... you can start from there down.

that by itself is a big advantage. believe it or not. hear it or not. pure math.

especially if you are working ITB.
Old 18th January 2009
  #12
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by The Beatsmith View Post
recording and mixing at 16 vs 24 has obvious and audible differences
I agree, but what about the end product, do you think 16 bits is good enough for the consumer? Let's say for anybody who listens to music?

Martin
Old 18th January 2009
  #13
I can't do the true ABX test, as I'm on Mac, and have yet to find ABX software that works (yes I've tried the Java app, doesn't recognize my Fireface). But I did put the two tracks into Cubase and did a Solo/Mute switch, with my eyes closed. I was getting it right about 90%. I also did a null test, and when phased reversed, my Fireface Mixer told me that the noise floor was around -120dB. This indicated that there was a discrepancy between the two files. I'd like to do a truer ABX, because I could just be guessing.

What seemed to stick out to me most was loss of reverb detail. As the reverb tail trailed off, it was less audible in the 16bit version. Same with low level details, less audible in 16-bit. This must be attributed to a raised noise floor.

I was monitoring on Beyerdynamic DT 770pro 80ohm heaphones, through a Fireface 800. While it wasn't something that stood out, it was audible upon closer inspection. There's no way somebody should be bothered by this, in the real world. Maybe on a 10,000 dollar playback system, but casual listening on modest home electronics you'd be hard pressed to really hear these things. Like the OP I'm still not really sure what is sounding so different, must be loss of high frequency hiss or something.

If anybody knows of a good ABX for Mac let me know.
Old 18th January 2009
  #14
Lives for gear
 
FeatheredSerpent's Avatar
 

Hang on. Martin, what was the dynamic range of the material you used for the file? I haven't listened to them yet btw.
I'm just thinking that for this test you would have to record an identical source at first 24 bit, then at 16 bit, no?

If you just convert a 24 bit file to a 16 bit file, it's removing the mantissa but you've still taken advatage of the top tiers of resolution at 24 bit, so it's not 'true' 16 bit anymore is it?

That would account for people only hearing differences in the quieter sections - reverb tails fading out etc.
Old 18th January 2009
  #15
When I nulled the two tracks in Cubase, at first I thought they completely nulled. Because I didn't hear anything. But I turned the tracks up to maximum and the master fader up all the way. Then I turned up my fireface mixer all the way. I could hear basically white noise. I thought that I heard a tiny bit of the music underneath all of the noise, but mostly it was just noise.

This might explain the 24bit to 16bit conversion. Which would add the dither noise. I don't fully understand dithering and bit depth conversion. But is this correct?
Old 18th January 2009
  #16
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by Sword in Hand View Post
What seemed to stick out to me most was loss of reverb detail. As the reverb tail trailed off, it was less audible in the 16bit version. Same with low level details, less audible in 16-bit. This must be attributed to a raised noise floor.
OK, that sounds familiar considering my usual experiences. With this laptop setup I can barely hear the reverb, so I had to find other changes to focus on. This was the first time I tried ABX testing, it was a very healthy experience. My first attempts failed miserably...

Quote:
Originally Posted by Sword in Hand View Post
There's no way somebody should be bothered by this, in the real world. Maybe on a 10,000 dollar playback system, but casual listening on modest home electronics you'd be hard pressed to really hear these things.
Agreed, except I just proved (to myself) that it makes a difference even on a lousy laptop. But I'm thinking:

- we can hear very subtle things, even some that "scientific research" can't seem to explain/accept for some reason. What's all that about?

- 16 bits is not transparent, it affects the musical experience. To what degree we can't say, but it's not a good thing. I'm also getting a feeling that people don't enjoy listening to music like they used to, maybe it doesn't taste as good as it could? (or did?)

Martin
Old 18th January 2009
  #17
elz
Gear nut
 
elz's Avatar
 

get a 16" x 16" box and a 24" x 24" box.
see how much stuff you can put into each.
Old 18th January 2009
  #18
Martin,

While you were listening through a low end sound card, your headphones are not very low end. At around $200, these are hardly what I'd call low end (except in gearslut land . Sure you can spend over a grand on headphones, but your headphones should be pretty revealing, even on a low end converter. Plus, converters and headphone amps are a much smaller factor in determining the quality of the sound. The transducer quality affects the sound much more.

I agree that 24-bit is extracting enough extra quality as to be audible. I just don't think it matters so much in the REAL WORLD. Like listening in the car, on an iPod with apple earbuds, or eek through a cellphone speaker (people do this!), you just aren't going to hear, or even care about that minute difference in quality. I certainly don't. I have to be fairly focused to hear these kinds of differences, on what consumers would call a high end listening environment. Audiophile playback systems probably reveal these differences really well. But that kind of investment is working on diminishing returns.

Keep in mind I'm talking about 24-bit as the final mixdown format. I'm not suggesting that recording and mixing in 24-bit doesn't have it's advantages. I record all of my projects in 24-bit and mix in 32-bit float. I'm just struggling with the true advantages of 24-bit, as a consumer medium.
Old 18th January 2009
  #19
Quote:
Originally Posted by Martin Kantola View Post
Here's what I did. Took a 24-bit (unprocessed) stereo recording and made a 16-bit version of it using high pass triangular dither.
I'm with FeatheredSerpent. I think you would have to record something simultaneously at both 16 and 24 bit, not make a 24 bit file 16 bit by some process.

Otherwise isn't this just a test of the process you used to make the 16 bit version?
Old 18th January 2009
  #20
Here's an analogy that, I think, parallels audio bit depth.

As a professional photographer, I work a lot with photographs, in photoshop. When I take photos in my DSLR, I have the option of shooting in Jpeg (16-bit) or RAW (24-bit). I always shoot RAW, because it allows me greater processing flexibility (much like 24-bit's higher headroom and lower noise floor). But, if I were to shoot a photo in both formats, look at them unaltered in Photoshop, I know I couldn't tell the difference (like a dithered 24-bit file and a 16-bit file). Same goes for 48-bit color vs. 24-bit color. You can't see the difference right out of the camera, but as soon as you start editing you have more freedom to push things or alter things. The end result is more natural looking.

The connections is how 24-bit audio (kind of like a RAW 48-bit photo) allows you more freedom in mixing. You can push things further, because you have more headroom and a lower noise floor. But once it all goes down to 16-bit audio the quality is preserved, through dithering, and you're left with a very high quality file.

Bottom line, the difference is pretty small to me. I'm happy listening on CDs and have yet to hear a 24-bit mix that blew a CD away, on standard playback systems.
Old 18th January 2009
  #21
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by Sword in Hand View Post
While you were listening through a low end sound card, your headphones are not very low end. At around $200, these are hardly what I'd call low end (except in gearslut land
No, they are not low end, but the laptop sure has a hard time driving them. Again, the original point was that we shouldn't be able to hear ANY difference on ANY setup... according to some claims.

I'm a sound engineer, and agree that these differences are tiny from a consumer perspective. If we are talking sound quality in traditional terms. Most people don't care, they happily enjoy medium bit rate mp3's. So do I.

But I'm more worried about long-term psychoacoustic effects, how are we going to be affected if the industry has now decided that the consumer formats we have are good enough. This is the first time ever progress has almost come to a complete halt (or reached the end of the line?) and the overall sound quality we are exposed to is rapidly declining anyway. Laptops, mp3-players and cellphones are the new standard. And what is all of this doing to our music? Nothing?

Martin
Old 18th January 2009
  #22
Lives for gear
 
travisbrown's Avatar
Quote:
Originally Posted by elz View Post
get a 16" x 16" box and a 24" x 24" box.
see how much stuff you can put into each.
Almost.

16" * n compared to 24" * n

Your analogy has 24bit as 2.25 times the capacity of 16bit when in fact it is 1.5 times. I know we are sort of mixing metaphors here.

One dimension remains constant. We are talking about depth, not width and depth. I suppose sample rate would be the width in this context.
Old 18th January 2009
  #23
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by pvoc3000 View Post
I'm with FeatheredSerpent. I think you would have to record something simultaneously at both 16 and 24 bit, not make a 24 bit file 16 bit by some process.
Think about it, how exactly would I record something in 16 bit? Do I need to find an old converter that's only capable of 16 bits resolution, or is it enough to only use 16 bits of whatever a new converter produces?

Martin
Old 18th January 2009
  #24
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by Sword in Hand View Post
The connections is how 24-bit audio (kind of like a RAW 48-bit photo) allows you more freedom in mixing. You can push things further, because you have more headroom and a lower noise floor.
Yes, that's how I see it too!

Quote:
Originally Posted by Sword in Hand View Post
But once it all goes down to 16-bit audio the quality is preserved, through dithering, and you're left with a very high quality file.
Not exactly, and that's what my little experiment hopefully shows, the full quality is not preserved intact. Of course, there are many different types of dither and noise shaping. But there seems to be more to it than just an increased noise floor we might or might not hear, since we can hear it in the musical performance somehow. Which I find interesting.

Martin
Old 18th January 2009
  #25
Lives for gear
 

Great initiative Martin. I'll do the test later today, must get something to eat first.


/Peter
Old 18th January 2009
  #26
Moderator
 
narcoman's Avatar
 

Quote:
Originally Posted by FeatheredSerpent View Post
I'm new enough to this game to never have recorded at 16 bit, always 24, interested to hear what differences you heard, Martin?

Also narcoman, can you just clarify for me what you mean when you say AES -18?

AES -18.... a standard by which 0dB(VU) is calibrated to be -18dB(FS).... gives you 18dB of "headroom".....Or in other words - you set up a forte music passage to sit at -18dB(FS) for ITB work.....
Old 18th January 2009
  #27
Lives for gear
 
Martin Kantola's Avatar
 

Thread Starter
Quote:
Originally Posted by Audiop View Post
Great initiative Martin. I'll do the test later today, must get something to eat first.
Cool, thanks Peter. Anybody else up for the challenge? We all know there's a clear difference, or do we...?

Martin
Old 18th January 2009
  #28
Gear Guru
 
Ethan Winer's Avatar
 

Quote:
Originally Posted by Martin Kantola View Post
Took a 24-bit (unprocessed) stereo recording and made a 16-bit version of it using high pass triangular dither.
I agree this is valid. There's no need to record the same source at 24 bits and 16 bits. If you got it 9 times out of 10 that's excellent!

One thing I noticed is that the soft part of the music peaks around -30 dB. So that means you're really comparing 11 bits to 19 bits. But that doesn't disqualify your test because level differences like this occur normally with orchestral music.

Good show Martin!

--Ethan
Quote
1
Old 18th January 2009
  #29
Lives for gear
 
FeatheredSerpent's Avatar
 

@narco - Ah, I guessed as much I've just not seen it written as 'AES -18' before. Thanks.

Quote:
Originally Posted by Martin Kantola View Post
Think about it, how exactly would I record something in 16 bit? Do I need to find an old converter that's only capable of 16 bits resolution, or is it enough to only use 16 bits of whatever a new converter produces?

Martin
The reason I think doing it the way you have done is slightly flawed, is that if you recorded at native 16 bit, the higher tiers of resolution wouldn't be available to you in the first place (I should just say that my approach here is theoretical).

So when you record at 16 bit, the highest tier (the top 6dB) gives you 32,896 steps of resolution (total of positive and minus changes available). Plus the steps from all the lower bits.
You can't 'see' any more steps than that.

But at 24 bit, the highest tier will give you 8,421,376 total steps. From -7 to -12 there are 4,210,688 steps and so on.
By recording at 24 bit first, then converting to 16 bit, you've outstripped what is possible at only 16 bit, because you're just deleting the bottom 8 bits, the signal is still residing in the higher tiers, and so is not the same as if you had recorded at 'native' 16 bit in the first place.
It's even more relevant (I think) if your recorded signal has a dynamic range of less than 88dB, because then there would be nothing in those bottom 8 bits anyway, all that would happen is that the noise floor got closer, the actual integrity/resolution of the signal would still be the same.


Don't know if I explained myself very clearly there but hopefully you'll catch my drift! (and set me straight if this is not correct).
Old 18th January 2009
  #30
I don't think recording at 16-bit on a 24-bit device is going to work. Correct me if I'm wrong, but wouldn't that be the same as converting a 24-bit file to 16-bit, by just chopping off the last 8-bits. 24-bit device are natively so, they work best at there intended word depth. And when the file is played back, it's spit out of the DA in that native word depth too. At least this is what I understand to be true.

So essentially for this test to be really valid, wouldn't you have to have two converters, one 16 and one 24, then, record the audio into them at 16 and 24-bit respectively, and finally, listen back to each, on their respective platforms? So no conversion is happening whatsoever.
New Reply Submit Thread to Facebook Facebook  Submit Thread to Twitter Twitter  Submit Thread to LinkedIn LinkedIn  Submit Thread to Google+ Google+ 
 
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Similar Threads
Thread
Thread Starter / Forum
Replies
hollywood_steve / Remote Possibilities in Acoustic Music & Location Recording
17
ocelot12 / Electronic Music Instruments & Electronic Music Production
3
Adebar / High end
76

Forum Jump
 
Register FAQ Search Today's Posts Mark Forums Read

SEO by vBSEO ©2011, Crawlability, Inc.