Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: 24bit recording listening test (Read 6335 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

24bit recording listening test

Ethan Winer is running a listening test about 24bit recording and dithered/truncated output.
http://www.ethanwiner.com/BitsTest.html

"Everyone knows that 24-bit recording is better than 16, right? Even though a 24-bit recording will be reduced to 16 bits for distribution on CDs, conventional wisdom says that recording at 24 bits is always better. Especially if you apply dither during conversion to 16 bits instead of just discarding the lower 8 bits by truncating them. But is recording at 24 bits really worth the added overhead of half again more disk space and CPU power? Can anyone really hear a difference? More important, can you hear any difference?  .....

  Again, one file was dithered from 24 bits to 16, another was truncated, and three others were reduced to (approximately) 13 bits, 11 bits, and 9 bits respectively."

Heated discussion @ Homerecording.com:
http://homerecording.com/bbs/showthread.php?threadid=46618
Juha Laaksonheimo

24bit recording listening test

Reply #1 –
I submitted my results to Ethan.  An interesting test.

ff123

24bit recording listening test

Reply #2 –
Yes, interesting, but i fear its missing the point. Everyone here would agree that 16 Bit resolution are fine and be able to copy with most stereo equipment being used by 'normal' users.

But this is not the point in 24 Bit recording. The main reason to use 24 Bit equipment is to make sure you really have 16 Bit resolution on your final CD. For a normal recording you need a dynamic headroom of at least 4 - 6 dB for every channel, so one bit is definitely lost ( gives 15 ). If you find one of the channels must be amplified by +6 dB compared to the others ( o.k., thats a lot ) in your mixer and you were using a lower quality A/D for this one with a SNR being smaller than 96 dB you will end up with 13 - 14 Bit Dynamics in the end ..... and i strongly believe this can be heard. Maybe not even with noise, but for sure you are running into other problems if your LSB is undefined, i often recognize harsh overtones and metallic sounding heights on elder recordings ( with significantly lower Dynamics than modern 20 or 24 Bit recordings ), and this is certainly not only caused by the A/D they had this time.

Another thing that is not at all coverd by Ethans test is sound processing. Why are we using SSRC in 24 or 32 bits floating point precision ? For sure not because we can hear 24 Bits dynamics, but to make sure rounding errors etc. are inaudible.

I personally believe we could live well with 16 Bits, but again the question is ( must be ) why not using 24 Bits for a high quality record ? SACD and DVD-Audio will both offer more than enough space, so why limit ourselves to 16 Bit ? We dont know what future will bring regarding our ears ( direct brain connection  ), so our beloved music should be worth it  .

Another thing is sampling rate. I was one of the first almost 15 years ago to stand up claiming the 44.1 KHz are not enough. We had created specs of a S-CD and X-CD standard with 20 Bit / 96 KHz resp. 24 Bit / 192 KHz this time .... long before DVD-Audio and SACD were discussed.

Unfortunately there were no storage media available offering the capacity and speed we needed ... also the only existing A/Ds ( for our own recodings ) at that time being precise and fast enough were used by SIEMENS for 'Kernspintomographie' ( dont know English word ) and not affordable ...

24bit recording listening test

Reply #3 –
And Creative created the Audigy which does 16bit recording and 24bit playback.

24bit recording listening test

Reply #4 –
Quote
Originally posted by tangent
And Creative created the Audigy which does 16bit recording and 24bit playback.


I wouldn't be too sure about that:

http://www.tomshardware.com/video/02q1/020...erratec-04.html

Quote
But be warned, the Audigy's ability to support 24-bit/ 96-kHz sound is very relative. Unlike the DMX 6Fire 24/96, the Audigy Platinum is not a "true" 24/96 and cannot play or record a file of this quality, nor even work on it. Actually, the only task it does in 96 kHz is a linkup with another device via the S/PDIF input. The card and rack components are all supposed to be 24-bit/ 96-kHz-compatible, but, in practice, the card is limited to 16 bits/ 48 kHz. Analog recording and restitution is possible in 24 bits/ 48 kHz, but, in fact, the processor downsamples in 16 bits and then upsamples. The card also has to be used with the sample rate conversion (SRC) software provided in order to work in 44.1 kHz. It's just a pity that Creative has not given this product an automatic SRC hardware option by using, say, a Cirrus Logic CS8420 chip. And it's really a pity that the card is not completely 24-bit/ 96-kHz-compatible. 


http://forums.winamp.com/showthread.php?s=&threadid=76742

24bit recording listening test

Reply #5 –
You have to love the Creative marketing engine:mad:

Download the WAV files now for the test.

24bit recording listening test

Reply #6 –
Yeah! I scored 100% correct. Is that statistically correct - what was the chance I was guessing? My feeling is that Ethan should have used a file randomiser for this test - it might have nade it harder for me.

----------------------
My reply to the test:

File 1 - Very clean throughout, even at tail of wav - no hiss
File 2 - Less clean than File 1 at tail of wav - ever so slight hiss
File 3 -
File 4 - Guitar string buzz on frets more grainy than File 1
File 5 - Lots of hiss

My guess - there in the order of highest to lowest quality - you haven't mixed them up at all.
Ruse
____________________________
Don't let the uncertainty turn you around,
Go out and make a joyful sound.

24bit recording listening test

Reply #7 –
Quote
Yeah! I scored 100% correct. Is that statistically correct - what was the chance I was guessing? My feeling is that Ethan should have used a file randomiser for this test - it might have nade it harder for me.


Cool!  But I would have expected file 3 (13 bit file) to have more noticeable hiss than 2.

When I did my mad challenge test (http://ff123.net/madchallenge.html), at 15 bits, I noticed the hiss from the non-noise-shaped dither, but no hiss from the truncated version.  So my expectation for Ethan's test would have been that file 1 should have more had more hiss than file 2, since he used SAWPro's type II dither (non-shaped, triangular dither).  This type of dither does not push noise out of the ear's sensitive frequency regions.

I heard no difference in files 1, 2, or 3, for this sample.  However, I did identify file 4 and file 5 as the 11-bit and 9-bit files, respectively.

I emailed Ethan about making the raw comments available so that I can perform an analysis on it.  Too bad he didn't make the ordering less obvious, though, as you point out.  I wonder how many people guessed like you (Ruse) that Ethan didn't mix things up at all?

ff123

24bit recording listening test

Reply #8 –
Quote
I wonder how many people guessed like you (Ruse) that Ethan didn't mix things up at all?
>Geoff,

> My guess - there in the order of highest to lowest quality -
you haven't mixed them up at all. <

Bingo!  You are the only person to get them all correct.

--Ethan<
Ruse
____________________________
Don't let the uncertainty turn you around,
Go out and make a joyful sound.

24bit recording listening test

Reply #9 –
I pointed this probability trivia out on the HomeRecording forum:

If one can identify the 11 and 9 bit files without trouble, then there is a 1 in 6 chance that he can just guess randomly and get all 6 correct and a 2 in 3 chance of getting at least 3 correct.  So given the number of participants (at least 16) I suspect that many people either didn't hear the differences at 11 and 9 bits, or (like me) didn't hazard guesses for all the files.

ff123

24bit recording listening test

Reply #10 –
Well, I identified the last two files (like ff123) and then guessed he might have just put them in order. But even with that knowledge, I can't hear a difference! Certainly 1 and 2 sound identical to me - I'll take them home on CD and listen on a different system, but I doubt it will help!

Cheers,
David.
http://www.David.Robinson.org/

24bit recording listening test

Reply #11 –
Quote
Originally posted by ff123
I pointed this probability trivia out on the HomeRecording forum:

If one can identify the 11 and 9 bit files without trouble, then there is a 1 in 6 chance that he can just guess randomly and get all 6 correct and a 2 in 3 chance of getting at least 3 correct.  So given the number of participants (at least 16) I suspect that many people either didn't hear the differences at 11 and 9 bits, or (like me) didn't hazard guesses for all the files.

ff123


mppdec16.exe supports some advanced technologies called
noise shaping. Using them makes it difficult to distinguish
9 bit from full 16 bit resolution when listing to typical pop music.
10 bit is not possible to distinguish. For some high dynamic
classic music this is between 13 and 14 bit.
These tests are made with 44.1 kHz.

When using fs=96 kHz so can noiseshape much better than
at 44.1 kHz, so for those classic music 10 bit is enough.

So at 96 kHz 16 bit for a transport media is overkill.
Don't ask for 96 kHz/24 bit or 192 kHz/16 bit.
--  Frank Klemm

24bit recording listening test

Reply #12 –
Quote
some advanced technologies called noise shaping

long live psychoacoustics!!!  so when are we going to FULL transparency at 7-8 bits/sample ? (human hearing entropy is about 1.6bits per sample using ideal system - I believe mr. Ivan Dimkovic once written here?)

24bit recording listening test

Reply #13 –
Quote
So at 96 kHz 16 bit for a transport media is overkill. 
Don't ask for 96 kHz/24 bit or 192 kHz/16 bit.


Interestingly, the ARA (the group of audiophiles who first campaigned for the DVD-audio specs) suggested 24-bit 48kHz or 16-bit 96kHz - they were well aware of the positive effects of noise shaping.

But the consumer judges quality on numbers - nothing else. 24-bit must be better than 16-bit, so they will buy 24-bit (and play it on their boom box!). For this reason, the industry will give us 24-bit 96kHz.


However, if storage space is not an issue, it seems foolish to take the output of a 24-bit 96kHz ADC, and to convert it to 16-bits, adding lots of (inaudible but possibley problematic) high frequency noise in the process. As Bob Stuart and others have often said "this isn't mp3 - we're trying to store as much as possible, not as little".

http://www.meridian-audio.com/ara/

Cheers,
David.

24bit recording listening test

Reply #14 –
Cheap and average converters sound BAD ! If 24/96 or 24/192 can help them in any way, I'm for it !

Now, about the test, like for everybody, 4 and 5 were obvious.

But it seemed to me, one time on headphones (AKG K-400), one time the day after on speakers (Dynaudio Gemini), that 2 was a little bit sharper and cleaner than 1 or 3.

I abxed File2 from File1 7 times out of 8.
Encouraged by this success, I tried 16 times, but couldn't succeed again (10/16). I was listening quieter the second time, because I was worried about the neighborhood standing the same bit of sound 100 times in a row at full power

24bit recording listening test

Reply #15 –
Quote
Originally posted by 2Bdecided


Interestingly, the ARA suggested
24-bit 48kHz or 16-bit 96kHz -

http://www.meridian-audio.com/ara/

I found this interesting (I was wondering about the thoughts behind those new audio formats like Super Audio CD and DVD Audio)
About the "precision" you mentioned they actually state:
Quote

As a guideline, this would imply a requirement of:

    20-bit precision for channels sampled at 48kHz
    14-bit precision for channels sampled at 96kHz

Because pre-emphasis is not helpful to packed audio channels, and because a 14-bit specification is unlikely to find favour, 16 bits at 96kHz is the minimum practical alternative
Another point I didn't expect is the strong recommendation to use lossless compression (as oposed to no compression).
--
Ge Someone
In theory, there is no difference between theory and practice. In practice there is.

24bit recording listening test

Reply #16 –
Quote
Originally posted by 2Bdecided However, if storage space is not an issue, it seems foolish to take the output of a 24-bit 96kHz ADC, and to convert it to 16-bits, adding lots of (inaudible but possibley problematic) high frequency noise in the process. As Bob Stuart and others have often said "this isn't mp3 - we're trying to store as much as possible, not as little".


I fully agree here ( see my comments above ). Neither DVD-Audio nor SACD are targeted to save sapce ... in fact there is plenty of it on a DVD  !

24bit recording listening test

Reply #17 –
Quote
Originally posted by ChristianHJW


I fully agree here ( see my comments above ). Neither DVD-Audio nor SACD are targeted to save sapce ... in fact there is plenty of it on a DVD  !


SACD is _pure_ noise shaping. Most of the signal
is ultrasonic noise (SNR is -9 dB). It is the
task of the player to make a good trade-off
between bandwidth and problematic noise.

Who worries about noise shaping should never use
a SACD.
--  Frank Klemm

24bit recording listening test

Reply #18 –
(a bit OT -I'm sorry)
Mr Klemm: is there any reliable info about SACD (technical standards, etc.) available on the 'net ?

24bit recording listening test

Reply #19 –
Someone who goes by the nick of Bobro has performed an ABX comparison of files 1 and 2 (the 16 bit files, dithered vs. truncated) and scored 19 of 24 correct (no cherry picking involved!) for a p-value of 0.004.

This is the first really convincing demonstration I've seen of the
audibility of truncation vs. dither at 16 bits (Chris Johnson's test in the rec.audio.pro newsgroup was marred by his having to cherry pick to get significant results).  Bobro described the truncated file as having more pizazz.  He was listening to the first few seconds of the files (not the end!)

See last messages in this thread:

http://homerecording.com/bbs/showthread.ph...4807#post364807

ff123

24bit recording listening test

Reply #20 –
I completed my analysis page.  It's at:

http://ff123.net/24bit/24bitanalysis.html

Two people produced positive ABX results comparing dither vs. truncation at 16 bits.  Both preferred the truncated version.

ff123

24bit recording listening test

Reply #21 –
OT:
"Kernspintomographie" German word, used by ChristianHJW.

The English word is NMR = nuclear magnetic resonace

yeea, there you need high resolution A/D.