• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Can you tell 8-bit file from 16-bit file in a DBT? Listening test, RLJ

OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
44.1/8-bit byproduct with the 44.1/16-bit original?

Such test with 44.1/8 file is pointless, with tons of audible noise. There is not enough space for noise shaping. Of course I did that.

My point is different - to show that even 8 bit resolution done with psychoacoustically proper dither does not degrade the perceived sound in the way that one would expect. Regardless the single number SINAD of less than 40 dB. I want to show that the single number 20kHz BW SINAD is pointless, from the perception point, without knowledge of signal spectrum.

signal_44_8b_dithered_-2dB.png

SINAD = 37.3 dB

signal_44_16b_to_96-8_shapeddithered.png

SINAD = 35.1 dB


Regarding your opinion of:
the variable of bit-depth
The bit depth is not variable here, the bit depth is 8 bit. Please read the post #36 again. ENOB is variable but only in case that we accept lower measuring bandwidth than 20kHz. For 20kHz BW, ENOB is low again.
 
OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
The concept of dither is so obscure

It is definitely not obscure. As in many fields of human activities, one must know what he is doing and must have some education in the field of work or field of interest. Not everything can be explained to everyone and understood by everyone.
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,488
Such test with 44.1/8 file is pointless, with tons of audible noise. There is not enough space for noise shaping. Of course I did that.

My point is different - to show that even 8 bit resolution done with psychoacoustically proper dither does not degrade the perceived sound in the way that one would expect. Regardless the single number SINAD of less than 40 dB. I want to show that the single number 20kHz BW SINAD is pointless, from the perception point, without knowledge of signal spectrum.

View attachment 224052
SINAD = 37.3 dB

View attachment 224053
SINAD = 35.1 dB


Regarding your opinion of:

The bit depth is not variable here, the bit depth is 8 bit. Please read the post #36 again. ENOB is variable but only in case that we accept lower measuring bandwidth than 20kHz. For 20kHz BW, ENOB is low again.

Oh. my. goodness. Testing with a 44.1k/8-bit file isn't "pointless" - rather, it just doesn't produce the outcome you want. That's an important difference.

"Even 8 bit resolution done with psychoacoustically proper dither does not degrade the perceived sound in the way that one would expect" - IF ONE FIRST UPSAMPLES THE SIGNAL IN ORDER TO GET A LOT OF THE NOISE OUT OF THE AUDIBLE SPECTRUM.

That qualifier is the point - and while it is refreshingly honest to see you say that this is yet another attempt at your dumb crusade against @amirm 's use of SINAD, it is also depressing and frustrating, as it creates confusion rather than enlightenment, per @danadam 's recent comment.

You pull this kind of BS over and over and over again here on these forums. It's so tiresome.
 
OP
pma

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,591
Likes
10,727
Location
Prague
Oh. my. goodness. Testing with a 44.1k/8-bit file isn't "pointless"

I do not want to argue and to debate. If you insist, here is the best 44.1kHz/8bit file I can do (regarding noise audibility), with C3 noise shaping.
It is a true 44.1kHz/8bit file.


And this is the dither used FYI. Contrary to the 96/8 file, there is no space to make effective noise shaping to get good result below 1kHz. However, audibly it is much less noisy than the usual TPD flat dither.

signal_44_8b_C3dithered_-2dB.png


DeltaWave results between original 44.1/16 and 44.1/8 music test sample files

1660478558973.png


1660478598250.png
 
Last edited:

bennetng

Major Contributor
Joined
Nov 15, 2017
Messages
1,634
Likes
1,692
Forgive my hijacking, but this thread seems highly relevant to an experimental audio format I made earlier, here is a snippet of the readme file:
UA-law is an experimental audio format inspired by the u-law and A-law companding algorithms, incorporated with noise shaping and optimized adaptive quantization to improve transparency, without using methods commonly found on typical perceptual codecs like MDCT, filterbank, ADPCM and so on. UA-law accepts .wav and .flac files with 16 and 24 bits and unlimited sample rates, mono to multichannel formats up to 4GB uncompressed PCM file size. UA-law uses 8 bits to encode 16-bit files and 12 bits to encode 24-bit files.
The executable and source code can be downloaded here:

So everyone can use their preferred audio files to perform listening tests and I can't cherry pick samples. I would like to see people posting their non-test signal audio samples and ABX results instead of FFT plots and such. In my own tests I can say the 8-bit version of UA-law may not be 100% transparent with some specific materials, but I can say my method has a much higher chance to achieve transparency than the traditional approach used by pma's version of Audition.
 

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
My point is different - to show that even 8 bit resolution done with psychoacoustically proper dither does not degrade the perceived sound
We already know, that even 1 bit is ok if you have enough bandwidth.

The bit depth is not variable here, the bit depth is 8 bit.
This is not what "to isolate the variable of bit-depth" means. When you claim to compare 8 and 16 bits, then bit-depth should be the only variable that changes. And you changed two variables, bit-depth and sampling rate.

But also, no, the bit-depth is not 8 bit throughout. When you convert 96/8 to 44, then first in this process is a low-pass filter. It creates new samples and those new samples need more than 8-bits, if they are to reproduce 0-22k spectrum of the 96/8 file. Sorry, something has to give. If you remove the noise from the 22k-44k band, then you either have to increase the bit-depth or accept that the noise floor in the 0-22k band will increase.

The only thing that can be said here, is that 44/16 reproduces the 0-22k spectrum of the 96/8 file, but to do so it needs more than 8 bits.

Frankly I don't understand what's the point of this smokescreen with 8 and 16 bit, if all you wanted to show is that 40 dB sinad can sound ok. If you named this thread "Can you tell 40 dB sinad from 90 dB", then no one would bat an eye. Well, at least I wouldn't :).
 

Madlop26

Active Member
Joined
May 2, 2021
Messages
189
Likes
331
Well , for what is worth, very easy test with my HD800s, it seems I can tell clearly the difference from 40 dB sinad from 90 dB, thank you for the demonstration of the importance of Sinad, lol
 

bhobba

Member
Joined
Oct 29, 2021
Messages
20
Likes
6
Um... I don't think that's correct way. With 8 bit at 96k the noise is spread through wider bandwidth. When you convert that to 16 bit at 44.1k, that lower noise floor remains and it is not something, that would fit an actual 8 bit at 44.1k.

For example, here I started with flat dithered silence at 16 bit and did your procedure with 2x, 4x and 8x upsampling, only without noise shaping. The more you upsample in the first step, the lower noise floor you end up in the second one.
View attachment 223850

This inspired me to try the following test. I took the very well-recorded DXD, Carmen Sings The Blues. Converted it to 8 bits with a triangle dither. Dither increased signal-to-noise ratio (SNR) from 48db to 64db.


This reduced the first song's FLAC file size from 280 to approximately 38 MB. Then, I played it using my direct stream DAC. That DAC upsamples everything to 10x DSD, ie for DXD 80 times oversampling. Now when you upsample dithered audio by two, you still have the same SNR overall, but that is spread out over twice the samples. Get rid of half the samples, and you increase SNR by 3db. To figure out the new SNR, since you can't hear past 20k the SNR is 9 db extra or 73 db. Since it is upsampled by 10x DSD, this is 80 times DXD or on continuously halving, we have 40, 20, 10, 5, 2.5. 1.25 or an extra 18db to give 91db SNR. This is nearly the SNR of undithered 16-bit audio, so no wonder the faint hiss was gone. However, to allow a bit of leeway in a real application, I would use 12-bit rather than 8-bit, and it would be 115db which is close to the 120db noise floor of the DAC.

Added Later: After further experimentation, the effect is mostly in the DAC, as expected. Taking some 88.2 content and reducing it to 8 bits with a triangle dither, I can hear the faintest hiss on my computer speakers. But nothing using the Direct Stream.

Thanks
Bill
 
Last edited:

danadam

Addicted to Fun and Learning
Joined
Jan 20, 2017
Messages
956
Likes
1,496
Converted it to 8 bits with a triangle dither. Dither increased signal-to-noise ratio (SNR) from 48db to 64db.
What? I guess you took it from the article but I'm rather sceptical of this article. First they write:
Notice that the lowest amount of noise has increased, so we’ve raised the noise floor in that regard.
which is fine, but then in the next sentence:
However, the highest peak of harmonic distortion has fallen almost entirely to the noise floor, improving our total SNR in the process and resulting in lower overall noise.
That sentence has no sense to me. Even if by "total SNR" and "overall noise" they actually mean THD+N, it's still wrong. Only THD will be lower, anything that includes noise will be larger. Here's what REW shows:
Code:
       no dither    with dither
THD       -62.3           -71.1
N         -50.3           -45.4
THD+N     -49.5           -44.9
sin250.8bit.nodither.png sin250.8bit.dither.png

In total, our 8-bit signal sees its SNR improve from 48 to about 64dB.
I don't know where they got those numbers from. I'm guessing 48 is from the 6*N formula and 64 they read from the FFT, but you can't read total noise from FFT, at least not without taking "FFT gain" into account.

Now when you upsample dithered audio by two, you still have the same SNR overall, but that is spread out over twice the samples.
No. It works like that only if upsample before applying dither. If you apply 8-bit dither to 44.1k file, you can upsample it even 1000x and you will still have -48 dBFS noise in the audible band.

so no wonder the faint hiss was gone.
Who was wondering? That's the first time you mention it :)
I don't know how you applied the dither but if you don't hear any difference, then I'm guessing it was not flat dither (like in my graphs) but rather dither with shaped noise (like in pma's examples).

With flat dither applied to DXD, i.e. 352k, i.e. 8x 44k, you should get noise floor, in the audible band, of about -57 dBFS (that's -48 from 8-bits and then -3 for each doubling of the total bandwidth)
 

j_j

Major Contributor
Audio Luminary
Technical Expert
Joined
Oct 10, 2017
Messages
2,267
Likes
4,758
Location
My kitchen or my listening room.
The concept of dither is so obscure that many in music production don't understand it either.

It's worse than that, I have more than one time had some producer or mixer person shout at me "YOU ARE AN IDIOT IF YOU THINK I AM GOING TO ADD NOISE TO MY MUSIC", coupled to rather loud applause from the crowd.

As to the source, I used our internal ABC/hr on it, but 20/20 is as conclusive as it gets, sorry. :) But I'm cheating, listening a very quiet room, over speakers, with good flat response. I could here the stationary noise at higher frequencies, which surprised me a bit, because I'm 70 years old.
 

G|force

Senior Member
Forum Donor
Joined
Apr 8, 2021
Messages
347
Likes
455
Location
Pioneer , CA
Thank you PMA for the original effort in August 2022
Please forgive me if what follows is daft ignorant or annoying.
Set ignore but set it to stun.

I have read the thread and was eager to take the test in FB2K. Reading into the thread I see objections that the test files are not appropriate for the test at hand. Amirm points out hearing acuity to 14kHz,
downsampling vs. upsampling, with different or adjusted files, and reposts to same.

I still want to take the test, what are the latest 'peer reviewed and approved' files to use?
 

Philbo King

Addicted to Fun and Learning
Joined
May 30, 2022
Messages
670
Likes
872
A challenge - can you tell 8-bit file from 16-bit file?

Based on a discussion in my thread God of Sinad
I have prepared a listening test with an original 44.1kHz/16bit file and the derivative from that file with 8bit resolution, with appropriate dither

The beloved Rickie Lee Jones, both files coded as 16-bit, but one of them is only in 8-bit resolution. Appropriate dither used. Here is the link, and I take only results from a DBT test, foobar ABX report that can be verified.

https://pmacura.cz/rlj8_16.zip

So, can you hear a difference? If you say yes, then please add your foobar ABX report.

FYI, this is what Deltawave calculates as a difference in time domain.

View attachment 223667
There is a vst plugin to measure this. Called Bitter.
 

Sebby

Active Member
Joined
Dec 18, 2022
Messages
116
Likes
84
Location
Milan, Italy
Una sfida: puoi distinguere un file a 8 bit da un file a 16 bit?

Basato su una discussione nel mio thread God of Sinad
Ho preparato una prova di ascolto con un file originale a 44.1kHz/16bit e la derivata da quel file con risoluzione 8bit, con opportuno dithering

L'amato Rickie Lee Jones, entrambi i file sono codificati a 16 bit, ma uno di essi ha solo una risoluzione a 8 bit. È stato utilizzato il dithering appropriato. Ecco il collegamento e prendo solo i risultati di un test DBT, rapporto foobar ABX che può essere verificato.

https://pmacura.cz/rlj8_16.zip

Allora, riesci a sentire la differenza? Se rispondi di sì, aggiungi il tuo report ABX foobar.

Per tua informazione, questo è ciò che Deltawave calcola come differenza nel dominio del tempo.

View attachment 223667
IEM Intime Sora 2 (50 Euro) + Topping G5 connected via USB to my PC. Tracks opened with Windows Music Player and at very low volume. I opened the firts track and knew from the first second that it was an 8 bit track. I opened the second track and the confermation came clear. The difference are so great throughout the song, bot in terms of background noise and (annoying) vocals and guitar, that not hearing them means being deaf. If you give me 100 tracks, 50 8bit and 50 16 bit, I could recognize them standing in the traffic of Rome
 

Attachments

  • P_20231025_001752.jpg
    P_20231025_001752.jpg
    151.1 KB · Views: 20
Last edited:

JakeK

Active Member
Forum Donor
Joined
Nov 9, 2022
Messages
137
Likes
122
Location
UK
I think the point @danadam is trying to make (and danadam, please correct me if I'm mistaken) is why not compare a 44.1/8-bit byproduct with the 44.1/16-bit original? That's the way to isolate the variable of bit-depth. If I read danadam's comments correctly, then by increasing the sample rate, you've effectively rigged the comparison in favor of the loss of 8 bits being less audible than it would otherwise be.
This is it in a nutshell IMO.

Comparing 44/8 to 44/16 is what I thought this thread was about initially and most people just skim reading would assume that same. I've done my own tests of different bit rates and depths and it was interesting in that you could almost get away with 44/12 and surprising that it still sounds like music even at quite low resolution so this thread at first looked interesting. It turns out you're comparing 96/8 to 44/16 which seems fairly pointless to me. You might as well try to compare 96/8 to 192/6 which you could do but to what purpose?

The thread title is false and designed to draw those in who, like me, think 'Of course I can tell the difference between 16 bit and 8bit!' but is actually a trick. Can the title be changed by a moderator?
 

tmtomh

Major Contributor
Forum Donor
Joined
Aug 14, 2018
Messages
2,636
Likes
7,488
This is it in a nutshell IMO.

Comparing 44/8 to 44/16 is what I thought this thread was about initially and most people just skim reading would assume that same. I've done my own tests of different bit rates and depths and it was interesting in that you could almost get away with 44/12 and surprising that it still sounds like music even at quite low resolution so this thread at first looked interesting. It turns out you're comparing 96/8 to 44/16 which seems fairly pointless to me. You might as well try to compare 96/8 to 192/6 which you could do but to what purpose?

The thread title is false and designed to draw those in who, like me, think 'Of course I can tell the difference between 16 bit and 8bit!' but is actually a trick. Can the title be changed by a moderator?

Yes, this is deeply misleading, and while @pma runs many good and educational tests that he writes about here, he also does some like this, with unnecessary confounding extra variables thrown in. I wish he'd be more accountable about this - or better yet, not do it at all.
 

Ageve

Active Member
Joined
Jun 13, 2021
Messages
192
Likes
640
Location
Sweden
Can you tell 8-bit file from 16-bit file in a DBT?

The question should be: Can you tell upsampled and noise shaped 8-bit file from 16-bit 44.1 kHz file in a DBT?

The answer is yes. I could tell the difference with HD650 connected to my M1 Macbook Pro (using the built-in headphone amp).

You can go even lower in Audition btw.

This is the original file resampled to 4-bit, 1 MHz:
ricke_4bit_1MHz.png

ricke_4bit_1MHz_freq.png


This is after converting back to 16-bit, 44.1 kHz:
ricke_4bit_1MHz_back_to_1644.png

ricke_4bit_1MHz_back_to_1644_freq.png


You can set the sample rate and bit depth manually. After conversion it will be treated as 8-bit (to avoid processing at 4-bit I guess).

4bitconv.png


Settings used:
convert.png
 
Top Bottom