HydrogenAudio

Hydrogenaudio Forum => General Audio => Topic started by: BlueDragon on 2010-11-02 02:28:28

Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-02 02:28:28
I would like to have the opinion of more knowledgeable people than me in transcodes...

Here the information for a piece of music I downloaded (flac):

Bitrate: 1'327 kbps
Spectrogram:



EDIT: I could not find a way to edit the topic header which should be 44.1kHz for 16bit and 24bit 
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: carpman on 2010-11-02 03:05:18
A spectrogram won't tell you about bit depth.
EDIT: This will explain why it's not 22.1kHz (http://en.wikipedia.org/wiki/Sampling_rate#Sampling_theorem)

C.

Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-02 03:24:17
Thanks carpman for your reply.

Did you notice the bitrate indication? This should be directly linked with the sample rate. Correct me if I'm wrong... 

EDIT: Besides anyway it's 44'100 Hz in both (bit) cases (I made a typo in the topic header but I found no way to correct it...)
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: 2Bdecided on 2010-11-02 09:48:24
A spectrogram is pretty useless for this task.

Even if you have the original file, unless the 16>24 conversion was carried out simply by adding 8 zeros (which you can check for*), then there's absolutely no way of knowing whether an arbitrary 24-bit file was captured directly at that bitdepth, or created from a 16-bit file by adding 8-bits of random noise. The end result will look, sound, and measure the same (assuming a source, like vinyl, where the noise is already above the 16-bit noise floor at all frequencies).

* - if the FLAC bitrate is that high (you reported 1.3Mbps), it's very unlikely that the 24-bit file has 8 zero LSBs, so you can discount that possibility

The most important fact is that, for a vinyl capture, it really doesn't matter either way.

Cheers,
David.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: dhromed on 2010-11-02 12:02:49
[the bitrate] should be directly linked with the sample rate.
Nope.[/s]

Edit: Ironically, I missed your bitrate indication. Disregard me!

And as 2Bdecided said; since it's from vinyl, it barely matters anyway.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-03 00:24:09
So if I understand you all correctly:

1. A spectrogram is absolutely useless in this situation (e.g. trying to find out if vinyl>24bit or if vinyl>16bit>24bit has been the transcode path).

2. The (high) bitrate tells me here that very likely the transcode has been done directly from vinyl to 24bit though it can basically not be 100% sure.

3. Conclusion: There's no way to be 100% sure if the final (24bit) result  directly came from a vinyl source but given the high bitrate (1.3kbps) it's rather very likely.

4. If that is correct then what are the bitrates (range) one should expect from various transcodings like:
a. vinyl > 16 bit
b. vinyl > 24 bit
c. CDDA > 16 bit
d. SCDA > 16 bit

Thank you to you all for enlighting me. 
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Soap on 2010-11-03 00:47:13
2. The (high) bitrate tells me here that very likely the transcode has been done directly from vinyl to 24bit though it can basically not be 100% sure.

3. Conclusion: There's no way to be 100% sure if the final (24bit) result  directly came from a vinyl source but given the high bitrate (1.3kbps) it's rather very likely.

No and No.
The high bitrate tells you the bottom 8 bits are likely not all zeros.
Period.

Tells you nothing else.
Not the original bitdepth of the rip, doesn't even suggest vinyl.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-03 01:01:54
Not the original bitdepth of the rip, doesn't even suggest vinyl.


Can you please elaborate, I don't understand, sorry...   

So I try to formulate it again in other words... There's abosutely no way to know if a 24bit transcode which you downloaded in a flac file format really comes from a vinyl source??...   
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Soap on 2010-11-03 01:09:30
So I try to formulate it again in other words... There's abosutely no way to know if a 24bit transcode which you downloaded in a flac file format really comes from a vinyl source??... 

correct.
No way to be sure.
Buy the frickin album if you want to know what you have.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-03 01:29:59
No way to be sure.


Ok no way to be sure but could at least the bitrate (if not the spectrogram) tell something about the likeliness??...   

Thanks
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Soap on 2010-11-03 01:31:51
The likeliness of it being a vinyl rip?  No.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: ExUser on 2010-11-03 02:02:59
If it's a vinyl rip, it doesn't even matter if it's encoded at 24-bit anyhow. There is no way you're getting more than the equivalent of 12 bits of signal-to-noise ratio; maybe even 8 bits with some good noise-shaping. 24-bit vinyl rips are a totally idiotic waste of space. Kick whoever made it for you in the nuts for me.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: WernerO on 2010-11-03 07:04:06
maybe even 8 bits with some good noise-shaping.


Could you quantify this? Given the 44.1kHz sample rate, what would be the 20kHz-wide SNR of the noiseshaped channel?
And how much noise signal power would be packed in the last 2kHz band? How would the tweeter like that?

Or is it just conjecture?

Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: ExUser on 2010-11-03 07:31:55
I read Arnold write it somewhere here. I don't have numbers.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: WernerO on 2010-11-03 08:09:21
I read Arnold write it somewhere here. I don't have numbers.


Ah, I see.

I did a quick experiment with a none-too-extreme noise shaper. Broadband sub-20kHz noise is then at -44dB (RMS), considerably worse than a decent LP, and very much worse when weighting is included. Super-20kHz noise contains frequent peaks up to -26dB(FS). That is pretty loud.

Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Arnold B. Krueger on 2010-11-03 11:27:58
maybe even 8 bits with some good noise-shaping.



That seems like a bit of a reach.

Quote
Could you quantify this? Given the 44.1kHz sample rate, what would be the 20kHz-wide SNR of the noiseshaped channel?


You're using the wrong criteria. Noise shaping doesn't drop the overall broadband noise floor. Instead, noise shaping moves the noise to the parts of the frequency spectrum where it is less noticeable. In oversampling, noise shaping is used to move the noise to frequencies that are ultimately filtered out. In dithering, @ 44 KHz, the noise is moved to the high end of the audible spectrum where the ear is far less sensitive.

So, noise shaping can be used to produce a noise floor with 16 bit samples that approaches the perceived noise levels of unshaped samples with 20 or more bits.

Of course the irony is that the basic source material has a far higher noise floor than even unshaped 16 bits.

If this sounds fishy, its really the same reason why LPs with noise floors comparable to 12 bit digital can sound as good as they do. The LP system has long been engineered to push noise to the frequency extremes where the ear is far less sensitive. This is one of the major benefits of RIAA equalization.


Quote
And how much noise signal power would be packed in the last 2kHz band?


If you look at the Fletcher-Munson curves you see that you don't have to pack all the noise into the last 2 KHz.  The sensitivity of the ear falls off rapidly above 5 Khz or so, particularly at low amplitudes.

Quote
How would the tweeter like that?


No problem since even with noise shaping, the actual noise levels remain relatively low.

Quote
Or is it just conjecture?


Noise shaping is technology that goes back to at least FM radio (late 1930s)  and 78 rpm recordings (similar time frame). In the old days we called it "de emphasis".  The RIAA playback curve includes massive noise shaping, and the FM deemphasis curve also does so, but to a lesser degree.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: ExUser on 2010-11-03 14:33:06
Yeah, I can't find a citation for that 8-bit figure. Let's pretend I never said it.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: 2Bdecided on 2010-11-03 16:20:44
Buy the frickin album if you want to know what you have.
LOL!. Yeah, but then you've got to digitise it yourself.

Seriously, 16 vs 24 bits is irrelevant in this context (and most contexts!). What does it sound like? Does it sound like pristine vinyl played on a top-notch record player? If so, be happy. If not, 24-bits won't rescue it anyway.

Cheers,
David.

Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: greynol on 2010-11-03 17:11:08
In oversampling, noise shaping is used to move the noise to frequencies that are ultimately filtered out.

Oversampling is used to help take the complexity away from the reconstruction filter.

If you use noise shaping and then filter out the noise that does the work to dither, what happens to the effect of the dither?
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Arnold B. Krueger on 2010-11-03 17:36:48
In oversampling, noise shaping is used to move the noise to frequencies that are ultimately filtered out.

Oversampling is used to help take the complexity away from the reconstruction filter.

If you use noise shaping and then filter out the noise that does the work to dither, what happens to the effect of the dither?


Memo to noise shaping design staff: Don't shape the noise so agressively that it gets  totally filtered out by later processing steps. This is facilitied by the fact that shaped noise is subject to some spectral speading along the way.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: JapanAudio on 2010-11-03 23:28:22
At first thought i would say that it is only possible to falsify the statement that it is a vinyl rip, meaning that you could only tell if it were NOT real. Like the 8 zeros thing (that would be pretty lame).

There is no actual relation between bit depth and sample rate, as others have said, but i would like to add that 24bit media is commonly seen sampled at 88.2, 96kHz, or higher. So i can see where you're coming from with your suspicion.

The only motivation i can see behind going from 16 to 24bit is to prevent quantization error during normalization. I guess that's fine, but your files are gonna be roughly 1.5 times bigger for the same quality.

But honestly, i wouldn't trust any vinyl rip, even knowing that it is genuine. Vinyl is meant to be spinning on a turntable for all i know. And no matter how good your equipment, recording analog audio from another recording equals to loss of fidelity in my book. That would be like taking a picture of a picture of the Mona Lisa, how cheesy is that?

Anyway. As for the sake of just wanting a higher quality version, i would go for an original 24bit digital download from the record company, if such a release exists. Go for Da Vinci himself.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-04 04:40:02
Ok thanks to all contributors for all these lively reactions.

I must honestly say that I did not understand a single word of all what has been discussed amongst the audio gurus out there... 
but I am pleased to see that apparently it's not that easy to understand.

Having understood what I could I dont't see then why it's a very wide spread habit to digitalise older vinyl audio material in 24bit. So apparently 90% of the people (of course it's just a guess but in any case many many people do it) do not understand anything... 

But maybe it's like the topic if one can hear the difference between mp3 cbr 320 and flac 16bit for the same source. So probably the discussion will never really end...   

Thanks again anyway! 
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: WernerO on 2010-11-04 06:28:04
If you use noise shaping and then filter out the noise that does the work to dither, what happens to the effect of the dither?


That depends entirely of the resolution/accuracy available in the domain where you do
that filtering.

Consider DSD with a signal chain ultimately feeding a loudspeaker with low bandwidth as a limit case.

--

To the OP: I'd say that about 1023 people in 1024 don't understand these things, so don't worry indeed.

Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: ExUser on 2010-11-04 06:31:56
But maybe it's like the topic if one can hear the difference between mp3 cbr 320 and flac 16bit for the same source. So probably the discussion will never really end...
Myself and several others like me have scientific proof that we can hear that difference, at least on certain samples. Generally though, 320 is good enough. I use LAME V5 on my portable/in my car sometimes. Sounds great.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: JapanAudio on 2010-11-04 07:35:33
But maybe it's like the topic if one can hear the difference between mp3 cbr 320 and flac 16bit for the same source. So probably the discussion will never really end...
Myself and several others like me have scientific proof that we can hear that difference, at least on certain samples. Generally though, 320 is good enough. I use LAME V5 on my portable/in my car sometimes. Sounds great.

Yes I did a test with a professional pianist with a great ear and she could tell pretty much every time which was which.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: greynol on 2010-11-04 08:05:36
But maybe it's like the topic if one can hear the difference between mp3 cbr 320 and flac 16bit for the same source. So probably the discussion will never really end...
Myself and several others like me have scientific proof that we can hear that difference, at least on certain samples. Generally though, 320 is good enough. I use LAME V5 on my portable/in my car sometimes. Sounds great.
Yes I did a test with a professional pianist with a great ear and she could tell pretty much every time which was which.

Are still talking about hi-res or are we veering further off-topic?  Other than the fact that the testing is the same and there are plenty of people who claim to be able to hear the difference but really cannot, detection of lossy artifacts and the ability to hear frequencies over 22kHz and/or quantization error at the 16th bit are really two different things.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: greynol on 2010-11-04 08:11:13
Consider DSD with a signal chain ultimately feeding a loudspeaker with low bandwidth as a limit case.

I'm talking about the noise used to dither and then it's removal.  If the noise used to dither is filtered, then there is no more dither.  If there is no audible distortion after the filtering then the dither was never necessary to begin with.  IOW, there is no free lunch.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: cliveb on 2010-11-04 08:28:37
Having understood what I could I dont't see then why it's a very wide spread habit to digitalise older vinyl audio material in 24bit.

There are two valid reasons to digitise vinyl at 24 bit:

There is another (completely invalid) reason to digitise vinyl at 24 bit:

Let me just say that I'm a long-time vinyl digitiser, and I do all my work at 16 bit. Having done some experiments at 24 bit, I couldn't find any advantage at all.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: lvqcl on 2010-11-04 08:31:34
I'm talking about the noise used to dither and then it's removal.

Do you mean something like:
24-bit -> reducing bit depth with noiseshaped dithering -> 16-bit -> lowpass filter -> 16 bit ?
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: greynol on 2010-11-04 08:41:38
I don't  think that's what Arnold was talking about, no.  Though, quite frankly I don't have the faintest idea why Arnold was lumping noise shaping and oversampling together.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: greynol on 2010-11-04 08:44:51
You're going to be doing many passes of post-recording processing and want to avoid the danger of rounding errors accumulating enough to bring them above the vinyl noise floor. I don't personally believe this is actually a danger - it would takes dozens of DSP operations at 16 bit to get to that stage.

Perhaps this is a good reason to perform processing at higher than 16 bits, but I don't see it as a reason to necessarily digitize at higher than 16 bits.  They aren't the same thing, of course.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: WernerO on 2010-11-04 11:19:29
I'm talking about the noise used to dither and then it's removal.  If the noise used to dither is filtered, then there is no more dither.  If there is no audible distortion after the filtering then...


Which is exactly what I meant, please read again (and lvqcl's post as well).

It all depends.

If the filtering domain remains at the same resolution, then the benign effect of the noise shaping is nulled and quantisation distortion rises.

If the filtering domain has a higher resolution than the noise shaped source stream, i.e. expanded to 24/32 bit or to the analogue domain, then the low-pass transfer of the filter smears the shaped noise energy through the passband, thus recreating the required intermediate signal states in the pass band, maintaining the source's resolution.

Again, DSD. It is 1 bit. Its noise-shaped dither part is low-pass filtered by the whole post-DAC signal chain. And yet what you hear is better than 1 bit in resolution. Are you going to argue that the noise shaping then wasn't necessary in the first place?


Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: pdq on 2010-11-04 12:56:48
Yes I did a test with a professional pianist with a great ear and she could tell pretty much every time which was which.

Sorry, but unless this was a true double-blind ABX test, this information is worthless!
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Goratrix on 2010-11-04 14:06:50
LOL!. Yeah, but then you've got to digitise it yourself.


well, he could just buy the CD (http://www.amazon.com/Raga-Jogeshwari-Ravi-Shankar/dp/B000007T07).
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: Arnold B. Krueger on 2010-11-04 15:14:23
I don't  think that's what Arnold was talking about, no.  Though, quite frankly I don't have the faintest idea why Arnold was lumping noise shaping and oversampling together.


Noise shaping is an integral part of why oversampling converters work so well.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: 2Bdecided on 2010-11-04 16:02:26
I don't  think that's what Arnold was talking about, no.  Though, quite frankly I don't have the faintest idea why Arnold was lumping noise shaping and oversampling together.


Noise shaping is an integral part of why oversampling converters work so well.
...but you can noise shape without oversampling.

...and you can oversample without noise shaping - but noise shaping makes it far more efficient - i.e. in an oversampled system you need far fewer bits to delivery a given SNR in the base band if you use noise shaping.

I'm sure you know this, but just for clarity!

Cheers,
David.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: JapanAudio on 2010-11-04 16:39:21
Consider DSD with a signal chain ultimately feeding a loudspeaker with low bandwidth as a limit case.

I'm talking about the noise used to dither and then it's removal.  If the noise used to dither is filtered, then there is no more dither.  If there is no audible distortion after the filtering then the dither was never necessary to begin with.  IOW, there is no free lunch.

I don't really see the free lunch...

Dither is used to pseudo-randomly spread out the quantization noise. Noise shaping will shift the noise power to different frequency bands. Sure if you shift most of it to the high end then apply a lowpass filter you may be getting a free candy, but not the whole lunch. On the other hand if you're not applying a lowpass then the noise is still there, subjectively less audible, but still objectively present.

Yes I did a test with a professional pianist with a great ear and she could tell pretty much every time which was which.

Sorry, but unless this was a true double-blind ABX test, this information is worthless!

I'm not asking you to believe me --It's not like I'm pretenting to be able to do it myself. I'm just reporting what I've witnessed.

TBH I'd rather look at a waveform, a spectrum and statistical data rather than relying on a person's opinion. And all the facts are there.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: JapanAudio on 2010-11-04 16:43:57
But maybe it's like the topic if one can hear the difference between mp3 cbr 320 and flac 16bit for the same source. So probably the discussion will never really end...
Myself and several others like me have scientific proof that we can hear that difference, at least on certain samples. Generally though, 320 is good enough. I use LAME V5 on my portable/in my car sometimes. Sounds great.
Yes I did a test with a professional pianist with a great ear and she could tell pretty much every time which was which.

Are still talking about hi-res or are we veering further off-topic?  Other than the fact that the testing is the same and there are plenty of people who claim to be able to hear the difference but really cannot, detection of lossy artifacts and the ability to hear frequencies over 22kHz and/or quantization error at the 16th bit are really two different things.

I'd love to see neuroimaging on people hearing higher-than-audible frequencies... Anyway, isn't hearing lossy artifacts enough of a motivation to go lossless?
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: DVDdoug on 2010-11-04 17:54:56
Quote
TBH I'd rather look at a waveform, a spectrum and statistical data rather than relying on a person's opinion. And all the facts are there.
The focus at HydrogenAudio is what we can hear rather than what we can measure.

Of course there is some correlation.  (i.e. Nobody is going to argue or demand an ABX test if you claim that 16-bits sounds better thatn 8-bits.)  The people who develop lossy encoders probably know more about human sound perception than anyone else and they should be able to make a good correlation between what we measure and what we hear. 


But, the audiophile community is full of people who claim to hear differences that they cannot hear in blind tests.  (The audiophile community is also full of people who say blind tests are invalid!!!)  Thus, we have TOS #8 (http://www.hydrogenaudio.org/forums/index.php?showtopic=3974)[/color] that says all claims of sound quality must be based on proper listening tests.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: BlueDragon on 2010-11-05 01:10:07
LOL!. Yeah, but then you've got to digitise it yourself.


well, he could just buy the CD (http://www.amazon.com/Raga-Jogeshwari-Ravi-Shankar/dp/B000007T07).


No, that's the problem. Apparently it exists only in (older) vinyl LP... 
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: krabapple on 2010-11-05 05:06:54
I'd love to see neuroimaging on people hearing higher-than-audible frequencies... Anyway, isn't hearing lossy artifacts enough of a motivation to go lossless?


primitive 'neuroimaging' at best, but since you asked...

http://jn.physiology.org/cgi/content/full/83/6/3548 (http://jn.physiology.org/cgi/content/full/83/6/3548)
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: JapanAudio on 2010-11-05 05:31:05
I'd love to see neuroimaging on people hearing higher-than-audible frequencies... Anyway, isn't hearing lossy artifacts enough of a motivation to go lossless?


primitive 'neuroimaging' at best, but since you asked...

http://jn.physiology.org/cgi/content/full/83/6/3548 (http://jn.physiology.org/cgi/content/full/83/6/3548)

Very cool.
Title: 16bit/22.1kHz or 24bit/44.2kHz
Post by: greynol on 2010-11-05 05:42:29
Despite attempts to do so, the results of that test have not been replicated.

Those who wish to discuss this study can do so here:
http://www.hydrogenaudio.org/forums/index....showtopic=68348 (http://www.hydrogenaudio.org/forums/index.php?showtopic=68348)

Any discussion about it in this thread will be binned per TOS #5.