[71 dB]

I do get what you are saying here.

There are a couple of other benefits of 24bit that I didn't delve into. When you conduct processing on music it generally sounds less artificial & develops less artifacts when using 24bit sources. Some recordings improve drastically to my ears when processed. I also enjoy listening to some recordings in multichannel which requires processing on sources where true multichannel mixes are not available. Also, when I compare 24/192 to 16/44, the overall presentation of the sound to my ears is much less harsh. For example, cymbals, trumpets and saxophones are much more lifelike & pleasing to my ears. I can abx a difference in these rates with foobar2000.

You are correct about 24 bit being "safer" bet compared to 16 bit if additional processing is done, althou as long as the processing isn't drastic 16 bit should be ok. When I say 44.1/16 is enough for consumer audio it assumes no additional processing or only mild processing (say cutting bass by 3 dB) is done.

Your DAC probably has different sound with 192 kHz and 44.1 kHz. You could try upsampling 44.1 kHz to 192 kHz and test if the sound changes. DACs are imperfect devices and always introduce something to the sound because of jitter, reconstruction filters and non-linearities. How audible it is is another issue.

 

[bigshot]

I can abx a difference in these rates with foobar2000.

I don't believe you. Why did you wait until now to say you had done controlled tests? You're just making stuff up.

 

[KeithPhantom]

I can abx a difference in these rates with foobar2000

Well, prove it by providing a log and your methodology. I don't think you can hear either quantization noise or the noise floor.

 

[bigshot]

He's googling other threads to see what he should say.

 

[miksu8]

Maybe a dumb question, but is conversion from 24 bit to 16 bit lossless when sample rate is the same? Scientific explanation would be nice.

 

[bigshot]

No, 24 bit has more zeros and ones, but it doesn’t matter because it’s been clearly proven that human ears can’t detect a difference. See the article in my dig called 16 bit is all you need.

 

[miksu8]

No, 24 bit has more zeros and ones, but it doesn’t matter because it’s been clearly proven that human ears can’t detect a difference. See the article in my dig called 16 bit is all you need.

If it's the last link in your sig it says not found.

 

[castleofargh]

You are correct about 24 bit being "safer" bet compared to 16 bit if additional processing is done, althou as long as the processing isn't drastic 16 bit should be ok. When I say 44.1/16 is enough for consumer audio it assumes no additional processing or only mild processing (say cutting bass by 3 dB) is done.

Your DAC probably has different sound with 192 kHz and 44.1 kHz. You could try upsampling 44.1 kHz to 192 kHz and test if the sound changes. DACs are imperfect devices and always introduce something to the sound because of jitter, reconstruction filters and non-linearities. How audible it is is another issue.

Most of those processing concerns can be addressed without starting with 24bit files(or high sample rate for that matter), and usually are without us having to lift a finger. The DSPs that still run at 16/44 do it as a deliberate choice(or were made by a guy in 5 minutes and probably have better alternatives). Even something as simple as volume control is handled at 32 or 64bit nowadays, and anything that can benefit from resampling will just do it. Then stuff are dithered (or not) and resampled when sent out to the DAC at whatever bit depth and sample rate we have set for it. Which usually should be 24 or even 32 if the DAC and driver happens to handle it.
All this will still happen with a 16/44 file. And a 24/96 probably still isn't the bit depth and sample rate used by heavy VSTs or whatever DSP or even the DAC itself that will probably end up with a handfull of bits and crazy high sample rate.

Obviously working with higher resolution makes sense for various reasons in production, but indeed for consumer audio we can have most of the benefits of higher everything while using 16/44 tracks.

 

[castleofargh]

Maybe a dumb question, but is conversion from 24 bit to 16 bit lossless when sample rate is the same? Scientific explanation would be nice.

The encoding format is lossless, the conversion to a lower resolution isn't. So you can put 16/44 data into Wav, Flac or any lossless format at the same resolution and you will be able to retrieve all that data as it was. You can decode it, re-encode it, and still you should get the same data. That's what the lossless denomination of the format describes.
But if you go from 24bit data to 16bit data, you've literally changed the resolution and removed some bits at the end of the code for each sample. That's not a lossless operation.

 

[gregorio]

There are a couple of other benefits of 24bit that I didn't delve into.
[1] When you conduct processing on music it generally sounds less artificial & develops less artifacts when using 24bit sources.
[2] Some recordings improve drastically to my ears when processed.
[3] Also, when I compare 24/192 to 16/44, the overall presentation of the sound to my ears is much less harsh. For example, cymbals, trumpets and saxophones are much more lifelike & pleasing to my ears.
[3a] I can abx a difference in these rates with foobar2000.

1. Processing virtually always occurs at 32bit or 64bit, regardless of whether the source files are 16bit or 24bit. So, the artefacts are at exactly the same level regardless of a 16bit or 24bit source!

2. Sure, but of course that's the freq response of your HPs (or speakers + room), the freq response of your ears and your personal preferences, which has nothing to do with the source file's bit depth!

3. Firstly, there is no mechanism/process which could cause that effect and Secondly, it is easily verified (with a null test for example) that the "sound to your ears" is identical with 16bit or 24bit. The only difference is ultrasonic content (above about 22kHz) and very low level dither noise, BOTH of which are inaudible (to your or anyone else's ears) at reasonable listening levels! The only rational explanation is therefore that you are "hearing" some difference that occurs AFTER the "sound to your ears", EG. Some bias in your perception.
3a. There are two conditions under which it can be possible to ABX a difference between 24/192 and 16/44:
A. Using a 24/192 recording with significant frequency content at 22kHz or higher that is causing IMD in your amp or speakers (within the audible spectrum) and comparing it with a 16/44 conversion, that obviously can't have any content above 22kHz and is therefore not causing any audible IMD. And ...
B. Using a very low level, non noise-shaped (TPDF) dithered test signal (or finding a very low level segment in a music recording) and amplifying it massively, so that the dither is audible.
In case "A", although one could discern an audible difference, the result is actually backwards. The 24/192 version is reproduced with unwanted distortion, effectively at a LOWER fidelity than the 16/44 conversion! In case "B" we have created an artificial scenario that CANNOT exist when we're listening to music recordings, in practice it would massively overload a system, blow your drivers and/or ear drums, as explained in the OP. Furthermore, the difference with this artificial scenario is the audibility of TPDF dither noise, NOT a "much less harsh presentation".
If you really can "ABX a difference in these rates" at reasonable levels, then you MUST provide the evidence, because you are contradicting a significant body of established scientific evidence!

Btw Rob Watts, who I think also knows something about this, says there is also an issue with transients with DSD.

Rob Watts does apparently "know something about this". Unfortunately though he misrepresents/lies about it!! From your linked post:

[1] .... This is the reason why -6dB is the max for DSD; but the innate noise floor modulation problem occurs well below -6dB.
[2] .. Dave's noise shaper will perfectly resolve a -301 dB signal - and this is essential for the perception of sound stage depth.
[3] My pulse array (5 bits, 2048FS) does not show this behavior on simulation - a -60dB step change has no consistent delay compared to 0dB step change.
[3a] This is due to the very high speed of operation,
[3b] the 5 bit resolution, and
[3c] the fact that one can properly dither a 5 bit delta sigma, but you can't dither a DSD system.

1. Obviously that's nonsense. If -6dB is the max for DSD, then how can a noise problem (or anything else) occur beyond the max? If -6dB really were the max, every SACD would be unlistenable because the very highest peaks could only be 6dB above the noise floor and the majority of the recording would be below the noise floor! The actual noise floor (in the audible spectrum) of SACD is NOT -6dB, it's about -120dB, which is easily verifiable.

2. If the last one wasn't bad enough, this assertion is way beyond nonsense, it's utterly ridiculous! Consider that the theoretical limit of 24bit is -144dB but it's only in theory, in practice we can't achieve anywhere near that level because if 0dB is say the sound level of a truck driving past from about 10ft away, then -144dB would roughly be the sound level produced by two hydrogen atoms colliding! Of course, the sound of two hydrogen atoms colliding is way, way below the ability of speakers/drivers to reproduce and of the human ear drum to detect. So, talking about a real world signal at -144dB is very silly indeed BUT here's Rob Watts talking about a "-301 dB signal", which is roughly 100 million times lower than -144dB! So what's 100 million times more than "very silly indeed"?? The best I can come up with is "utterly ridiculous beyond imagination" which apparently, in Rob Watts marketing BS language, translates to "essential for the perception of sound stage"! It makes Monty Python seem entirely reasonable ... you've got to laugh!!!

3. And neither does an "off the shelf" $2 DAC chip with a standard linear phase filter!
3a. True but obviously you don't need more speed than a stock DAC chip.
3b. That's interesting, many/most pro audio ADC/DACs use 6 bit resolution at similarly high sample rates, I wonder why Rob Watts "cheaps out" with only 5 bits?
3c. Nonsense, of course you can dither a DSD system, if you couldn't all SACDs would be swamped with noise and un-listenable. It's true that you can't dither a 1bit DSD system adequately enough to linearise all quantisation distortion but you can reduce it to below audibility.

Sorry, but it's one of the oldest audiophile marketing tricks in the book: Take some issues that are inaudible, purely theoretical or miniscule to the point of laughable, FALSELY describe them as "massive and unique problems", explain how your new/super-duper DAC solves them and is therefore "massively" better than other DACs and worth it's MASSIVELY over-inflated price!!!

G

 

[KeithPhantom]

But if you go from 24bit data to 16bit data, you've literally changed the resolution and removed some bits at the end of the code for each sample. That's not a lossless operation.

And increased the noise floor by adding more quantization errors and decimating the theoretical dynamic range of the music.

 

[danadam]

If it's the last link in your sig it says not found.

Here's the correct link: https://xiphmont.dreamwidth.org/57937.html

@bigshot You should update your sig

EDIT: Ups, it just leads to the bigshot's link
Wayback Machine then: https://web.archive.org/web/20200426202431/https://people.xiph.org/~xiphmont/demo/neil-young.html

 

[bigshot]

@bigshot You should update your sig

Fixed! Thanks!

 

[71 dB]

And increased the noise floor by adding more quantization errors and decimating the theoretical dynamic range of the music.

If you simply truncate the 24 bit sample points into 16 bit samples then this happens. If you add 16 bit dither noise to the 24 bit file before truncation, you mitigate problems: quantization errors get randomized into noise which doesn't correlate with the signal and also you prevent increased distortion. In that sense the original signal still has it's original dynamic range, but is masked partly by the added dither noise. Reverb tails for example decay into the noise the same fashion analog sound works so that signal levels below the dither noise level can be heard (if the listening level was insane) whenever the dither doesn't completely mask it. A 16 bit audio file produced from 24 bit file can theoretically have a "ananog-like noisy" dynamic range of 120 dB depending on what kind of dither is used. It's good to remember bits do not quarantee dynamic range. If I digitize old noisy C-cassettes into 24 bit files I certainly don't get huge dynamic range. A 24 bit file might have only 18 bits worth of dynamic range* for example so that turning it into "ananog-like noisy" 16 bit file which has a "noisy" 20 dB dynamic range doesn't mean much. In consumer audio about 80 dB of dynamic range is all we need, about 13 bits worth of dynamic range. In this sense even 16 bit is overkill by a few bits, but that's just a nice safety margin.

* Try and record acoustic instruments at 24 bit dynamic range and come back to tell us how well that went.

 

[gregorio]

[1] But if you go from 24bit data to 16bit data, you've literally changed the resolution and removed some bits at the end of the code for each sample.
[2] That's not a lossless operation.

1. Careful here. What do we mean by "resolution"? Effectively, we haven't changed the resolution, the resolution of the output signal is the same, what has changed is the amount of noise which accompanies the output signal but at 16bit, that noise is inaudible.

2. It IS an audibly lossless operation though. We could even argue that in practice we have not "lost" any resolution because at least the last 10 LSBs of a 24bit recording are just noise, of which we loose 8 and add 1 (16bit) LSB of .... noise. So, we've "literally changed the resolution" purely in terms of exchanging inaudible noise for the same amount of inaudible noise! Is that really lossy?

Therefore:

[1] And increased the noise floor by adding more quantization errors and
[2] decimating the theoretical dynamic range of the music.

1. Not really. There are no quantisation errors, what we have instead is dither noise, which *would* have "increased the noise floor" ONLY IF the noise floor of the 24 bit recording were below about -120dB. Do you know of any commercial 24bit recordings with a noise floor lower than -120dB? The lowest noise floor commercial recording I've heard of, has a noise floor about 300 times higher than that and the vast majority 1,000+ times higher. Bare in mind, it's standard practice to use noise-shaped dither when converting from >16bit to 16bit, which puts the digital noise floor of the 16bit version at about -120dB.

2. That depends on what you mean by "theoretical dynamic range of music". Your assertion could be true (in theory) for only one very specific form of music: Music comprised SOLELY of signals digitally synthesised (at >20bit), but even then it is only "theoretical" because AFAIK, there are no music recordings in practice that actually employ a dynamic range greater than 120dB. All other forms of music, even including almost all electronic music, is constrained by either analogue inputs, analogue modelling, acoustic recording conditions or all of the above, EACH of which have a noise floor above or way above -120dB. So even theoretically, the dynamic range of music is NOT "decimated", except in the one *potential* case just mentioned. Incidentally, almost all electronic music, even purely "generated in the box" electronic music (without any recorded instruments or vocals), relies on at least some "digital synths" which aren't really synthesizers, they're reliant on sample playback and therefore constrained by the noise floor of the samples.

G