How do you master a DSD recording?

[71 dB]

I do know about this, the higher bit depth yields a lower noise floor, thus a higher dynamic range, it is true that 16 bits is already pretty sufficient, but like I mentioned, maybe, just maybe, loud orchestra pieces might benefit with a dynamic range over 96dB since it can reach 100 or more.

Even the most dynamic orchestral music doesn't need more than 80 dB of dynamic range (with safety margin, in practise even less). Why?

Reason 1: You don't listen to music in "total silence". You have background noise which is in a quiet living room about 30 dB.

Reason 2: Your hearing has a dynamic range of about 70 dBs are any moment in time. If you are listening to loud sounds, your threshold of hearing rise and in order to hear very quiet sounds (below 20-30 dB) you need to stay in silence for a while. That's why you can't experience the large dynamic changes in music beyond about 70 dB even if you listened to it in a very quiet anechoic chamber.

Reason 3: Electronics has a noise floor too. Getting more than 90 dB of electronic dynamic range is challenging. But your headphone amp is rated at 110 dB of signal to noise ratio you say? Yeah, maybe, but that's a marketing measurement meaning maybe in an optimal situation for optimal signal you might get something like that, not with any random signal in typical listening.

Reason 4: Changes in dynamics more than 60 dBs with music is annoying. There is a reason why popular music has so flat dynamics, but compressing the dynamics too much makes the music dead too, so there is a optimal range for dynamics in music of 15-30 dB.

Vinyl gives 60 dB (10 bits worth) of dynamic range at best. Vinyl-lovers are happy with that. My rule of thumb is that consumer audio doen't need more than 13 bits (78 dB) of dynamic range if used well. That's why 16 bits is even overkill and definitely enough. In music production things are different and 24 bits have benefits, but consumers need only 13 well used bits (whole scaled used + properly dithered).

 

[Chimmy9278]

Even the most dynamic orchestral music doesn't need more than 80 dB of dynamic range (with safety margin, in practise even less). Why?

Reason 1: You don't listen to music in "total silence". You have background noise which is in a quiet living room about 30 dB.

Reason 2: Your hearing has a dynamic range of about 70 dBs are any moment in time. If you are listening to loud sounds, your threshold of hearing rise and in order to hear very quiet sounds (below 20-30 dB) you need to stay in silence for a while. That's why you can't experience the large dynamic changes in music beyond about 70 dB even if you listened to it in a very quiet anechoic chamber.

Reason 3: Electronics has a noise floor too. Getting more than 90 dB of electronic dynamic range is challenging. But your headphone amp is rated at 110 dB of signal to noise ratio you say? Yeah, maybe, but that's a marketing measurement meaning maybe in an optimal situation for optimal signal you might get something like that, not with any random signal in typical listening.

Reason 4: Changes in dynamics more than 60 dBs with music is annoying. There is a reason why popular music has so flat dynamics, but compressing the dynamics too much makes the music dead too, so there is a optimal range for dynamics in music of 15-30 dB.

Vinyl gives 60 dB (10 bits worth) of dynamic range at best. Vinyl-lovers are happy with that. My rule of thumb is that consumer audio doen't need more than 13 bits (78 dB) of dynamic range if used well. That's why 16 bits is even overkill and definitely enough. In music production things are different and 24 bits have benefits, but consumers need only 13 well used bits (whole scaled used + properly dithered).

Very valid reasons.

That 10 bit bit depth of vinyl as are also what makes them have a hiss, because you are hearing the noise floor if my understanding is correct, couple that with dust and what not. So in practice, I would like a bit more of depth for a lower noise floor, and like you said, 16 is indeed enough.

24 but in consumer might be over kill.

 

[71 dB]

24 bit in consumer might be over kill.

It definitely is, because 16 bit already is. If shaped dither is used, 16 bit digital audio can reach amazing perceptual dynamic range (up to 120 dB, which is 40-50 dB more than any consumer needs).

 

[gregorio]

Unfortunately you’ve gone back to linking to BS videos again Although to be fair, it’s more difficult to spot and what he states is largely correct or only slightly wrong. It’s what he fails to mention that makes it BS! Yes, there are artefacts with filters, the pre and post ringing he shows, HOWEVER: He fails to mention that most of the energy of that ringing is near the Nyquist point, so even using a 44.1kHz sample freq the ringing is inaudible. Also, he demonstrates this “filter ringing” using an illegal signal, a signal that never exists in the real world, which exacerbates the effect. Using real world signals, we rarely see this ringing and when we do, it’s very low level, so even more inaudible.

I think that some ADCs might already be doing this, when you pick per say 48 or 44.1, it would be sampling at twice or thrice that, and downsamples for the output.

The first pro ADC I bought was nearly 30 years ago, initial sampling was 128 times higher than the 48 or 44.1 sample rates. Modern ADCs operate 4 or more times higher again. This allows a very gentle analogue filter and then a digital “decimation” filter to arrive at the desired sample rate/bit depth.

DACs essentially do the process in reverse, a “reconstruction” filter and then a gentle analogue filter. Standard, cheap, steep linear phase filters are absolutely fine, with inaudible artefacts.

Maybe hi res is worth it, forgot to consider its higher bit depth earlier too. Some orchestra pieces can reach 130dBs is what I heard.

Don’t confuse peak levels with dynamic range. Orchestras can in fact reach much higher than 130dB, I’ve measured up to 138dB but that’s right in the middle of the orchestra, just 2-3 feet in front of the trumpets. In an ideal seat in the audience however, around 100dB or so is about the most you’ll get and as others have mentioned, even with a very low noise floor of say 30dB, that leaves us with a max dynamic range of probably no more than about 70dB.

You might find this thread: 24bit vs 16bit useful, which explains the issues.

G

 

[71 dB]

Hans is one of the worst of the charlatans. He puts just enough technical language around his subjective opinions to bait audiophiles.

This man has produced videos claiming that Ethernet Cables and Ethernet Switches make an audible difference. It takes an amazing lack of technical knowledge to make such a claim.

Best advice - avoid all Hans B. videos - nothing but snake oil wrapped in a phony layer of "science"

Agreed. He talks in a way that makes others easily think what he says is 100 % fact-based unless you know yourself enough to know better. A talented con-man.

As long as people fall for snake-oil there will be snake oil sellers. Unfortunately that's forever, because only some people can be saved from snake-oil scams.

 

[Chimmy9278]

Unfortunately you’ve gone back to linking to BS videos again Although to be fair, it’s more difficult to spot and what he states is largely correct or only slightly wrong. It’s what he fails to mention that makes it BS! Yes, there are artefacts with filters, the pre and post ringing he shows, HOWEVER: He fails to mention that most of the energy of that ringing is near the Nyquist point, so even using a 44.1kHz sample freq the ringing is inaudible. Also, he demonstrates this “filter ringing” using an illegal signal, a signal that never exists in the real world, which exacerbates the effect. Using real world signals, we rarely see this ringing and when we do, it’s very low level, so even more inaudible.

The first pro ADC I bought was nearly 30 years ago, initial sampling was 128 times higher than the 48 or 44.1 sample rates. Modern ADCs operate 4 or more times higher again. This allows a very gentle analogue filter and then a digital “decimation” filter to arrive at the desired sample rate/bit depth.

DACs essentially do the process in reverse, a “reconstruction” filter and then a gentle analogue filter. Standard, cheap, steep linear phase filters are absolutely fine, with inaudible artefacts.

Don’t confuse peak levels with dynamic range. Orchestras can in fact reach much higher than 130dB, I’ve measured up to 138dB but that’s right in the middle of the orchestra, just 2-3 feet in front of the trumpets. In an ideal seat in the audience however, around 100dB or so is about the most you’ll get and as others have mentioned, even with a very low noise floor of say 30dB, that leaves us with a max dynamic range of probably no more than about 70dB.

You might find this thread: 24bit vs 16bit useful, which explains the issues.

G

Well, despite the harshness received, receiving harshness did help me, who was tricked by this con show seeing what is true. Criticism deserved, will continue to link BS videos.

All jokes aside, the 24vs16 is a nice thread, I will give it a thorough read soon. A thing I have in mind is, won't a higher sample rate native file make the over sampler do less work, or make the oversampling more accurate? Since the extra info is recorded instead of artificially over sampled later.

@71 dB I just want to say, I do enjoy music without crossfeed most of the time, except some badly mixed tracks. Prefer the wide sound stage, and less of the squished center image, personally.

 

[gregorio]

A thing I have in mind is, won't a higher sample rate native file make the over sampler do less work, or make the oversampling more accurate? Since the extra info is recorded instead of artificially over sampled later.

If anything, it’s the other way around. With high sample rate recordings you’ve got more data to move around and process more quickly so there’s more chance of error there. Oversampling is trivial computationally, CD players were doing it back in 1984, my first pro DAC was doing 128x oversampling in 1993 and computing power has moved on massively since then. In practice you don’t really get any errors and there’s no accuracy benefit.

G

 

[Chimmy9278]

"The only difference between 16bit and 24bit is 48dB of dynamic range (8bits x 6dB = 48dB) and nothing else." (https://www.head-fi.org/threads/24bit-vs-16bit-the-myth-exploded.415361/).
Since we can only perceive around 70dBs of dynamic range in a live orchestral performance because of noise floor and biological limitations, there is no need for anything more than that.
While tape, reaching 70dBs is already pretty good, 10-11 bits in digital world, CD goes beyond that to 14 bits. Overkill, and ensures you maximum fidelity!

 

[Chimmy9278]

If anything, it’s the other way around. With high sample rate recordings you’ve got more data to move around and process more quickly so there’s more chance of error there. Oversampling is trivial computationally, CD players were doing it back in 1984, my first pro DAC was doing 128x oversampling in 1993 and computing power has moved on massively since then. In practice you don’t really get any errors and there’s no accuracy benefit.

G

I see it in this light, the waveform is already reproduced perfectly, any over sampling done can be accurate based on this inference. While I do not have the computer science knowledge to understand if it is a trivial matter to be done perfectly in real life, the fact that it could be done at such a high rate back then means it is not hard to over sample.

I would not go so far as to say the files generate errors, because that would mean the chip is designed badly in ingressing that data, which the silicone manufacturers should have accounted for.

Rather, since the over sampling process is: implemented at a higher rate than the 192khz most of the time, done based on a perfect model, thus has the possibility to be perfect, there is no need to carry around more data when a smaller amount of it is already sufficient.

This kind of blind test, between 44.1 and 176.4kHz can be done easily in software, and should be done!

 

[gregorio]

Since we can only perceive around 70dBs of dynamic range in a live orchestral performance because of noise floor and biological limitations, there is no need for anything more than that.

In practice, exceedingly few recordings use 70dB of dynamic range, it’s extremely rare to find recordings that use more than 60dB and the vast majority are 50dB or less.

While tape, reaching 70dBs is already pretty good, 10-11 bits in digital world, CD goes beyond that to 14 bits. Overkill, and ensures you maximum fidelity!

The very best studio multi-track tape recorders could manage up to about 80dB. The problem is that the process of mixing and mastering requires “bouncing down” (re-recording) several times and each time you loose another 4-6dB of dynamic range. So you’ll end up with less than 60dB even under ideal conditions. You don’t loose any of this with digital. Your conclusion is correct, 16bit is already overkill.

I would not go so far as to say the files generate errors …

As I mentioned, in practice you don’t really get any errors, although anti-imaging filters can struggle at high sample rates. Typically we are looking for over 100dB attenuation beyond the Nyquist point, with the 192kHz sample rate and higher we often find only 60-80dB attenuation.

G

 

[71 dB]

@71 dB I just want to say, I do enjoy music without crossfeed most of the time, except some badly mixed tracks. Prefer the wide sound stage, and less of the squished center image, personally.

Your comment about crossfeed comes "out of the blue" in this thread, but okay. No problem. Each to their own. Personally I listen to without crossfeed only some "binaural-like" recordings which are a small minority of all recordings out there.

 

[Chimmy9278]

https://i.imgur.com/JmKvMv3.png
https://i.imgur.com/VF6zPq6.png
https://i.imgur.com/YHKjCC5.png

Sorry, was busy and did not have time to properly respond to the articles yet, but came across this on discord. Person was using Foobar ABX plugin and claimed this. Thoughts?

 

[gregorio]

Thoughts?

1. It’s easy to ABX the difference between 16/44.1 and 24/96 or 24/192. Find a very quiet part of the song/track, whack the volume up and listen for the dither noise.
2. Did the tester actually ABX the difference between the different formats or between different masters? This is a very common methodology failure, sighted or ABX.
3. Some other methodology failure, such as improperly converted and edited files, IMD, ruling out chance, etc.
4. The tester is lying or has cheated the ABX test.
5. We cannot absolutely rule out the possibility that the tester did in fact pass a valid test and is super human. Although as they claimed it was “easy”, they would need to be another step beyond super human, mega-human maybe? We’d need some extremely robust, independently scientifically verified evidence to rule out the 4 probabilities above, before even considering the minuscule possibility there are mega-humans amongst us.

G

 

[71 dB]

Thoughts?

The same that has been said a million times before here: If a difference between CD quality and hi-rez is heard, they must be different masters.

 

[Chimmy9278]

1. It’s easy to ABX the difference between 16/44.1 and 24/96 or 24/192. Find a very quiet part of the song/track, whack the volume up and listen for the dither noise.
2. Did the tester actually ABX the difference between the different formats or between different masters? This is a very common methodology failure, sighted or ABX.
3. Some other methodology failure, such as improperly converted and edited files, IMD, ruling out chance, etc.
4. The tester is lying or has cheated the ABX test.
5. We cannot absolutely rule out the possibility that the tester did in fact pass a valid test and is super human. Although as they claimed it was “easy”, they would need to be another step beyond super human, mega-human maybe? We’d need some extremely robust, independently scientifically verified evidence to rule out the 4 probabilities above, before even considering the minuscule possibility there are mega-humans amongst us.

G

Agreed.
1. I don’t think he did that, the plug in just plays the track at one equalized volume.
2. He said he used sox to convert the audio, so they are the same master
3. The plug in is, so far, closed source, so I can’t review its code. Though headfi had a thread on this years ago. (In the wiki) https://wiki.hydrogenaud.io/index.php?title=Foobar2000:Components/ABX_Comparator_(foo_abx)
4.absolutely possible, since it is not peer reviewed or supervised
5. He is confirmed mega-pro human.