Poll: 48kHz vs 96kHz

[Sonik Wiz]

I do 24/96 because that's the resolution I record at. Storage is cheap, why not keep it the same as the source?

In the future we may discover some things we don't yet understand about sound.

I thought you did stuff in 8 bit??

 

[Mr. Afternoon]

In the future we may discover some things we don't yet understand about sound.

"We here at Youth Audio have developed an exciting new way to listen to music. Forget about 8D Audio or MP3, we've invented a revolutionary new format! Half-bit music! Who needs PCM or DSD when you can have pure noise!"

 

[Sonik Wiz]

"We here at Youth Audio have developed an exciting new way to listen to music. Forget about 8D Audio or MP3, we've invented a revolutionary new format! Half-bit music! Who needs PCM or DSD when you can have pure noise!"

I thought Half-bit music was called .5bit DSD?

 

[Electric Moo]

In my case, it's more like "Half-wit" music: Those Orcs were rude, stubborn & disorganized!

 

[Sal1950]

In the future we may discover some things we don't yet understand about sound.

Yep, way past time for the general audiophile public to learn you can't hear the difference in anything better than properly done RedBook

 

[THX1136]

That is an idea that has been floating around forever, it even sounds plausible to me. I don't think that it has ever been proven nor disproven. Likewise the use of less steep reconstruction filters should also have a positive effect.

I would hope that all here would at least agree that while high sample rates such as 96 and 192 kHz might be overkill, their use can only have positive effects (at worse no effect). With hard drive space so cheap and plentiful these days the only good reason to down sample or use lower rates is for system compatibility.

To me it's like headroom when related to the frequencies one can hear from a digital source. Whether one can hear audible effects of higher sampling rates or not does not change the benefit of more headroom, so to speak. One has decreased the possibility of 'low headroom' negatively affecting your listening experience.

 

[THX1136]

Wanted to add, I know that conversion is more solid now than say 30 years ago, but I've always been a bit concerned at sampling at rates that are 'video' related instead of 'audio' related rates. 88.2 makes sense to me as it makes the math for conversion much simpler than converting from 96 to 44.1. It's a circuit design issue sort of from my pov. Simple math make the conversion simple providing less opportunity for rounding errors. Of course if one is staying in the 'video'realm 96 down to 48 is simple math also.

 

[DuncanS]

Wanted to add, I know that conversion is more solid now than say 30 years ago, but I've always been a bit concerned at sampling at rates that are 'video' related instead of 'audio' related rates. 88.2 makes sense to me as it makes the math for conversion much simpler than converting from 96 to 44.1. It's a circuit design issue sort of from my pov. Simple math make the conversion simple providing less opportunity for rounding errors. Of course if one is staying in the 'video'realm 96 down to 48 is simple math also.

The 44.1kHz sample rate for CDs came about because for two reasons, firstly it met Nyquist's sampling theory so could reproduce 20kHz audio, but secondly because they used video machines to produce the master tapes.
Why Are The Sample Rates Based Around 44.1K or 48K? | Production Expert (pro-tools-expert.com)

 

[THX1136]

If you are referring to DAT that's somewhat true and somewhat not true. Consumer machines were limited to only being able to record at a sampling rate of 32k or 48k. This was an effort to assuage the industry's fears of average folks like you and I making millions from duplicating our CDs and selling them to others on DAT tapes. The first consumer machines could only playback at 44.1k, but not record at that rate. That allowed the owner to play back all those thousands of DAT tapes the industry was going to produce (and didn't). Pro DAT allowed recording at 44.1k. In this instance it had little to do with video making it unique to DAT as afar as sampling rates go.

I remember back in the day that there was a way to utilize a VCR with an add-on converter that would allow digital recording, but my memory of the details is not up to snuff to remember much more. That may be what you are referring to which makes sense. The first 'pro' multitrack was 3M's 32 track beast which included a mastering portion as part of the packge, but I'm remembering it was a stationary head machine, not spinning head like the video machines were predominantly. That was a 16 bit 50k machine. As there were no 16 bit converters at the time, they used a 12 and 8 bit together to accomplish 16 bit performance. (Among the first notable releases using the 3M system were The Nightfly and Ry Cooder's Bop Til You Drop.) I recall the spinning head used in video to be a major hurdle that the industry was trying to get around - a stationary head was the goal at that time which I believe the 3M machine accomplished.

In the second edition of of Ken Pohlman's excellent book entitled 'Principals of Digital Audio' (1989) he made the point that the current technology was limited by what was possible at a cost effective level. High sampling rates were easier to achieve as it was easier to manipulate time as opposed to bit length of converters to come to a better quality of recording/playback systems. This was at the beginning of the idea on 1 bit which theoretically ran at very high sample rates to achieve a more accurate 'recording' of an audio signal. Plainly, it was easier to manipulate time. The more samples, the closer to the analog signal one was recording or playing back.

A lot of this is fuzzy for me as it's been literally years (30+) since I knew this stuff in more detail. You are absolutely right about the Nyquist theory of course. When all the other aspects came into play like error correction and the like, there was a lot of things to be considered when designing digital audio devices to record and playback what we heard in the analog world.

Thanks for your comment Duncan! Always good to stir up thought and talk about this stuff.

 

[DuncanS]

I'm referring to the relationship between the frame rate & no. Of lines per frame, no. Of samples per line, as this set the sample rate at 44.1kHz. They used VHS tapes to transfer the master recordings to the pressing plant.
US video 3x samples per line x 490 lines x 30 frames per second = 44.1kHz
UK video 3x samples per line × 588 lines x25 frames per second = 44.1kHz

 

[THX1136]

Thanks for the clarification, Duncan! Went back and reviewed the article you referenced. I was unfamiliar with that aspect of things, being only aware of DAT and, later, linear tape like the 3M system. Learn something all the time.

 

[THX1136]

Just took the time to check out an article published via the AES on the history of the early days of digital recording. Funny how stuff comes back when one sees the mention. The Sony 1600/10/30 UMatic VTRs were used for mastering many early CD releases (obviously what the author of the earlier article was referring to which you post the link to). Also a mention of the PCM-F1 which was the device I mentioned in the earlier post. Thanks again, Duncan! You got me digging into history which brought back stuff I knew and had forgotten. Enjoyed the exchange, sir!

Article link: https://www.aes.org/aeshc/pdf/fine_dawn-of-digital.pdf

 

[ssully]

I do 24/96 because that's the resolution I record at. Storage is cheap, why not keep it the same as the source?

In the future we may discover some things we don't yet understand about sound.

Sure. By that token, we may discover that hi rez is detrimental.


Anyway rather than hypotheticals, there is one 'known' reason to capture analog tape sources at hi rez. That's if useful information exists as a signal out of the audible range. Plangent processing uses high frequency carrier signals in analog sources (e.g. bias signal in tape) to identify and correct speed drift, wow, and flutter artifacts. (Whether the result is worth the effort, I leave to the listener.)

Similarly, if a matrixed quad source employs a carrier signal, you'll want to capture that.

Not because you can hear it, though.

 

[Electric Moo]

I've been quoted a couple times so I sorta feel obligated to reply: I'm not interested in arguing about CD vs hires audio, if that's what @ssully & @Sal1950 are aiming to do?

I listen to & enjoy CD's. If you believe digital audio peaked in 1982 & there's nothing new to discover about sound in the future, that's fine.

I stand by my statements though. I record at 24/96, don't use analog tape much ( although that might change if I get a Studer A800 ) and prefer to keep the source resolution. Why even mess with dithering, filter slopes, Nyquist, conversions etc if it's not necessary? It's also another step where data could get corrupted.

 

[boxhead]

I'm not suggesting that, but they affect sound waves within the audible spectrum, affecting what one does hear.

I have wondered about this for a long time. Logic suggests that we don't hear anything above say 20k, but what impact, if any, do ultrasonics have on sounds within the audible spectrum? Ssully suggests that "You aren't hearing any of those ultrasonics" in reference to a violin, rather that "If you *hear* something from your playback system when it's playing ultrasonic frequency content, it is distortion in the audible range caused by the presence of content your playback system was not designed to handle."

I did an experiment using Adobe Audition, but any similar program could probably do the same. When I mention square waves below, please understand that I know that music does not usually contain square waves (some electronic music notwithstanding), but the principle of musical harmonics and resulting waveforms can easily be observed using these square waves.

Firstly, I created a 5kHz square wave at 44.1/16. I then created a second 5kHz square wave at 96/24. The two samples sound very different. If you zoom in on these samples you can also see a difference, with the 96/24 sample being a much closer approximation of a square wave. A square wave consists of a base frequency sine wave with ideally an infinite number of even harmonics. The more (and higher frequency) the harmonics the closer the waveform approximates a real square.

So why do they sound different? I'm open to suggestions, but my initial observation is the the added harmonics allowed due to the higher 96kHz sample rate are impacting how the waveform sounds in the audible spectrum. Of course it may be something else altogether....

 

[tonyE]

Firstly, I created a 5kHz square wave at 44.1/16. I then created a second 5kHz square wave at 96/24. The two samples sound very different. If you zoom in on these samples you can also see a difference, with the 96/24 sample being a much closer approximation of a square wave. A square wave consists of a base frequency sine wave with ideally an infinite number of even harmonics. The more (and higher frequency) the harmonics the closer the waveform approximates a real square.

Actually, I believe the square wave is made with the fundamental and odd harmonics, note the (2k-1) term:

Using Fourier expansion with cycle frequency f over time t, an ideal square wave with an amplitude of 1 can be represented as an infinite sum of sinusoidal waves:

 

[boxhead]

Actually, I believe the square wave is made with the fundamental and odd harmonics, note the (2k-1) term:

Using Fourier expansion with cycle frequency f over time t, an ideal square wave with an amplitude of 1 can be represented as an infinite sum of sinusoidal waves:

Quite right - got my odds and evens mixed up!

 

[kap'n krunch]

Actually, I believe the square wave is made with the fundamental and odd harmonics, note the (2k-1) term:

Using Fourier expansion with cycle frequency f over time t, an ideal square wave with an amplitude of 1 can be represented as an infinite sum of sinusoidal waves:

shhhhhhhhhhhhhhhhh (sounds like white noise to me!)
oh, man, I used to be a god in the acoustics class... now I don't remember diddley squat!...oh well...

 

[jimfisheye]

So why do they sound different? I'm open to suggestions, but my initial observation is the the added harmonics allowed due to the higher 96kHz sample rate are impacting how the waveform sounds in the audible spectrum. Of course it may be something else altogether....

I feel like this is one of those trick engineering questions I should know the answer to!

It has to be an artifact of the system and/or analysis. I don't say that to be dismissive. I know we can take the final result audio (ie the mix) and convert it back and forth and not hear any gross change. Which case is the artifact? Are we hearing something skewed from aliasing in one of those test cases?

Maybe if I try to state this thought:

Digital has a weirdness around the edges. Analog fades out into noise around the boundaries of the system. Quiet levels fade down into hiss, for example. Our digital recording system has a data window where everything is made to be linear but it truncates at the edges. Instead of sound getting noise or muddy around the limits, we get weird artifacts. Pixilation kind of stuff.

So I'm looking for an explanation with that in mind.

 

[ssully]

Firstly, I created a 5kHz square wave at 44.1/16. I then created a second 5kHz square wave at 96/24. The two samples sound very different. If you zoom in on these samples you can also see a difference, with the 96/24 sample being a much closer approximation of a square wave. A square wave consists of a base frequency sine wave with ideally an infinite number of even harmonics. The more (and higher frequency) the harmonics the closer the waveform approximates a real square.

So why do they sound different? I'm open to suggestions, but my initial observation is the the added harmonics allowed due to the higher 96kHz sample rate are impacting how the waveform sounds in the audible spectrum. Of course it may be something else altogether....

Keep in mind that any 'square wave' your brain processes is a 20kHz lowpassed version, no longer square. So their different appearance, particularly the 96kHz version, doesn't mean much. It absolutely 'should' look a lot closer to a perfect square wave than the 44kHz (i.e.,. 'lowpassed' at 22 kHz) waveform.

I didn't say that ultrasonic harmonics can't impact the audible range. If ultrasonic content is causing your playback setup to distort, that distortion can have components in the audible range.

Increasing sampling bandwidth (e.g. from 22 to 48 kHz as you've done by increasing sample rate from 44 to 96) does not inherently improve the digitization of content below the lower bandwidth. It just captures more (higher) frequencies. That's just Nyquist. It's possibly the hardest aspect to get one's head around. There are caveats involving antialiasing/imaging filtering, but if done right this should not cause the 'very different' audio you are hearing.

Beyond that, you'd need to supply lots more detail , and the files themselves, to diagnose what's causing the artifact you are hearing.