Taperssection.com

Gear / Technical Help => Ask The Tapers => Topic started by: checht on September 13, 2020, 06:19:20 PM

Title: 48 kHz vs 44.1 kHz sample rate
Post by: checht on September 13, 2020, 06:19:20 PM
Recently had a bit of time to think through my recording practices 😀. A few years ago I switched to recording 24/48 rather than 24/96, feeling like the trade offs made sense for me.

Now I’m wondering about 44.1. Mostly, trying to figure out if there’s a significant quality difference from 48 kHz, one that’s large enough to offset any quality reduction from downsampling. I’m unclear on quality difference between the rates, and whether downsampling algorithms degrade sound quality.

Thoughts? What else should I be considering?
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: DSatz on September 13, 2020, 08:34:59 PM
Are you asking about down-sampling some existing 48 kHz recordings to 44.1 kHz? There could be reasons to do that, but I wouldn't do it unless it was necessary, e.g. if a bunch of your friends unearthed a trove of CD players, and wanted to throw a 1980s party and listen to your recordings that way.

As far as 44.1 vs. 48 kHz for an original recording is concerned, there shouldn't be enough sonic difference to be detectable by a human listener. If there is, the equipment (or your process of making the judgment) is suspect IMO. Once the sampling rate exceeds 2x the highest frequency being sampled, the only audible differences should be in the side effects of the anti-imaging and anti-aliasing filters. The whole decades-long saga over sampling rates is really about filtering, not sampling. In general there are tradeoffs between the frequency-domain and time-domain behavior of any filter (analog or digital). There are dynamic range and distortion considerations as well.

For digital filters, some important optimization constraints relax if the sampling frequency is raised. But the <10% difference between 44.1 and 48 kHz isn't enough to matter in that way. You'd need to go to (maybe) 60 or 64 kHz before you're totally "in the clear" for real-world program material. Thus we have 96 kHz as a professional standard--plus the usual assortment of people who think that no sampling rate is ever high enough, because they misunderstand sampling theory, i.e. they imagine that it becomes "closer to analog" the higher you raise the rate, which isn't how digital audio works.

All the potential sonic problems of filters become less if their design is less aggressive (fewer "poles"). Few live, acoustic signals have significant energy at 20 kHz or above, and those that do are rarely recorded close-up by consumers using microphones capable of conveying 20+ kHz signal components. So as a generality, there should be less need for aggressive filtering, and correspondingly less "need" for 96 kHz sampling in consumer recording equipment (i.e. to shove the filter problems brute-force up out of the audible range). For better or worse, though, manufacturers tend to design equipment for worst-case scenarios, and unfortunately, this aspect of recording equipment, though readily measurable, isn't usually described in spec sheets or on-line reviews.

--best regards
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: jerryfreak on September 13, 2020, 09:25:19 PM
id go by end rate of your project

id take native 44.1 over 48K resampled to 44.1 personally

Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Sebastian on September 14, 2020, 04:24:24 AM
I went from 96 kHz to 48 to 44.1 in the last couple of years. 96 kHz just took up too much space and required too much processing time with no audible benefits. And since most of my recording have 16 bit/44.1 kHz as their final format, it made sense to me to also record in that format. And I can avoid resampling this way.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: checht on September 14, 2020, 08:40:38 AM
I went from 96 kHz to 48 to 44.1 in the last couple of years. 96 kHz just took up too much space and required too much processing time with no audible benefits. And since most of my recording have 16 bit/44.1 kHz as their final format, it made sense to me to also record in that format. And I can avoid resampling this way.

This was my contemplated path prior to starting the thread. Now reconsidering my practice of releasing 16/44.1 versions. DSatz pointed out my logical error; not likely that anyone's making optical discs any more, except of those vintage Duran Duran shows...

Hmmm, maybe 16/48 releases so they're easy to sync with video, which does happen once in a while.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: rippleish20 on September 14, 2020, 10:12:11 AM
I record and post at 24/48 ; it seems to me i nthis day and age people should be able to handle this...
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Papaphunk on September 14, 2020, 10:20:05 AM
I record in 24 Bit/44.1 gives me about 2 hours 15 minutes recording time for each set. 24/48 gives me about 2 hours 5 minutes, and alleviates the headaches of having to add on that extra encore to the 24/48 recording I've found. Relistening difference is negligble imho.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: heathen on September 14, 2020, 11:07:46 AM
I record and post at 24/48 ; it seems to me i nthis day and age people should be able to handle this...

Up until recently I was distributing everything in 16/44.1 (recording in 24/48), but then I came to the realization that you're pointing out.  I record in 24/48 and I'll generally distribute stuff in 16/48.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: EmRR on September 14, 2020, 11:33:47 AM
Is anyone still burning CD's?!?  That's the only reason I can think to stick with 44K1.  48 works with video and all media players. 

Following on what DSatz said, there IS a lot of suspect equipment out there, very EXPENSIVE suspect equipment you wouldn't suspect of being.....suspect.  I've heard lots of cases where cymbals and acoustic instrument treble was noticeably better at 48 over 44K1, observing the long pattern results from multiple recording sessions with the same equipment, different sample rates.  As to high rates, I used to have multiple converters that clearly sounded their best at 88K2 or higher, now there's much less observable difference.   

I worked at 44K1 for years when everything was going to CD, as it allowed me to mix directly to the same tape medium with the mix in sync with the multitrack.  That was the ONLY reason I ever stuck with 44K1.  I've worked at 88K2 or 96K for 14 years now.  In my few interactions with record labels, they all want 96K files. 
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: morst on September 14, 2020, 12:21:22 PM
44.1 is for compatibility with wasteful and outmoded plastic.
Video Sync is easiest at 48k.
If your target audience includes dogs and birds, or the intended use of your recordings is to pitch shift them downwards more than one octave, then I suggest using f(s) rates higher than 48k, else not.
I have finally settled at recording at 24/48 and distribution of finalized work at 16/48, as none of my recordings exceed 96dB s/n.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Sebastian on September 14, 2020, 12:49:18 PM
This was my contemplated path prior to starting the thread. Now reconsidering my practice of releasing 16/44.1 versions. DSatz pointed out my logical error; not likely that anyone's making optical discs any more, except of those vintage Duran Duran shows...

Hmmm, maybe 16/48 releases so they're easy to sync with video, which does happen once in a while.

I guess there is no right or wrong here. Everybody is entitled to do with their recordings whatever they want. I chose 44.1kHz because it's the sample rate that is compatible with the most playback hardware, including CDs. Even though I haven't burned a CDR in over a decade myself. And since I couldn't care less about video, I'm just sticking to the audio standard.

Following on what DSatz said, there IS a lot of suspect equipment out there, very EXPENSIVE suspect equipment you wouldn't suspect of being.....suspect.  I've heard lots of cases where cymbals and acoustic instrument treble was noticeably better at 48 over 44K1, observing the long pattern results from multiple recording sessions with the same equipment, different sample rates.  As to high rates, I used to have multiple converters that clearly sounded their best at 88K2 or higher, now there's much less observable difference.

I guess my initial point still stands: If it sounds fine on my playback equipment at 44.1kHz, I'll continue putting out (and recording) my tapes in that format. Why should I care about some suspect equipment I don't even own?

Caveat emptor: I have been to way too many loud shows to hear a difference between all those formats, so maybe that's why I couldn't care less about all those supposedly better sampling rates. ;) Your hearing may be better than mine. In this case: more power to you! :)
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: goodcooker on September 15, 2020, 10:27:41 AM

When I got my first "real" recording deck, a Marantz PMD670, it topped out at 16/48. Many people had already moved on to 24 bit and much higher sampling rates but memory cards and storage drives were still expensive. My next recorder was 24 bit and I stuck with 48kHz for a while than moved it down to 44.1. I'm still there.

I can't tell a difference and I have pretty good playback gear. Not entirely convinced any longer that processing sampling rate conversions would have more adverse affect than just going with the lower sampling rate to begin with.

If I'm recording audio and I know for sure that it is to be used in a video I'll switch to 48. If I was recording audio for a professional release I would go to 96 to give the post production folks downstream more resolution.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: voltronic on September 15, 2020, 12:50:27 PM
My default has been 48 kHz for a long time.  The video sync is a big reason for that.  Even if you aren't doing a camera audio replacement, you may find yourself doing an edit of someone's phone video with some music you have previously recorded, or you want to lay down tracks with a friend who lives far away and they only have a phone or tablet to record on.  Those devices are going to be running at 48k by default, so for maximum compatibility with whatever my tracks might ever get combined with, that's the sample rate I choose.  Having 44.1 and 48 k sources in the same project can get hairy.

For projects that I'm doing 100% at home in my DAW that have a combination of live and MIDI-fed VSTs, I run at 96 kHz for the greater precision, but final delivery is 48 kHz.  Honestly, 96 kHz is the default for new projects in my DAW (Reaper) and I never cared to change it.

44.1 kHz or a multiple of it is what you should choose if you are going to deliver to CD because of the reduced risk of SRC errors.  (This is probably less of a problem as it once was.)  If you're not making CDs, I would go with 48 kHz.

I have never heard a sonic benefit to acoustic recordings with sample rates higher than 48 kHz, so I don't use them.  I even have tracks I have bought from high-res online music stores that were delivered at 96 or 192 kHz, and I downsampled them to 48 kHz to save hard drive space.  The funny thing is if you put a lot of those tracks into a spectrogram, you'll find that they may have been recorded at a much lower rate and then upsampled! (caveat emptor)

As far as bit depth, I would avoid 16-bit for tracking, unless you have no other choice.  There's no reason not to use 24-bit and its wider dynamic range.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: jerryfreak on September 15, 2020, 01:17:34 PM
As far as bit depth, I would avoid 16-bit for tracking, unless you have no other choice.  There's no reason not to use 24-bit and its wider dynamic range.

you can make equally good recordings with 16 bit, but with good equipment, 24-bit is useful as you can achieve comparable results without riding the levels as much, which is risky
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: EmRR on September 15, 2020, 06:51:56 PM
Any of the MOTU converters before the 2014 updates, for example, sound obviously worse up top at 48 and under.  Conversely, the Aurora converters of then and earlier apparently did sound best at 48. 
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: voltronic on September 15, 2020, 08:23:48 PM
As far as bit depth, I would avoid 16-bit for tracking, unless you have no other choice.  There's no reason not to use 24-bit and its wider dynamic range.

you can make equally good recordings with 16 bit, but with good equipment, 24-bit is useful as you can achieve comparable results without riding the levels as much, which is risky

Definitely.  24-bit just gives you a bit more margin of error on both ends of the dynamic range, so why not take advantage?
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Chrisedge on September 20, 2020, 01:24:16 PM
I use 48/24 since I do a lot of video stuff too. Just made sense from the BD days and having the same sampling rate as BD.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: guitard on September 21, 2020, 06:56:39 PM
I use 48/24 since I do a lot of video stuff too. Just made sense from the BD days and having the same sampling rate as BD.

I also do a lot of video.  I tend to release audio a lot more than I do my videos.  So I usually record in 44.1/24, release the audio, and then convert to 48 and sync with the video.  I don't notice a difference in audio quality after converting to 48, so I stick with this work flow.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 21, 2020, 08:42:30 PM
I want to go against the grain a bit here for what's being said. Before we get too bogged down on a few topics here, my thesis statement: you should record in as high a digital fidelity (both bit depth and sample rate) as it is convenient for you to store. If you are recording to massive internal hard drives or digital memory, you have little reason not to record at 24/96.

There are a lot of misconceptions about digital audio, stemming from its misrepresentation to the public as well as its misinterpretation by the public. It's a bit esoteric from the academic side, but that's because to really understand what these numbers like bit depth and sample rate really mean requires a few semesters of calculus plus a course in discrete math and another in linear algebra, then you need a course in continuous-time (analog) and discrete-time (digital) signal processing and information theory to understand the why. And in the other direction, academia is pretty deaf to the questions, criticisms, and comments from the layperson - for example, a claim like "I don't hear a difference between 16/44.1 and 24/96" is rarely met with anything less than scoffs from engineers.

Before going further, a few points I want to highlight from this thread:

Quote from: checht
I’m unclear on quality difference between the rates, and whether downsampling algorithms degrade sound quality.

A downsampling algorithm when used incorrectly can induce distortion in the phase of the signal for non-integer downsample rates, and you might perceive that as less treble detail, a shift in soundstage (or worse - a constantly shifting soundstage for some instruments), or some level of "splatter" and "time smearing" - sorta like those cheap 90s karaoke microphones that sounded like they had a spring inside of them.

If you are doing integer down-sampling of 2:1 or 4:1 - so for example, going down to 48 kHz from 96 kHz or 192 kHz sample rates - and your down sampler is just throwing out every other sample rather than doing some form of linear phase interpolation, then you should hear no discernible difference between the higher sample rate and the lower sample rate of the exact same audio content, assuming the digital-to-analog converter (DAC) performs identically at all sample rates.

Quote from: DSatz
As far as 44.1 vs. 48 kHz for an original recording is concerned, there shouldn't be enough sonic difference to be detectable by a human listener. If there is, the equipment (or your process of making the judgment) is suspect IMO. Once the sampling rate exceeds 2x the highest frequency being sampled, the only audible differences should be in the side effects of the anti-imaging and anti-aliasing filters. The whole decades-long saga over sampling rates is really about filtering, not sampling. In general there are tradeoffs between the frequency-domain and time-domain behavior of any filter (analog or digital). There are dynamic range and distortion considerations as well.

For digital filters, some important optimization constraints relax if the sampling frequency is raised. But the <10% difference between 44.1 and 48 kHz isn't enough to matter in that way. You'd need to go to (maybe) 60 or 64 kHz before you're totally "in the clear" for real-world program material. Thus we have 96 kHz as a professional standard--plus the usual assortment of people who think that no sampling rate is ever high enough, because they misunderstand sampling theory, i.e. they imagine that it becomes "closer to analog" the higher you raise the rate, which isn't how digital audio works.

All the potential sonic problems of filters become less if their design is less aggressive (fewer "poles"). Few live, acoustic signals have significant energy at 20 kHz or above, and those that do are rarely recorded close-up by consumers using microphones capable of conveying 20+ kHz signal components. So as a generality, there should be less need for aggressive filtering, and correspondingly less "need" for 96 kHz sampling in consumer recording equipment (i.e. to shove the filter problems brute-force up out of the audible range). For better or worse, though, manufacturers tend to design equipment for worst-case scenarios, and unfortunately, this aspect of recording equipment, though readily measurable, isn't usually described in spec sheets or on-line reviews.

One reason I love DSatz - that first paragraph is BANG right on the money.

To expand on the second paragraph, I want to clear the air here once and for all: why 48 kHz vs 44.1? Or why 96 vs 88.2? Well in the 80s when much of digital media was till the Wild West, we established independent standards for audio and video - rightfully so, as they are two very different media. Digital sampling theory (going back to Nyquist-Shannon in Nineteen Twenty Fucking Seven - it always blows my mind how ahead of the curve the paper these guys' wrote really is) tells us a few things. Namely: if we have continuous, analog media, then we capture it at discrete intervals and then try to reproduce it accurately and without distortion, then we must capture data at a rate at least twice the highest frequency (or twice as fast of a period, in the time domain) as the fastest signal contained within the original continuous media. The top range of healthy, young human listening is approximately 20 kHz; so per Nyquist-Shannon, we need to capture data at a rate of at least 40 kHz. As David points out, the implications of needing to go from digital to analog when incorporating "real-world/environmental" factors are that we can't design a perfect re-constructor, so we over-sample a little bit at 44.1 kHz to account for this. We call this 44.1 kHz sample rate 'red book" since it was possible in the 80s to produce converters between the digital and analog domain relatively inexpensively at 44.1 kHz - any faster, and you need a more robust (and thus more expensive) converter.

48 kHz comes from the video world - playing by the same rules, video engineers' standard is to capture data at 24 frames per second. For those of you playing along at home, I want to highlight this mismatch not just in number, but also in magnitude, since I fight video engineers regularly about this point and why audio processing is such a bigger and more ignored challenge than video processing.

Audio: 44,100 Hz
Video:        24 Hz

Yeah. Digital audio filtering - especially if you want to do it in real time, or "on the fly" without inducing any lag - is a helluva lot harder.

But where was I... Ah, RIGHT! Well, as you've gathered, resampling algorithms are a MASSIVE can of worms, and aren't always guaranteed to work sans phase distortion, unless the re-sample ratio is an integer multiple.

I think you can see where this is going.

48 kHz became a standard in audio, because it's just flat-out easier to sync up with video feeds. The numbers 24 and 48000, even if off by a few magnitudes of each other, are a nice, elegant, integer ratio. So for that reason alone, 48 kHz became another standard used around audio and to this day it tends to be a common format for video and gaming platforms.

44.1 is a carryover from when digital audio first became a thing in the 80s - it was the standard, and today we just have enough knowledge of how signals operate at this sample rate - and it's even part of many digital audio standards - that we frequently still record, process, and listen to music at 44.1 kHz. 48 is slowly taking over, especially as mobile phone DACs move to 48 for standardization across video platforms like people watching YouTube and Netflix on their phones. One isn't better than the other inherently, it's down to what you prefer and if you work with video engineers.

For the last paragraph above, DSatz says "All the potential sonic problems of filters become less if their design is less aggressive (fewer "poles")." This one is an attempt at a broad generalization that contains a lot of caveats to really say; it's in the right ballpark, but given the layperson's penchant to misinterpret the implications of this sentence I'd flag it as "with conditions."

Quote from: jerryfreak
id take native 44.1 over 48K resampled to 44.1 personally

Again, that's heavily dependent on the resampling algorithm and method used. If it's a pretty high order linear phase filter, I think you'd be hard-pressed to find a massive difference between native 44.1 kHz content and properly resampled 48 kHz down to 44.1 kHz content on the same converter.

Quote from: EmRR
Is anyone still burning CD's?!?

Following on what DSatz said, there IS a lot of suspect equipment out there, very EXPENSIVE suspect equipment you wouldn't suspect of being.....suspect.  I've heard lots of cases where cymbals and acoustic instrument treble was noticeably better at 48 over 44K1, observing the long pattern results from multiple recording sessions with the same equipment, different sample rates.  As to high rates, I used to have multiple converters that clearly sounded their best at 88K2 or higher, now there's much less observable difference.   

Yep - me. Though I am admittedly pretty niche. Speaking of which, if anyone has old music CDs officially released by artists, or blank CD-Rs taking up space in their home, drop me a line - I'll take 'em off your hands.

Regarding that second paragraph - this more likely has to do with how the actual converters were designed - the analog circuitry around them is likely more optimized to work at 48 rather than 44.1 kHz with lower distortion, and that's likely what you're hearing more so than any property of the sample rate of the file.

Quote from: morst
If your target audience includes dogs and birds, or the intended use of your recordings is to pitch shift them downwards more than one octave, then I suggest using f(s) rates higher than 48k, else not.
I have finally settled at recording at 24/48 and distribution of finalized work at 16/48, as none of my recordings exceed 96dB s/n.

Quick side tangent: there are lots of signal processing techniques we can use to get around the octave limit for pitch shifting up or down outside of oversampled material. At the risk of accidentally breaking some NDAs, I'll just comment that companies have been doing pitch shifting of significant ratios since the late 70s, and getting it down pretty clean (even by 2020 standards) by the 80s. What we can do today with pitch shifting is really darn cool, but the methods to clean it up are held pretty tight by industry since a good pitch shifter is a lucrative algorithm.

The mention of bit depth and SNR of 96 dB is useful, but I'll loop back around to that later when I discuss why I think 24 is better than 16 for recording.

Quote from: goodcooker
I can't tell a difference and I have pretty good playback gear.

You had me at the first half, and lost me on the second. At the end of the day, this really boils down to if YOU can hear a difference. That's all that should matter. But you can't generalize your hearing and your playback equipment (as good as it may be) to everyone and every playback rig on this particular note, even the very expensive ones - is your converter designed to run with lower distortion at one clock rate over another?

Quote from: jerryfreak
you can make equally good recordings with 16 bit, but with good equipment, 24-bit is useful as you can achieve comparable results without riding the levels as much, which is risky

Yep, one of a few great reasons to record at 24 bits. The extra headroom to avoid clipping is a great perk.

Quote from: EmRR
Any of the MOTU converters before the 2014 updates, for example, sound obviously worse up top at 48 and under.  Conversely, the Aurora converters of then and earlier apparently did sound best at 48.

Yep, sounds about right - the converter circuitry is probably optimized for lowest distortion at 48 kHz, knowing that's likely to be the most commonly used. FWIW, I've never met a MOTU product I liked so this doesn't surprise me, but I want to avoid bashing them - I'll just say I've never had one of their units work easily or correctly as-advertised in any system I've worked on.






ANYWAY... I mentioned you should record as high as possible for both bit depth and sample rate. So, why do I recommend you do that? Even if you can't hear a huge difference?

Post production

I come to taping from studio production. I started recording my own music on a computer in high school, then I started producing in college as I learned more about digital filtering and computer music. Eventually friends asked me to produce for them too as they liked my production sound, and I've been down the digital filter wormhole ever since. Eventually, before a show one day I just walked up to the tapers' section and happened across some very kind faces who answered all my questions, and I leveraged my production chops to start doing remix and mastering work.

The way digital filters work, ESPECIALLY a digital filter emulating an older analog filter - they approximate. They can internally upsample to help improve their performance and there's more wizardry under the hood we can run there, but even still; the higher the sample rate, the more precise and accurate the filter sounds. This just sounds SO much cleaner, better, and brighter. This filter accuracy CAN be heard both at the higher sample rates at 96 kHz, as well as after that 96 kHz master has been down-converted. It's not as subtle as you'd think, it's very audible.

A good way to think about this: if a filter is a knife, the higher your bit depth the more precise your cut location, and the higher your sample rate the sharper your knife is. Both have implications for how accurately your knife is cutting - correct location AND the shape of the cut, straight lines are straighter and curved lines are closer to the curve's spec.

I encourage you to try this out for yourself, and trust your own ears. That's a better sell for the general idea for this than any amount of math and numbers I could throw at you. I've been meaning to make a comp of this to help prove my point in general to tapers, maybe this is good motivation for me to set aside some time in the next month or two and make a comp of A-B for both raw tapes AND tapes with processing...

So take it from me. You might not hear a huge difference today... but do you know how much I would MURDER for 24/96 recordings of my favorite Phish and Dead shows from the 90s? And I wasn't a thought on any taper's mind back then.

A few more points, so I can wrap this up for the moment and cook dinner:

-I never went deep into bit depth and why it's important. A quick note: if you are adding or multiplying two digital numbers together, the more bits you have the more accurate your addition (and that is the fundamental operation of filtering - delaying, multiplying, adding).

-How much more space are these files taking for you? Digital storage is cheap.

-Sure it takes more work for you to arrange and join files (especially if you record more than two channels), but the extra bit of work will be worth it. I think 24/192 is a bit unwieldy given WAV's 4 gig file size limit so needing to splice a set even for just two channels can be frustrating, plus if I'm recording an entire run of shows those file sizes do add up quickly to fill a 64 gig SD card. HENCE, my statement at the beginning: record in as high a fidelity as it is convenient for you. For me, that's 24/96.

-For the raw tape, if you're doing your re-sampling and bit reduction correctly you shouldn't hear a significant difference. I have been double blinded and I consistently sniff higher resolution files from a mile away, but to my ear the differences in the raw files with no processing applied are slight, and I'm not convinced I'm hearing any benefit from the file instead of converters designed to operate at higher sample rates. My personal DAC at home does internal upsampling to a ridiculous degree and domain anyway, so 16/44.1 usually works just as well as 24/96 for playback for me. AS SUCH, I personally record at 24/96, do any processing at that rate, and then bounce down to 16/44.1 or 16/48 for release.

If anyone has questions, or criticisms of any of this: fire away.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: justink on September 21, 2020, 10:17:42 PM
I record and post at 24/48 ; it seems to me i nthis day and age people should be able to handle this...

same.  I've been recording 24/48 and only posting the same for 10 years now. 

cars don't even have cd players anymore.  so 16/44.1 is meaningless.

24 bit recording has plenty of benefits (even more in 32bit FP). 

just post 24/48 and make it easy on yourself.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: heathen on September 22, 2020, 10:23:52 AM
If anyone has questions, or criticisms of any of this: fire away.

Thanks for taking the time and effort to drop serious knowledge on us.  Very much appreciated!

Question: Other than file size and CD compatibility, is there any reason to convert from 24 bit to 16 bit for the "released" version of a recording?
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 22, 2020, 11:56:26 AM
Thanks for taking the time and effort to drop serious knowledge on us.  Very much appreciated!

Question: Other than file size and CD compatibility, is there any reason to convert from 24 bit to 16 bit for the "released" version of a recording?

Cheers, mate! Glad it’s appreciated.

There are some reasons. Off the top of my head right now, pre-coffee:
-I’ve seen data sheets for some mobile DACs (designed for phones, tablets, laptops, cars, etc) where the DAC has better performance with lower distortion at 16 bits as opposed to 24 bits. Given how much media is consumed these days on mobile DACs, it’s a possibility that many people are listening on such DACs, and those with mobile DACs tuned for 24 bits wouldn’t be too bothered by the “hit in performance” of their DAC running at 16 bits, compared to the other way around.
-Speed and likelihood of distribution. If you’re trying to be quick with a release after a show (as I often am), a smaller file size set means it takes less time for people to download your show, and dissemination to the public happens a lot faster.

I’m sure there’s more too, but a healthy dose of caffeine would be needed first.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Gutbucket on September 22, 2020, 12:35:55 PM
Got coffee yet?

On sampling rate-
Post production

[..snip..]
The way digital filters work, ESPECIALLY a digital filter emulating an older analog filter - they approximate. They can internally upsample to help improve their performance and there's more wizardry under the hood we can run there, but even still; the higher the sample rate, the more precise and accurate the filter sounds. This just sounds SO much cleaner, better, and brighter. This filter accuracy CAN be heard both at the higher sample rates at 96 kHz, as well as after that 96 kHz master has been down-converted. It's not as subtle as you'd think, it's very audible.

[..snip..] I've been meaning to make a comp of this to help prove my point in general to tapers, maybe this is good motivation for me to set aside some time in the next month or two and make a comp of A-B for both raw tapes AND tapes with processing...

Yes please! I strongly encourage you to do this!

I ask because I long ago decided to record in 24/48, primarily for storage space reasons.  Storage cost continues to drop, yet there is also the management of accumulated storage to consider.  When recording 4 to 8 channels of raw data every outing, archiving the raw files, the fully edited output, and perhaps an intermediary working stage or two.. over the course of decades of taping, things add up.  Twice as much is significant regardless of storage cost.  Granted I made that decision before 96kHz became the defacto standard it now is, but I have a hard time justifying doubling the required storage space for what seems a rather incremental gain - one I don't doubt is real, yet have not clearly heard in the admittedly few recordings I made when determining what standard recording rate I would use (granted, those recordings were not post processed).  Still, I'd consider recording at 96kHz for some things if I knew it would make a significant difference in my recordings, either now or in the future.  I've yet to convince myself that's worthwhile, but remain open to changing my mind.

Quote
The way digital filters work, ESPECIALLY a digital filter emulating an older analog filter - they approximate. They can internally upsample to help improve their performance and there's more wizardry under the hood we can run there, but even still; the higher the sample rate, the more precise and accurate the filter sounds. This just sounds SO much cleaner, better, and brighter. This filter accuracy CAN be heard both at the higher sample rates at 96 kHz, as well as after that 96 kHz master has been down-converted. It's not as subtle as you'd think, it's very audible.

Does it make any difference if the the entire raw recording is upsampled upon loading in the editor prior to filtering or is doing so essentially the same achieved by upsampling within the filtering being applied, except being less efficient?


On bit-depth-
Quote
A quick note: if you are adding or multiplying two digital numbers together, the more bits you have the more accurate your addition (and that is the fundamental operation of filtering - delaying, multiplying, adding)

Isn't the provision of sufficient space for calculation accuracy a separate issue from the space required to capture the full dynamic range of during the initial sampling?  I ask because I think it is important to clearly differentiate storage space requirements from processing space requirements, and editing software is working natively at higher bit-depths. Are mathematical operations which require additional space to avoid rounding errors and preserve accuracy conducted as part of the initial sampling in the ADC?  Is this what things like Sony Bit Mapping and other schemes to incrementally improve the dynamic range performance of 16bit recording were doing?

Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: heathen on September 22, 2020, 03:14:06 PM
Follow up question...  How picky does one need to be when converting from 24 bit to 16 bit?  I've been doing it in Audacity when I export a finished file, paying no attention to how it's actually done.  Are there different algorithms or whatever for doing that conversion?  If there are, should I care or should I just stick with the default?
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: DSatz on September 22, 2020, 05:02:55 PM
Very few people have the opportunity to record in surroundings where the actual dynamic range of their recording ends up exceeding the capacity of 16 bits. Thus in nearly all practical cases, when you are going from 24 to 16 bits, all that you're really doing is normalizing your levels so that the peaks will be as close to full scale as possible; the noise floor of your recording will fit quite comfortably within 16 bits.

In such cases, all you need to do is set levels and dither properly. Noise shaping isn't recommended because (a) it doesn't help in such situations and (b) you never really know when you or someone else may want to process the recording in some other way in the future.

In the very rare cases in which a 24-bit live recording has a greater-than-16-bit actual dynamic range, you might consider some gentle form of noise shaping as you level-shift and dither. But that's advisable only if you're fairly sure that you're delivering the recording in what will be its final form. And in any such situation, don't throw away the original.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 22, 2020, 07:35:26 PM
Quote
The way digital filters work, ESPECIALLY a digital filter emulating an older analog filter - they approximate. They can internally upsample to help improve their performance and there's more wizardry under the hood we can run there, but even still; the higher the sample rate, the more precise and accurate the filter sounds. This just sounds SO much cleaner, better, and brighter. This filter accuracy CAN be heard both at the higher sample rates at 96 kHz, as well as after that 96 kHz master has been down-converted. It's not as subtle as you'd think, it's very audible.

Does it make any difference if the the entire raw recording is upsampled upon loading in the editor prior to filtering or is doing so essentially the same achieved by upsampling within the filtering being applied, except being less efficient?

Hooray for breaking up nested quotes! Wonder how many sub-nests we can get within a few posts as the topics simultaneously diverge...

The short answer is yes there's a difference, and no the same result is not achieved by upsampling the raw data before you hit it with a filter. The underlying intuition for this is as follows:

Take a discrete sequence (in this case, the samples containing audio data). Now throw out every other sample; you have just simulated a downsampling rate of 2:1. Now, try to up-sample back to the original sample rate. The resampling algorithm here used to go back up to the original sample rate may or may not deviate from the original samples you had. That error is HEAVILY dependent on the content of the original signal, and signals are generally mixed in the probability the content between two samples is linear/exponential/sinusoidal/a mathematically elegant reconstruction function.

I've seen some resampling algorithms leverage properties of the DAC to get lower overall error, but once you lose the samples the error is there - and that error gets compounded down the effects chain (for example, by the equalizer).



On bit-depth-
Quote
A quick note: if you are adding or multiplying two digital numbers together, the more bits you have the more accurate your addition (and that is the fundamental operation of filtering - delaying, multiplying, adding)

Isn't the provision of sufficient space for calculation accuracy a separate issue from the space required to capture the full dynamic range of during the initial sampling?  I ask because I think it is important to clearly differentiate storage space requirements from processing space requirements, and editing software is working natively at higher bit-depths. Are mathematical operations which require additional space to avoid rounding errors and preserve accuracy conducted as part of the initial sampling in the ADC?  Is this what things like Sony Bit Mapping and other schemes to incrementally improve the dynamic range performance of 16bit recording were doing?

They are separate issues, yes. Both are independent consequences of bit depth.

There are encoding tricks we can play at capture, and there are different tricks we can play in-the-box to try and expand the encoding of information.

A really good conversion technique involves moving from PCM to another domain (PDM and PWM are common), and this elegantly ties in to sigma delta modulation. But the finer distinctions of those topics is less applicable to what we do as tapers, and most of what we use means we're stuck with PCM.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 22, 2020, 07:37:52 PM
Follow up question...  How picky does one need to be when converting from 24 bit to 16 bit?  I've been doing it in Audacity when I export a finished file, paying no attention to how it's actually done.  Are there different algorithms or whatever for doing that conversion?  If there are, should I care or should I just stick with the default?

Very few people have the opportunity to record in surroundings where the actual dynamic range of their recording ends up exceeding the capacity of 16 bits. Thus in nearly all practical cases, when you are going from 24 to 16 bits, all that you're really doing is normalizing your levels so that the peaks will be as close to full scale as possible; the noise floor of your recording will fit quite comfortably within 16 bits.

In such cases, all you need to do is set levels and dither properly. Noise shaping isn't recommended because (a) it doesn't help in such situations and (b) you never really know when you or someone else may want to process the recording in some other way in the future.

In the very rare cases in which a 24-bit live recording has a greater-than-16-bit actual dynamic range, you might consider some gentle form of noise shaping as you level-shift and dither. But that's advisable only if you're fairly sure that you're delivering the recording in what will be its final form. And in any such situation, don't throw away the original.

This is mostly what I'd say.

I personally use noise shaping dependent on the source material, but the noise shaping used in dither is a subtler impact than moving the microphone 5 mm in any direction.

The studio mantra is, it's better to apply light touches to a great recording, than it is to aggressively polish a good recording. The same applies to noise shaping in dither here - if you hear a big difference between them, you should revisit the capture method.

As a concrete answer: dither if you're going down from 24 to 16 bits. Otherwise, you don't need to worry once you're down to 16 bits.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: checht on September 22, 2020, 08:32:24 PM
TS community is amazing; thanks all.

Sooo, for the less-technical among us I’ll try to serve as chief summarizer in chief:

48 kHz vs 44.1 kHz
Unlikely to hear a difference unless DAC or analog circuitry very optimized for one or the other. Note that many mobile DACs are optimized for 48.
Downsampling can degrade sound when done incorrectly.
CDs are so done, unless you’re wforwumbo in which case all your CDs are belong to him.

96 kHz
Might be able to hear a difference today, but real benefit is for future listeners.
Helps a lot in post production.
Downside: double the storage required, though that cost is diminishing.

16 vs 24 bits
Might as well record with 24 bit words, makes it easier to get the max dynamic range without riding the faders.
Release how you like, noting that few folks are burning CDs.

What’d I get wrong? Please let me know and I'll update this post.

I’m now considering going back to 24/96 for recording as I did in the oughts vs sticking with 24/48. Will likely continue releasing both 16/44.1 and 24/48 based on response from folks on bt.etree.

edited to supplement/clarify based on feedback.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 22, 2020, 08:43:15 PM
Yeah! Glad this is what you absorbed. The only thing I'd change, is:

Downsampling can degrade sound when done incorrectly.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: EmRR on September 23, 2020, 08:40:52 AM
I'd make some argument that even though live recordings may not encounter greater than 16 bit dynamic range, a 24 bit capture gives you better information for forensic work in RX7/8/whatever.  If you've got hum/buzz/whatever that shows up in occasional quiet moments, those tools are going to work better with more info and dither further down under it.  All that feeds into the 32 bit float recorder wormhole......which we've done in other threads......
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Gutbucket on September 23, 2020, 10:38:37 AM
Does it make any difference if the the entire raw recording is upsampled upon loading in the editor prior to filtering or is doing so essentially the same achieved by upsampling within the filtering being applied, except being less efficient?

The short answer is yes there's a difference, and no the same result is not achieved by upsampling the raw data before you hit it with a filter. The underlying intuition for this is as follows:

Take a discrete sequence (in this case, the samples containing audio data). Now throw out every other sample; you have just simulated a downsampling rate of 2:1. Now, try to up-sample back to the original sample rate. The resampling algorithm here used to go back up to the original sample rate may or may not deviate from the original samples you had. That error is HEAVILY dependent on the content of the original signal, and signals are generally mixed in the probability the content between two samples is linear/exponential/sinusoidal/a mathematically elegant reconstruction function.

I've seen some resampling algorithms leverage properties of the DAC to get lower overall error, but once you lose the samples the error is there - and that error gets compounded down the effects chain (for example, by the equalizer).

I think you misunderstood my question.  Not asking about the difference between processing a recording originally made at, say, 96kHz sample rate without oversampling performed within the plugin filter and one made at 48kHz sample rate with 2X oversampling applied within the filter (which is what I'd love to listen for in the potential comparisons you floated in your original post), but rather of two different ways of processing the same 48kHz source recording: Is there any difference between upsampling the entire file to 96kHz first and processing without oversampling by the filter, verses processing the 48kHz file with 2X upsampling done within a filter?

Follow up question-
Up sampling seeks to determine the value of additional points along the waveform using mathematical interpolation, and reflects a mathematically rigorous determination of what the new values should be.  That's a logical approach to limiting error.  However, I can imagine other predictive schemes which might better reflect actual acoustic behavior, albeit at the risk of introducing other forms of error.  Adding harmonic distortion above the cutoff frequency, excited by a range of signal just beneath cutoff for instance. Or perhaps in a more targeted way by identifying harmonic series' which appear to have been truncated by the anti-aliasing filtering and extending those in such a way as to mimic natural harmonic progressions and decays typical to real world sounds, without adding spurious harmonic distortion to the entire signal in that range just below stop-band cutoff.

Do any up-sampling techniques do this?  Perhaps is is a hidden proprietary plugin filter upsampling strategy?
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 23, 2020, 12:51:09 PM
Does it make any difference if the the entire raw recording is upsampled upon loading in the editor prior to filtering or is doing so essentially the same achieved by upsampling within the filtering being applied, except being less efficient?

The short answer is yes there's a difference, and no the same result is not achieved by upsampling the raw data before you hit it with a filter. The underlying intuition for this is as follows:

Take a discrete sequence (in this case, the samples containing audio data). Now throw out every other sample; you have just simulated a downsampling rate of 2:1. Now, try to up-sample back to the original sample rate. The resampling algorithm here used to go back up to the original sample rate may or may not deviate from the original samples you had. That error is HEAVILY dependent on the content of the original signal, and signals are generally mixed in the probability the content between two samples is linear/exponential/sinusoidal/a mathematically elegant reconstruction function.

I've seen some resampling algorithms leverage properties of the DAC to get lower overall error, but once you lose the samples the error is there - and that error gets compounded down the effects chain (for example, by the equalizer).

I think you misunderstood my question.  Not asking about the difference between processing a recording originally made at, say, 96kHz sample rate without oversampling performed within the plugin filter and one made at 48kHz sample rate with 2X oversampling applied within the filter (which is what I'd love to listen for in the potential comparisons you floated in your original post), but rather of two different ways of processing the same 48kHz source recording: Is there any difference between upsampling the entire file to 96kHz first and processing without oversampling by the filter, verses processing the 48kHz file with 2X upsampling done within a filter?

Follow up question-
Up sampling seeks to determine the value of additional points along the waveform using mathematical interpolation, and reflects a mathematically rigorous determination of what the new values should be.  That's a logical approach to limiting error.  However, I can imagine other predictive schemes which might better reflect actual acoustic behavior, albeit at the risk of introducing other forms of error.  Adding harmonic distortion above the cutoff frequency, excited by a range of signal just beneath cutoff for instance. Or perhaps in a more targeted way by identifying harmonic series' which appear to have been truncated by the anti-aliasing filtering and extending those in such a way as to mimic natural harmonic progressions and decays typical to real world sounds, without adding spurious harmonic distortion to the entire signal in that range just below stop-band cutoff.

Do any up-sampling techniques do this?  Perhaps is is a hidden proprietary plugin filter upsampling strategy?

Ah, now I see your question. I had assumed you were asking with the condition that the filter was performing internal upsampling regardless.

It's a hard question to generalize, as the answer is dependent on how the filter goes about its business. Depending on the method of coefficient calculation, it could have an impact. But *generally* speaking, you shouldn't hear as big of a difference. The exceptions are for specific digital filters doing analog modeling which try to incorporate some of the non-linear and time-variant properties of said real-world analog filters.

Regarding filter error on your follow up question, the problem we run into is that filter error is Gaussian. This means that the error induces distortion even below the cutoff frequency. The level of this error goes down with some signal processing techniques (sigma-delta modulators in conversion and converting from PCM can minimize this), but the error will almost always be Additive White Gaussian Noise, given enough samples for the error to be analyzed. Side note, worthy of another thread sometime: noise analysis is VERY different in the short-time (i.e. frame-to-frame) as opposed to the long-term, integrated noise response.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: morst on September 24, 2020, 01:08:28 PM
Take a discrete sequence (in this case, the samples containing audio data). Now throw out every other sample; you have just simulated a downsampling rate of 2:1.

Is this how downsampling is implemented? I would have guessed that averaging two (or three?) samples would be the method, although it would take more CPU.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: wforwumbo on September 24, 2020, 01:22:16 PM
Take a discrete sequence (in this case, the samples containing audio data). Now throw out every other sample; you have just simulated a downsampling rate of 2:1.

Is this how downsampling is implemented? I would have guessed that averaging two (or three?) samples would be the method, although it would take more CPU.

The answer is “sometimes”

Good downsampling algorithms that can recognize an integer ratio will do this. But more often than naught, they don’t.

What’s more common in a downsampling algorithm is:
-a block of samples are up-sampled to a common denominator between the native (old) and target (new) sample rates
-the samples “in between” where no native data exists are interpolated. This can vary from “connect the dots” which is similar to the averaging you mention, to “create a quadratic (second order function, which can create curves) or higher order function to model this block of 3-5 samples and use this function to ‘fill in the blanks’” method. Splines are common, and there are some advanced techniques from noise communications like using Lagrange polynomials to generate the missing sample data.
-Use the function you just generated to calculate the value at the target sample rate where no data previously existed.

The upside to the benefit to this method is that it safely assumes you’re not resampling between integer values (for example, resampling from 48 to 44.1) and thus there WILL be some level of error, so you can pick your poison of error in the function used to “connect those dots” - which themselves will be wrong, but you can expect and minimize the wrong-ness. The downside is complex source material - such as music, which contains lots of concurrent frequency and phase data - can behave erratically and unpredictable as samples get shuffled forward/backwards in time and error. This can create the perception of slightly muted treble, moving sources, or just pure mush in the soundstage. Effectively, the recording sounds unstable.

We can play some tricks to try and recover or mitigate the distorted phase, but that’s also a “pick your poison” deal and brings its own caveats. The issue is that resampling distortion is not inherently linear across frequency, and thus no ideal filter exists to perfectly reconstruct the signal except for at integer ratios and that’s almost never guaranteed. Look at this thread - most people (myself included) are likely to go from 96 to 44.1 and there’s no plan to change that; the implication is that unless you control the resampling manually yourself, even going from 88.2 to 44.1 will induce some phase distortion due to common resampling algorithms with phase compensation that doesn’t need to be there and sadly can’t be undone after the fact.

Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on September 24, 2020, 07:25:04 PM
For some additional useful thoughts on bit depth, I dug up an old post of mine: http://taperssection.com/index.php?topic=184569.msg2251598#msg2251598

Copy-pasting it here, for posterity's sake and a consolidated/collected base of this information. Note that, like most of what I have said in this thread so far, this is a simplification of the problem meant to demonstrate the concept and not by any means a complete reflection of the underlying principle.

For a simple example, let's say I had a system that wanted to add two numbers. The numbers are 1571 and 0448. The result for any human adding these two numbers together is 2019. The effect of a 16-bit system would be having the two numbers as 1500 and 0400, which adds up to 1900. SOME 16-bit systems can try and adjust this during recording as 1600 and 0400, which gives us 2000 and is a bit closer, but that requires some additional processing power on the A/D converter, most of which doesn't occur except in more expensive 16-bit recorders. If your converter is capable of 24-bit or 32-bit float and you're recording in 16-bit mode or manually converting to 16-bit in post, you're probably getting the former. A 24-bit system is the same as going to 1570 and 440, which adds up to 2010 - a lot closer than 1900.Your Samplitdue system, at 32-bit float and importing in 24-bit files, would be similar to having 1570.00 and 0440.00, which allows for decimal rounding at the end after processing - not of note during recording, but DEFINITELY useful in post, particularly pre-bit reduction and dither, as the equalizers can come MUCH MUCH closer to what the real values should be.

This is obviously an imperfect example, and a touch exaggerated. But it gets the point across.
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: morst on September 25, 2020, 01:57:47 PM

Fascinating, I would have guessed that the answer was almost never or never. Thanks for the detailed reply, explaining the use of quadratic / higher order interpolation.

Take a discrete sequence (in this case, the samples containing audio data). Now throw out every other sample; you have just simulated a downsampling rate of 2:1.

Is this how downsampling is implemented? I would have guessed that averaging two (or three?) samples would be the method, although it would take more CPU.

The answer is “sometimes”
Good downsampling algorithms that can recognize an integer ratio will do this.
But more often than naught, they don’t.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on September 25, 2020, 04:39:47 PM
^^ Can you please elaborate on your example a little? Maybe I am being a little thick, but I don't understand how it applies to the recording itself. If the dynamic range of the source (plus headroom) "fits" into 16 bits, wouldn't that be sufficient? Can't both 1571 and 0448 be described exactly with 16 bits? I can see that performing operations on that data might push it outside of the 16-bit range, but, at least in the case of tapers, barely any of us are doing that during the recording process. When we import it into a DAW to process, the extra bits to handle operations will be available there.

I do understand the benefit of 24-bit recording and I can see the point for processing, but I don't get the reason why a less than 16-bit source would benefit from a longer word (assuming properly set gain).

[EDIT: Removed a typo.]
Title: Re: 48 KHz vs 44.1 KHz sample rate
Post by: Gordon on September 25, 2020, 05:27:57 PM
I record and post at 24/48 ; it seems to me i nthis day and age people should be able to handle this...


I just started only releasing 24/48 about a year ago.  Exactly one person has asked for a 16/44.1! 
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: jerryfreak on September 25, 2020, 07:47:40 PM
^^ Can you please elaborate on your example a little? Maybe I am being a little thick, but I don't understand how it applies to the recording itself. If the dynamic range of the source (plus headroom) "fits" into 16 bits, wouldn't that be sufficient? Can't both 1571 and 0448 be described exactly with 16 bits? I can see that performing operations on that data might push it outside of the 16-bit range, but, at least in the case of tapers, barely any of us are doing that during the recording process. When we import it into a DAW to process, the extra bits to handle operations will be available there.

I do understand the benefit of 24-bit recording and I can see the point for processing, but I don't get the reason why a less than 16-bit source would benefit from a longer word (assuming properly set gain).

[EDIT: Removed a typo.]

that example is a bit exaggerated. 24 vs 16 bit is more like adding 1571.36453235439801090341 and 0448.59372906135484319835 vs 1571.364532354398 and 0448.593729061354, when in the end your required result cannot discern doesnt depend on much of anything past maybe the 9th or 10th decimal digit anyway

now an example of recording at very low levels (assuming no hardware limitations, 0000.0000 0000 0123 4567 8901 and  0000.0000 0000 0987 6543 2109 would look like  0000.0000 0000 0123  and 0000.0000 0000 0987 respectively in 16 bit.

if you add those and THEN normalize, in the 16 bit case you end up with 1110.00000000000000 vs 1111.111110100000

that difference would be more understandable as distortion - ie % of difference in data from input to output of device or process



Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on September 27, 2020, 02:49:56 PM
Maybe my last post wasn't clear enough, but I am not really interested in operations. What I am asking is this: "If the source is less than 16-bits and properly recorded to ISOs at 16-bits, is there any advantage to recording at a higher bit rate?" As I mentioned, I see the advantages of not having to ride levels or for processing. Also, aren't PCM samples signed integers?
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on September 30, 2020, 11:54:15 PM
aaronji, to your previous question, if a recording "fits" within 16 bits then by definition there is no benefit to be gained by extending it to 24 bits, as long as you don't alter the recording otherwise.

If, however, the recording really uses the full range of the available 16 bits (I think it happened to me only once ever, at a percussion ensemble concert), then if you wish to do certain kinds of processing on it, there might be a tiny, tiny, tiny (and almost certainly not audible) theoretical advantage to adding another bit or two. To make the difference even potentially audible, you would need to play back the recording at a level allowing you to hear those bottom bits, and then the highest-order bits would be about 100 dB louder, which is not gonna happen in a critical listening situation.

--You can view the digital samples as signed integers as you want, but then you end up with full scale = +32767 or -32768, which are rather arbitrary numbers if you're not a binary geek. To me it makes more sense to view the samples as binary fractions, i.e. analogous to decimal fractions such as 0.45 or -0.234, with the bit values based on the series 1/2, 1/4, 1/8, etc. -- then full scale can be reckoned either as +/- 1 or, if you prefer, +/- 1/2 so that the entire range is then 1 from peak to peak.

In either case, because of the particular binary notation involved (so-called "two's complement" arithmetic), the 0 sign bit for the value 0 places that value in the positive half of the range. So there is one possible extreme sample value in the negative direction that is one step farther than exists on the positive side. This is equivalent to saying that in integer arithmetic (say, for an eight-bit byte), the possible values are from -128 through 0 to +127.

Many years ago I worked with a computer system (the "Adage Graphics Terminal") that used "ones' complement" arithmetic; it was weird because it had both +0 and -0 that were logically distinct values, even though they were quantitatively equal. And its maximum positive and maximum negative values were identically far from 0. But that type of arithmetic was rarely used even then, and by convention PCM audio is always twos' complement.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 01, 2020, 09:54:41 AM
aaronji, to your previous question, if a recording "fits" within 16 bits then by definition there is no benefit to be gained by extending it to 24 bits, as long as you don't alter the recording otherwise.

If, however, the recording really uses the full range of the available 16 bits (I think it happened to me only once ever, at a percussion ensemble concert), then if you wish to do certain kinds of processing on it, there might be a tiny, tiny, tiny (and almost certainly not audible) theoretical advantage to adding another bit or two. To make the difference even potentially audible, you would need to play back the recording at a level allowing you to hear those bottom bits, and then the highest-order bits would be about 100 dB louder, which is not gonna happen in a critical listening situation.

Thank you, DSatz. That was my intuitive understanding and exactly the sort of answer for which I was looking. Lots of dubious "information" on this topic out there on the web...

With respect to the binary, binary fractions are fine with me. The difference with respect to signed integers is just the scale (as you mentioned). I was actually specifically referring to the example that was using numbers that can't be represented in binary at all, or, rather, only as infinitely long binary numbers fractions.   

[EDIT: For clarity.]
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: morst on October 01, 2020, 02:56:26 PM
In either case, because of the particular binary notation involved (so-called "two's complement" arithmetic), the 0 sign bit for the value 0 places that value in the positive half of the range. So there is one possible extreme sample value in the negative direction that is one step farther than exists on the positive side. This is equivalent to saying that in integer arithmetic (say, for an eight-bit byte), the possible values are from -128 through 0 to +127.
:o :o

In order to have a "zero crossing" in a binary system, there is a necessary asymmetry!?
This is obviously true but it makes my brainzzzz explode!
 :zombie03:


PS the Sound Summit day one is complete and will continue to be available at this link https://www.youtube.com/watch?v=E1iUbxfJmQQ
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: 108Ω on October 03, 2020, 01:30:51 AM
When intent is only CD, I record at 24/88.2
These days, with media so inexpensive, I record at 24/192, and archive the master.
With a high initial sampling rate, I'm totally comfortable with a resample to 44.1.
24/96 --> 16/44.1 (or 20/44.1) has been industry standard for decades.

Worthy of mention, DAT 16/48 was responsible for at least a few fast tapes in the 80's.
I'm not sure how the consumer decks did it, but the 48kHz rate was cited as the reason in a few DATHeads posts.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: morst on October 03, 2020, 10:49:19 AM
When intent is only CD, I record at 24/88.2
These days, with media so inexpensive, I record at 24/192, and archive the master.
With a high initial sampling rate, I'm totally comfortable with a resample to 44.1.

24/96 --> 16/44.1 (or 20/44.1) has been industry standard for decades.
It ain't about the rate, it's about a whole number ratio being completely different and much simpler math
Since you keep your masters you can always redo everything later.


Edit on Mon oct 5:
YOW!
Looks like my statement, now highlighted in red, is an incorrect assumption on my part!
See DSatz post on the next page.

Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on October 05, 2020, 08:10:38 PM
No, if you're downsampling then you MUST filter first (or simultaneously), and there's no special, more straightforward approach that applies to simple integer ratios. You might just as well go from 96 kHz to 44.1 kHz (or any other, arbitrary rate, even if it's not an integer!) as to 48 kHz. There is no gain or loss in efficiency, accuracy, dynamic range, distortion, ethical purity, sexual desirability, followers on social media, or anything else either way.

If you're upsampling, then part of the code could be made simpler for the special case of integer ratios of sampling rates. However, no actual sampling rate conversion hardware OR software that I've ever seen, smelled, tasted, or heard of uses such an approach. I'm trying and failing to find a suitable metaphor to explain why this is, since metaphors seem to convince people where actual descriptions of reality do not.

Yes, those special cases would be easier to program than the general case. But the writers of the software or firmware would STILL need to support the general case as well, test it, AND THEN have the "easier" way standing by for whatever percentage of cases in which it could be used. They'd also have to build in additional logic to detect the special cases that could use the simpler code and switch over to it. But that's a worse situation for the software or firmware developer, not a better one. Yes? The inconvenience and risk that they would have to take on in order to code, test and integrate their implementation of the "easier" approach, would come in addition to the inconvenience and risk of handling the more general case that they have to handle anyway.

So no one that I've ever heard of, or can realistically imagine, does that when writing sample-rate conversion software.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: morst on October 05, 2020, 11:36:32 PM
Yes, those special cases would be easier to program than the general case. But the writers of the software or firmware would STILL need to support the general case as well, test it, AND THEN have the "easier" way standing by for whatever percentage of cases in which it could be used.
Dang, I guessed wrong!
Thanks for the straight info, and note that I have edited my post above to reflect this.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: Gutbucket on October 06, 2020, 06:04:12 PM
"I'm trying and failing to find a suitable metaphor to explain why this is, since metaphors seem to convince people where actual descriptions of reality do not."
^
Thanks for this gem dejour, which very much rings true.. but then, isn't all language inescapably metaphoric? 
Even mathematics, which seems odd. 
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on October 07, 2020, 02:47:15 PM
I don't know about all language being metaphorical; maybe. This gets into Kant and Hume and whether or not we ever experience "reality itself". All I know so far is that I respect those questions, and can't answer them for sure.

And I understand the impatience of Ayn Rand and her followers, who refuse to respect those questions--which, however, doesn't make the questions go away; it just makes "Objectivism" another branch of religion, in which people choose what they will believe because of how it makes them feel when they believe it. Hulk smash: I want life and the universe to work a certain way, therefore I claim that they should and do work that way.

Yeah, so, metaphors and visualizations--they're a huge help, but they can also cement mistaken impressions into place (see that metaphor there?). Sooooooo many people imagine the sampling process in A/D conversion in a way that makes it "perfectly obvious" that the higher the sampling rate, the more accurate the reconstruction will be. And that's just not a fact, given the Shannon/Nyquist bandwidth constraints. But as long as you're thinking by reference to the visual metaphor, you can't see what's true and what's not true about that model or outside of that model, where the sampling and reconstruction processes actually live.

It gets to where people won't even listen to evidence from the real world, because their mental model is so precious to them. Not just occasionally, but really often.

I think that's a lot of why Aristotle was the official philosopher of Western Christianity for so many centuries despite being a pagan. His whole system was made up of assertions that feel right, as long as you don't have any empirical knowledge of the subject at hand. And back then, not many people did. As a result, empirical knowledge itself wasn't respected; think of the people who refused "on principle" to look through Galileo's telescope.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: noahbickart on October 07, 2020, 04:38:54 PM
I think that's a lot of why Aristotle was the official philosopher of Western Christianity for so many centuries despite being a pagan. His whole system was made up of assertions that feel right, as long as you don't have any empirical knowledge of the subject at hand. And back then, not many people did. As a result, empirical knowledge itself wasn't respected; think of the people who refused "on principle" to look through Galileo's telescope.

Of Medieval Judaism and (especially) Islam as well....
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 08, 2020, 06:03:01 AM
My little vote is for Gutbucket. They taught me that language, words are concepts. For me, math and thoughts belong there too. Unfortunately, we need words. Ha ha, for most of my friends, "Shannon/Nyquist bandwidth constraints" is a boring word at best. But for me it's an interesting word thanks to Taperssection.

I would have a question if I imagine the following just like the others:

According to Shannon / Nyquist - 44.1kHz sampling is sufficient for accurate signal reconstruction when the recorded signal does not contain frequencies higher than 20kHz. My microphones are cheap and I can't hear even 15kHz. So 44.1kHz is perfectly sufficient for me.

For the reason of higher sampling, Wforwombo mentioned post-processing. Does this mean that some post-processing also takes into account frequencies above 20kHz and the processing of these high frequencies will affect sound that I already hear - 15kHz and below? I use only basic plugins like eq, limiter, compressor, sometimes multi-band compressor.

Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 08, 2020, 12:02:41 PM
On the whole, what DSatz has said since I last checked this thread is right.

I’ve seen resampling code that DOES have the “special integer case” resampling method in the field, BUT it is a safe assumption that any publically available resampling code is NOT doing this.

For the reason of higher sampling, Wforwombo mentioned post-processing. Does this mean that some post-processing also takes into account frequencies above 20kHz and the processing of these high frequencies will affect sound that I already hear - 15kHz and below? I use only basic plugins like eq, limiter, compressor, sometimes multi-band compressor.

This is a good question - but it’s confusing the symptom with the cause.

The “band limiting” of 20 kHz has to do with *accurately reproducing the signal in the analog domain*

Digital signals DO have data outside of that bandwidth of 0-20 kHz. They actually have a lot of information outside of that range defined from -infinty to +infinity, and even more technically the baseband has info from -20k to +20k (don’t think too hard about that one...) but this is another conversation for another day.

Here’s what Nyquist-Shannon are ACTUALLY saying in their theorem, paraphrased by me (I encourage you to seek out their 1927 paper and read it yourself, I’m happy to answer any questions and discuss differences in interpretation):

If you have an analog signal, and want to sample it into a digital domain, and then reproduce it back in the analog domain... in order to do so without any distortion or loss of information, you must sample the signal at a rate that is at least twice the highest frequency that is contained within the signal.

So you need to disentangle the “a 44.1k signal is only capturing up to 20k” from the sample rate. The sample rate is a property of your recording that has effects and impacts - one of them being, in a theoretically ideal world you could only reproduce data with a bandwidth of 0-22,050 Hz without distortion if you do everything else right.

The post processing argument has to do with a reduction in error, especially for filtering. The degree of audibility of that error is up for debate. But the idea here is that you’re no longer reproducing the original signal, you’re now modifying it - and that has impacts elsewhere in the chain. These impacts are a function - symptom of - sample rate, not necessarily a cause.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 08, 2020, 04:10:18 PM
Thank you Wforwumbo. I think I am starting to understand. 44.1k sampling is sufficient for accurate reproducing a signal to analog domain, e.g. for playback up to 20k. But digital post-processing is modifying a signal in digital domain. And apply here, the more accurate the samples the better.

I have two more theoretical question:

Could we record at 44.1k/16bits and resample the signal later in a case someone would like to do a fine digital post-processing? If I understand correctly: The result should be very similar to the recording that is recorded from the beginning at 96k/24bits. The only difference is that there are two additional d/a and a/d converters in the recording chain. Is it right?

I know that DAW can internally resample signals into the higher rate because more accurate calculations. Is there a difference between an internal resampling in DAW and physical d/a and a/d converters?

I understood not all the posts in this thread, I apologize if anyone has already explained this.

Digital signals DO have data outside of that bandwidth of 0-20 kHz. They actually have a lot of information outside of that range defined from -infinty to +infinity, and even more technically the baseband has info from -20k to +20k (don’t think too hard about that one...) but this is another conversation for another day.
I prefer not to read this paragraph.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 08, 2020, 04:40:40 PM
Thank you Wforwumbo. I think I am starting to understand. 44.1k sampling is sufficient for accurate reproducing a signal to analog domain, e.g. for playback up to 20k. But digital post-processing is modifying a signal in digital domain. And apply here, the more accurate the samples the better.

I have two more theoretical question:

Could we record at 44.1k/16bits and resample the signal later in a case someone would like to do a fine digital post-processing? If I understand correctly: The result should be very similar to the recording that is recorded from the beginning at 96k/24bits. The only difference is that there are two additional d/a and a/d converters in the recording chain. Is it right?

I know that DAW can internally resample signals into the higher rate because more accurate calculations. Is there a difference between an internal resampling in DAW and physical d/a and a/d converters?

I understood not all the posts in this thread, I apologize if anyone has already explained this.

Assuming ideal conversion between the analog and digital domain, I don’t think you’d hear a difference re: sample rate. Taking 44.1 content, converting it to analog, and then sampling that at 96 - assuming no distortion - should give the same result as native content at 96k. Then you could do your processing at the higher sample rate without issue. This is because the audible frequency range is captured at both sample rates with no distortion.

Bit depth is different; once the bits are gone, they’re gone. You can’t really replace them. IF the dynamic range of the original content is less than 96 dB AND the noise floor of your system is greater than 96 dB and induces no distortion, then I don’t think you would see an appreciable difference post-reconversion, but those are two big ifs.

The last question depends on the DAW and resampling algorithms of choice.

How much any of this is appreciable or noticeable is still up for debate. One of these days I’ll do an A-B...

Addendum: technically in the resampling process, there will be distortion between the two cases. But that distortion will occur between 22.050 kHz and 48 kHz. And if I were a gambling man I’d put money down that you couldn’t perceive such distortion.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 08, 2020, 04:55:13 PM
Digital signals DO have data outside of that bandwidth of 0-20 kHz. They actually have a lot of information outside of that range defined from -infinty to +infinity, and even more technically the baseband has info from -20k to +20k (don’t think too hard about that one...) but this is another conversation for another day.

Would you be so kind as to explain this in greater detail? If not here then in a new thread? I can understand why this would be the case in the mathematical realm, but I can't wrap my head around the idea of negative Hz in the physical realm. It seems, basically, impossible to me.

Bit depth is different; once the bits are gone, they’re gone. You can’t really replace them. IF the dynamic range of the original content is less than 96 dB AND the noise floor of your system is greater than 96 dB and induces no distortion, then I don’t think you would see an appreciable difference post-reconversion, but those are two big ifs.

That first one is definitely not a big if for tapers. The maximum SPL of the music minus the ambient SPL is probably almost always less than 96 dB in our world. DSatz mentioned above that he (thinks he) has only encountered a recording situation that actually exceeded 96 dB once.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 08, 2020, 05:23:38 PM
Thank you very much Wforwumbo for nice explanation. I understand.

My question about resampling in DAW was meant with regard to - If someone has a good DAW with good resampling algorithm, can this replace d/a and a/d physical converters to resample the recording from original 44.1k to 96k due to fine digital post-processing?
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 08, 2020, 05:33:13 PM
Digital signals DO have data outside of that bandwidth of 0-20 kHz. They actually have a lot of information outside of that range defined from -infinty to +infinity, and even more technically the baseband has info from -20k to +20k (don’t think too hard about that one...) but this is another conversation for another day.

Would you be so kind as to explain this in greater detail? If not here then in a new thread? I can understand why this would be the case in the mathematical realm, but I can't wrap my head around the idea of negative Hz in the physical realm. It seems, basically, impossible to me.

Bit depth is different; once the bits are gone, they’re gone. You can’t really replace them. IF the dynamic range of the original content is less than 96 dB AND the noise floor of your system is greater than 96 dB and induces no distortion, then I don’t think you would see an appreciable difference post-reconversion, but those are two big ifs.

That first one is definitely not a big if for tapers. The maximum SPL of the music minus the ambient SPL is probably almost always less than 96 dB in our world. DSatz mentioned above that he (thinks he) has only encountered a recording situation that actually exceeded 96 dB once.

Re: the second point, for the vast majority of situations we are in, yes you are correct. I'll note that I've been in rooms with NC-5 (meaning that the "background hum" is less than 5 dB) and peak volume of a performance was at 127 dB-C, but that's definitely an exception to a rule. It was definitely interesting having the self noise of my microphones and the dynamic range of my mic pres be the bottleneck in the dynamic range chain... but again, exception to the rule.

Re: the first question... yes. More to follow later this evening when I have some time to draw out diagrams, since it actually gets at the core of why the sampling theorem exists.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 08, 2020, 05:35:55 PM
Thank you very much Wforwumbo for nice explanation. I understand.

My question about resampling in DAW was meant with regard to - If someone has a good DAW with good resampling algorithm, can this replace d/a and a/d physical converters to resample the recording from original 44.1k to 96k due to fine digital post-processing?

Short answer is "no", the less succinct answer is "with lots of caveats, it's possible", and the best answer is "just record at 96k in the first place to avoid such a headache."

There are tricks we can play in terms of "modeling the transfer from digital back to the analog domain" with upsampling, but our tricks can only get us so far. There are some standards that try to model this (such code is used to accurately display dB meters in DAWs, for example), but because of the shortcomings of resampling algorithms they're already inherently a bit flawed.

Edit: If your question is more about understanding the philosophy behind sample rates and resampling, then yes your intuition is barking up the right tree here. But in practice... just record at the higher sample rate.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 08, 2020, 05:56:18 PM
Re: the second point, for the vast majority of situations we are in, yes you are correct. I'll note that I've been in rooms with NC-5 (meaning that the "background hum" is less than 5 dB) and peak volume of a performance was at 127 dB-C, but that's definitely an exception to a rule. It was definitely interesting having the self noise of my microphones and the dynamic range of my mic pres be the bottleneck in the dynamic range chain... but again, exception to the rule.

I think there is a vanishingly small probability of that happening in a taper situation! Even in a studio setting, I suppose 5 dbSPL is a real rarity. For us, I think we are lucky if the background is only 50 dB higher...

Re: the first question... yes. More to follow later this evening when I have some time to draw out diagrams, since it actually gets at the core of why the sampling theorem exists.

I would definitely appreciate it! I think I have a pretty good layman's understanding of Nyquist-Shannon, but I like numbers and would appreciate the deeper insight.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 08, 2020, 06:54:22 PM
Thank you!

I was asking because of both philosophical and also practical. I am doing amateur, audience recordings at 44.1k/24bit. I am archiving the original recordings after level adjustment at 44.1k/16bit. My dynamic range is lower than 96 dB. I don't record extremely loud music and the background noise at concerts is relatively high. I hope 16bit won't hurt. Probably 44.1k is also enough for me because I'm not good at post-processing, I am not able to distinguish fine details. But I will try to compare post-procesing with 44.1k vs 96k.

What's good I learned here is that if I find some treasure in the archive, whoever would do quality digital post-processing in the future can resample my original with good a/d convertors. And the loss will not be so big.

Of course, I understand now that those who do well post-processing or those who record professionally record and store at 96k/24bit.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: heathen on October 10, 2020, 08:32:43 AM
This thread is like a free graduate school course in digital audio. Thanks all!
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: checht on October 10, 2020, 04:56:09 PM
...... What's good I learned here is that if I find some treasure in the archive, whoever would do quality digital post-processing in the future can resample my original with good a/d convertors. And the loss will not be so big.

Curious why you seem to prefer the dac -> adc path to convert vs upsampling in a DAW. Seems to me that there are more possibilities for problems going out of the digital domain and depending on perfect cables and connections in analog domain, and also more digital processes in that process, versus a singlue upsample process.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on October 11, 2020, 01:11:06 AM
Let's please keep in mind that audio sampling is just slicing a continuous signal into discrete ... samples. It doesn't include the subsequent conversion of those samples into numerical values; that's quantization. Back when sampling theory was first developed, quantization at speeds useful for audio wasn't readily available. But sampling itself is inherently analog. So maybe it can be simpler if people just think about sampling WITHOUT quantization just for the duration of this one message.

Like, you all know (or will know in a moment) that a capacitor can hold an electrical charge constant for some amount of time. If you take a capacitor and connect its two terminals to each other, any charge that it was holding will very soon be discharged, because the two "plates" of the capacitor will then be at the same level of charge as each other, and thus will have zero voltage (potential) relative to each other. (This is ignoring some strange low-level effects that happen in the real-world; let's say this is an ideal capacitor that wants to be included in thought experiments and has signed all kinds of release forms, so leakage and internal resistance and dielectric absorption and inductance are gonna sit this one out, OK?)

OK. So let's say that you get yourself a VERY large number of these ideal capacitors, and you set them all up in a row and tie one side of each to a common ground. They're sitting there like brand-new babies, just waiting for the world to give them information to store in their bellies. That information will be in the form of a certain level of charge relative to ground--but initially, they've all been discharged fully. Tabula rasa.

Now let's take a varying signal voltage that consists entirely of frequencies below, say, 1 kHz. Shannon's sampling theorem says that if you connect the signal at low impedance to each capacitor in turn, switching (noiselessly and in zero time--good luck) from one capacitor to the next every 0.5 msec, then after (say) 10 seconds you will have charged 20,000 capacitors in sequence, each according to the amplitude of the signal during its special little interval* of time. And since these are all perfect capacitors that volunteered to be part of this test, they can all hold their little breaths for as long as you want. Go get a sandwich or something.

Then you can come back and do the same thing in reverse: Discharge each capacitor in turn, spending exactly 0.5 msec on each one, into an amplifier that has a reasonably high input impedance. You'll get switching noise and several other kinds of crap above 1 kHz, but below 1 kHz where the original signal was, you should get a perfect** reconstruction of it--which you can recover for practical use by filtering at 1 kHz again. Since you can never do that perfectly in the real world, that's why there's a somewhat substantial guard band (that's part of why the CD sampling rate is 44.1 kHz rather than 40.0000000000000000000000000001 kHz).

In this process, time got divided into discrete intervals, but the amplitude of the signal remained continuous; no quantization or digitization occurred in the making of this thought experiment, which was supervised by the local chapter of the American Society for the Prevention of Stairstep-levels in Audio (ASPSA).

And that type of scenario is what the sampling theorem was originally about, and what the discussion of sampling rates is based on. Not digitization, which is a further dimension of processing that depends on sampling.

--best regards

______________________
* note "interval of time", not "moment in time"--that's key to one of the great misconceptions about digital audio, but I'll save that for another message some day.
** "perfect" here means: You can achieve any level of quality that you're willing to pay for; the process of sampling as such won't limit you--though available real-world electrical components, etc., surely will.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 11, 2020, 08:54:05 AM
In this process, time got divided into discrete intervals, but the amplitude of the signal remained continuous; no quantization or digitization occurred in the making of this thought experiment, which was supervised by the local chapter of the American Society for the Prevention of Stairstep-levels in Audio (ASPSA).

Where can I sign up???

Just kidding. Thanks for the interesting post...
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 11, 2020, 05:35:49 PM
I read DSatz's post. I disobeyed, I didn't go to get a sandwich, but I kept reading. It was a mistake, it took me a while to deal with the capacitors (and their full bellies). I finally understood, thanks a lot!

Checht, if I understood Wforwumbo well, accurate upsampling the signal in digital domain (in DAW) is not an easy task. Simulating dac and adc in DAW is probably difficult. But it is possible, see Wforwumbo's post. Professionals can probably upsample in DAW. For me, I think, it would be easier and safer to use dac and adc.

Thank you very much Wforwumbo for nice explanation. I understand.

My question about resampling in DAW was meant with regard to - If someone has a good DAW with good resampling algorithm, can this replace d/a and a/d physical converters to resample the recording from original 44.1k to 96k due to fine digital post-processing?

Short answer is "no", the less succinct answer is "with lots of caveats, it's possible", and the best answer is "just record at 96k in the first place to avoid such a headache."

There are tricks we can play in terms of "modeling the transfer from digital back to the analog domain" with upsampling, but our tricks can only get us so far. There are some standards that try to model this (such code is used to accurately display dB meters in DAWs, for example), but because of the shortcomings of resampling algorithms they're already inherently a bit flawed.

Edit: If your question is more about understanding the philosophy behind sample rates and resampling, then yes your intuition is barking up the right tree here. But in practice... just record at the higher sample rate.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 11, 2020, 07:22:43 PM
I would also like to ask about quantization. These are only theoretical questions. I have recorded in 24bit. And I used to convert it to 16bit. This was due to space and also due easier file handling. I normalized master recordings to 0 dB FS and then I converted it to 16bit. Now, out of curiosity,  I am thinking about it in more details:

16 bits are 96db dynamic range. A quantization noise is a half of the last bit, so it is 3dB. This is less than microphone or background noise. When the master recordings does not exceed a dynamic range 96dB and after a level normalization, can I always save the master recordings to 16bit without degrading it?

What happens when the recording has a dynamic range greater than 96dB and I save it in 16bit? Will the quantization noise increase? Eg dynamic range 108db (96db+12db), quantization noise is 12db?

Am I thinking about it right?
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: checht on October 11, 2020, 10:05:40 PM
Checht, if I understood Wforwumbo well, accurate upsampling the signal in digital domain (in DAW) is not an easy task. Simulating dac and adc in DAW is probably difficult. But it is possible, see Wforwumbo's post. Professionals can probably upsample in DAW. For me, I think, it would be easier and safer to use dac and adc.

Kuba, thanks for your reply, and your great questions.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on October 11, 2020, 11:56:40 PM
kuba e, when you convert a 24-bit recording to 16 bits, your software should definitely redither at the 16-bit level prior to truncation; otherwise the truncation process would add quantization noise. It's highly improbable that you would actually hear that noise, but not impossible; certainly you can create situations in which the noise is audible if you raise the playback gain far enough, and it's not at all a nice kind of noise, since its level is highly program-dependent (i.e. it sounds like analog tape modulation noise that's having a particularly bad day). It's not too "expensive" (in terms of dynamic range lost, CPU time, or anything else) to use dither to ensure that no quantization noise whatsoever is added--so fortunately, that has become standard practice.

The exact numbers for the maximum theoretical dynamic range of a 16-bit recording, and the minimum noise increase for adequate dither, aren't exactly 96 and 3 dB for technical/math reasons, but those are within a dB or two of the right numbers. The discrepancy is partly because a factor of 2 isn't exactly 6 dB but rather 20 * log10 2 = 6.0206... dB, and partly because the greatest undistorted amplitude that can be represented is the rms value of a sinusoid whose peaks just reach full scale, rather than a sinusoid whose rms amplitude itself is at full scale. If I recall correctly the actual value is 97.3 dB maybe, while the minimum lost dynamic range due to dither can be closer to 2 dB if you're clever about its characteristics ("probability density"), but it's been years since I saw the math spelled out completely.

But those are theoretical values based on simplified models of both the signal and the noise, both in terms of the distribution across the frequency spectrum, and the relative amount of time that each one spends at each possible level within its range (i.e. its varying levels over time). A random noise will have occasional moments (which may last only one sampling interval, but still ...) of being significantly higher in level than its long-term average. The ear/brain does a certain amount of time-averaging, but is also sensitive to sudden shifts. As a result, rms values are rather poor indicators of what we hear in live situations, unless our life consists of listening to sine waves. This is why marketing people love to quote A-weighted, rms values of equivalent noise for microphones--typically those are 10 to 12 dB lower than the much more revealing CCIR-weighted quasi-peak values, even though the latter also contain some averaging (it's just a much more perceptually relevant, carefully tested, shorter-term form of averaging).

As a result, there can be real-world cases in which the noise floor of a recording system may be exposed in a given range of frequencies somewhat more readily than you might expect from the simplified/schematized ideal values alone. This is why I was a fan of Dolby "A"-type noise reduction even when I was recording live concerts at 15 ips on a well-maintained Nagra recorder, and eventually telcom c4 noise reduction, which is even more powerful. And it's one reason I get impatient with people who say that The Right Way is to set your levels so that the highest signal peaks are at -12 or even lower (I've seen someone say -16 to -18), and don't worry because "24 bits". That approach could work in a given case (i.e. not add any unnecessary audible noise to the recording) but it could very definitely fail (i.e. add unnecessary audible noise to the recording, possibly in quite significant amounts); without further details, it's impossible to know which.

--best regards
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on October 12, 2020, 12:40:11 AM
kuba e, after I wrote my previous message about dynamic range I re-read your messages, and I think I should explain what is and isn't meant by quantization noise. You appear to be thinking in terms of a steady noise floor, which it isn't inherently.

Imagine a 16-bit A/D converter that's perfect (no noise, no non-linearity of any kind) except that it also has no dither. Say you feed it a steady, pure 400 Hz tone at a level so low that only the lowest-order three bits of the A/D toggle on and off in response. There are only eight possible signal levels available from three bits, so you definitely have the famous stairstep problem; at any given moment, there's likely to be a rather significant inaccuracy in the sample value relative to the (very low!) amplitude of the tone--but numerically, that inaccuracy is always less than the value of your lowest bit, so you're stuck with it.

Let's stop at this point and look more closely at that inaccuracy as if it were a signal all its own--an "error signal". Subtract the original, pure sinusoid from the stairstep result (or vice versa, doesn't matter which) and you'll get basically a patterned series of broken curves peaking just below the level of the lowest-order bit. I say patterned because it's not at all random; its shape and content will vary cyclically as the sinusoid's momentary amplitude passes through each of the eight available values that the system can represent exactly. So the error signal is a recurring (but very strange and "toothy") waveform at 400 Hz, plus it has rather strong overtones at integer multiples of 400 Hz. Thus you can think of it as noise or you can think of it as distortion. Neither view is wrong, nor do the two views have to exclude one another.

If you were to view the frequency content of this A/D's output on a spectrum analyzer, you would see a vertical line at 400 Hz, and further vertical lines at 800 Hz, 1200 Hz, 1600 Hz and so on, with their heights gradually decreasing toward higher frequencies, but continuing several octaves for sure. It's messy. What dither does is to take that error signal and randomize it. Not drown it out, since the dither is applied to the analog signal BEFORE quantization, so the patterned noise/distortion never has a chance to occur in the first place.

If you were to repeat the experiment and gradually inject dither in slowly increasing amounts prior to quantization, what you'd see on the spectrum analyzer is quite remarkable (I saw this done, and intellectually I knew what was going to happen, but it was still remarkable to experience it). At first you see what I described earlier--the forest of regularly-spaced vertical lines. As the dither creeps up in level, you can see a tiny, tiny turbulence that it causes in the noise floor--but the vertical lines above 400 Hz, which are well above that noise floor, start to get shorter and shorter while the 400 Hz line stays the same height. And when the optimal level of dither is reached, all the other vertical lines have shrunken to where you can't see them any more. It's not that the noise floor has risen to their level--rather, the distortion components have decreased (a lot farther and a lot faster!) to below its level.

So quantization noise is also known as quantization distortion, and it basically is that "error signal". It will consist of benign noise when dither is properly applied, while it will be program-dependent noise (that calls huge attention to itself and sounds grainy/granular beyond what you would ever expect, given how low in level it is on an rms basis) when dither is absent or insufficient, etc.

--best regards
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 12, 2020, 08:55:07 AM
Thank you DSatz. It's kind of you to explain all this clearly. Also thanks for the alert for the dither. I used to remember the lesson that we have to use the dither always when we change the bit's depth. But I forgot about it. I thought that my DAW (Reaper) internally calculate in 32bit and dither automatically. Looks like no, I have to check it.

Let's stop at this point and look more closely at that inaccuracy as if it were a signal all its own--an "error signal". Subtract the original, pure sinusoid from the stairstep result (or vice versa, doesn't matter which) and you'll get basically a patterned series of broken curves peaking just below the level of the lowest-order bit. I say patterned because it's not at all random; its shape and content will vary cyclically as the sinusoid's momentary amplitude passes through each of the eight available values that the system can represent exactly. So the error signal is a recurring (but very strange and "toothy") waveform at 400 Hz, plus it has rather strong overtones at integer multiples of 400 Hz. Thus you can think of it as noise or you can think of it as distortion. Neither view is wrong, nor do the two views have to exclude one another.
I like this sentence. I would slightly improve it:
Neither view is wrong, neither view is right, nor do the two views have to exclude one another, nor do the two views have to complement one another.
I found pictures of what DSatz describes: http://www.skillbank.co.uk/SignalConversion/quanterror.htm (http://www.skillbank.co.uk/SignalConversion/quanterror.htm)

I record music in pubs or clubs. There it is possible to hide many of my mistakes behind not ideal sound, background noise, etc. Those who record in a strict environment, for them, these mistakes are much bigger than for me. But the theory is also important to me because when I am recording then it is more interesting. And everyone records sometimes in a critical environment.

Heathen, you are right. Also for me, this thread is like a free graduate school course in digital audio.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 12, 2020, 02:38:59 PM
Digital signals DO have data outside of that bandwidth of 0-20 kHz. They actually have a lot of information outside of that range defined from -infinty to +infinity, and even more technically the baseband has info from -20k to +20k (don’t think too hard about that one...) but this is another conversation for another day.

Would you be so kind as to explain this in greater detail? If not here then in a new thread? I can understand why this would be the case in the mathematical realm, but I can't wrap my head around the idea of negative Hz in the physical realm. It seems, basically, impossible to me.

So, I wanted to pick back up on this since I never got around to answering it.

I spent a few days drawing up proofs and theorems, and realized most of the math would likely confuse more than it would assist. So I want to try simplifying the concept a bit more, before going "down the mathematical rabbit hole" - which I am happy to do, but want to use as a last resort here.

When you look at a frequency response plot, you are actually looking at the magnitude frequency response plot. This is because the way we "remap" data from the time domain to the frequency domain is through the Fourier transform. The Fourier operator is very useful, and the mapping variable for that is actually a complex (as in, sqrt(-1) complex/imaginary) exponential. The Fourier transform of a signal looks like:

(https://latex.codecogs.com/gif.latex?F%28%5Comega%29%20%3D%20%5Cint_%7B-%20%5Cinfty%7D%5E%7B%5Cinfty%7D%20f%28t%29%20e%5E%7B-%20j%20%5Comega%20t%7D%20%5Cdelta%20t)

where F(w) is the frequency content, f(t) your test function, j is the sqrt(-1) quick historical side note: mathematicians use i for sqrt(-1), electrical engineers already use i for current so j gets the sqrt(-1) honors in our equations w or omega is the angular frequency (2 * pi * f), t is time, and dt is the differential operator with respect to time. NOW... what this means, is that you are actively re-mapping the data using that exponential construct. When you are looking at a frequency response plot (for example, from a manufacturer) you are often looking at the absolute value of this function, and it is frequently limited from 20 Hz - 20 kHz, since looking at data outside of that is rarely relevant.

BUT... what happens when we evaluate the magnitude spectrum outside of that range? Well, the negative frequency range can be though of as an expression or manifestation of the negative phase of waves. I don't say they are explicitly representative of both positive and negative phases of the waves, since that has implications for power analysis (which is another topic for another time...). But that is part of how you can wrap your head around what the negative frequency data is getting at - when you take the absolute value of a part-real, part-imaginary number, you get the magnitude by using pythagorean theorem on the real and complex parts to get the "total magnitude" effect of a system on the magnitude-frequency spectrum.

THAT is often what we are looking at.

Phase response is something that is harder to explain without a bit of experience, I'll leave it off the table and comment for now that when I'm doing engineering work about 75% of the time the real answer to debugging my systems comes from the phase response and not the magnitude response. But it's something that hurts to wrap your head around, especially if you're not familiar and comfortable with imaginary numbers.

I don't want to distract too much from the topic at-hand, but as a fun thought experiment, what happens mathematically speaking if we remove the negative frequency domain information, so that our signal is exclusively 20 Hz - 20 kHz, and NOTHING else? Or, what if it's 0 Hz - 20 kHz? Well to construct a time-domain signal for which the above Fourier transform holds true, you actually need an infinitely long signal stretching out to before "t = 0" - which means you have a non-causal system that requires knowledge of events before measurement as well as theoretically infinite energy. These signals do exist, but they often are not represented by simple frequency domain analysis - you need to expand to the aforementioned power analysis to get an idea of what's up. I will leave that to another topic.

Now... everything I have mentioned thus far is for analysis of a finite, causal (meaning, no info is processed before we "hit the record button"), band-limited, analog signal. We can think of the magnitude spectrum of our ideal audio signal plotted in frequency as a rectangle - flat magnitude with unit 1 between -20 kHz and +20 kHz, and zero outside of that.

Sampling has this interesting property in the frequency domain in that it copies that magnitude spectrum and pastes it infinitely in either direction, centered around the sample rate. So for a quick thought experiment, let's sample the signal at 50 kHz. The bandwidth of our signal is 40 kHz, so for all real integer n * 50 kHz we see the frequency spectrum pasted. This means we have our nice rectangle from -20 to + 20 k, centered at 0 Hz, pasted and showing up centered at 50 khz from +30 to +70 kHz, centered at 100 kHz from +80 to +120 kHz, etc. going in the positive direction, and pasted again going in the negative direction centered at -50 kHz and going from -70 to -30 kHz, centered at -100 kHz going from -120 to -80 kHz, and so on.

A semi-reasonable graphical interpretation looks like this, except imagine the content is a flat rectangle (it's drawn as-such for other mathematical reasons not worth getting into right now...):

(https://i.stack.imgur.com/I6MGC.jpg)

Sampling theory in the frequency domain insures that the negative data from a copy does not interfere with the upper-end of the positive spectra of our original baseband signal (so that the first copy that starts at +30 kHz does not interfere with our data going up to +20 kHz). IF we were to lower our sample rate to, say, 35 kHz, then the data from +15 to +20 kHz in our original signal gets distorted by the copy in the frequency domain that is a result of sampling.

Another good illustration that google is able to provide quickly:

(https://www.tutorialspoint.com/signals_and_systems/images/perfect_sampling.png)

BUT WUMBO! Why then do we sample at 44.1 kHz? Why that extra 4.1 kHz?

Well, this is all still theoretical. Again, our digital signal theoretically has information above 20 kHz and below -20 kHz; when it gets converted back to the analog domain, that info is still there. To prevent that energy from sending signals at frequency ranges outside of what our equipment is rated for, we generally low-pass filter the data so that back in the analog domain we ONLY have data from -20 kHz to +20 kHz. We cannot do that in the digital domain, ONLY the analog domain. And real-world filters can be pretty sharp and steep, but they have implications - especially in the phase spectrum. To insure minimal distortion, that extra 4.1 kHz of sampling is a bit of a buffer to assume audio equipment may not have the sharpest filters.

I know I threw a lot of conceptual and cerebral info out here, and I know that some of it may not be thoroughly explained or concretely approachable. Feel free to ask follow-ups, this stuff takes everyone a while to fully wrap their heads around.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 13, 2020, 07:26:58 AM
^ Thanks for this post! I have some knowledge of Fourier transforms, so it even (kind of) makes sense. I will have to read it again, and think about if for a minute, but this is what I was getting at when I said I could only conceive of it mathematically and not physically.

I also appreciated the note on imaginary numbers. I have always seen, and been taught, that i is the square root of -1. It's interesting to see how the same thing often has different notation in different fields...
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 14, 2020, 05:16:21 AM
It is very nice explanation! It's nice when someone can simply explain in words a theory that is based on complex mathematics. These are explanations of the details, but when I create at least a rough idea of these details and what is happening when sampling and quantizating, it's useful. From a practical point of view, it may not be necessary, but it's good to know what I do and why I do it when I record. Then recording is much more interesting for me.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: seethreepo on October 14, 2020, 12:34:07 PM
I kept up with page for the first few pages but I could now use a Sample rate for dummies version :laugh:
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 14, 2020, 12:38:42 PM
Another comment I want to make here:

DSatz is as expected 100% right on the money in all of his posts.

However, to avoid a VERY commonly confused misconception... stair stepping is not exactly a 100% accurate representation of the signal in the digital domain, nor is it necessarily a representation of the signal from the digital domain back into the analog domain. It's a useful visual tool, but does not inherently or necessarily exist in either the digital signal or the converted-to-analog signal.

The expansions of points in this video run slightly counter to the advice I tend to give, but it's a good explanation and overview of LOTS of concepts realized and visualized in DSP. I NEED to reinforce this here, for you to keep in the back of your mind while watching this video - ignoring any differences in the actual performance of a digital to analog converter, it is highly unlikely you would sonically hear an appreciable or noticeable difference between 16/44.1 and 24/96 (or higher) ON PLAYBACK for rock music. There are other reasons outlined earlier in this thread for recording at higher bit depths and sample rates that I encourage.

https://www.youtube.com/watch?v=cIQ9IXSUzuM
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 14, 2020, 01:01:33 PM
I kept up with page for the first few pages but I could now use a Sample rate for dummies version :laugh:

Here's the core of sampling theory, in layman's terms.

The overall question we want to ask is: if we have information in the analog world and we want to somehow sample the information, how quickly do we need to capture each individual "slice" or "moment" of the analog information, so that when we try to reconstruct the original analog world from our slices/moments we can do so without distortion?

The practical analogy for this that's often given in signal processing classes is taking photos of a tire. If you watch videos of a wheel accelerating, at first it looks like you would expect it to - the wheel is turning, and each successive frame the spokes follow one another in a logical pattern. However, if the wheels start turning too fast, then the spokes complete more than what our brains can process as a "logical turn" and it looks like the spokes are moving backwards. This is because the rate of spinning of the wheel is over half as fast as the frequency in which you are taking pictures. So sampling theorem says: you need to take at least two photos for each fastest turn of the wheel.

Now the topics that are involved in this thread are a follow up to this concept. Once you are over that "two photos per wheel turn" do you get an appreciable difference? Given that it takes up more space to take more photos, and that most people may not hear a significant difference, is it really worth our time/effort/energy to capture the information any faster than twice-per-fastest-wheel-turn?

My arguments in this thread are effectively "if you're just listening to the raw information, it's pretty unlikely. But it's still useful to take more photos per fastest wheel turn, because let's say some video engineer in the future wants to sharpen your photos by individually processing red, green, and blue (an analogy here - please don't get hung up on the details of this precise example) portions of your image. The more photos you have taken, the more accurate each individual color can be processed such that the end transformation and re-addition that a post-production video engineer has done will be cleaner if you've taken more photos, because red/green/blue exist in different frequency bands and interact differently based on their properties."

That's a VERY high-level summary of what I've been trying to get at here for recording at 24/96 (or higher). At the end of the day, for playback you are totally fine listening to and distributing online 16/44.1 data. However, if you have an easily-accessible archive of 24/96 or 24/192, it allows post production people doing remixing work or mastering work a LOT more precision to do their magic, and the end result of the signal processed with filters at 24/192 that is then bounced down to 16/44.1 sounds significantly better (to my ear, at least) than info that was bounced down then processed at 16/44.1.

And this extends to technologies that aren't exactly mainstream - I'll point to the major tech companies investing in spatial audio filtering and processing. So just because you may not appreciate any benefits from higher sample rates doesn't mean you shouldn't still do it.

Does this clarify things for you?
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: seethreepo on October 14, 2020, 11:14:46 PM
 Thanks for dumbing things down a little for me. My comment  was mostly tongue  in cheek. I think I have a decent grasp on things. I was lost when the math equations came up and I don’t think any amount of dumbing down the math will help, as I just don’t have the patience for it.
Cheers
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 15, 2020, 10:18:02 AM
It is very nice video. Wforwumbo, I understand. It is good to think of the output analog signal as the original analog signal + frequency bandwidth limitation (due to sampling) + quantization noise (due quantization). And the imagine of ​​stairs may be useful when we are modifying the digital signal in our DAW.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 15, 2020, 11:11:36 AM
That's a VERY high-level summary of what I've been trying to get at here for recording at 24/96 (or higher). At the end of the day, for playback you are totally fine listening to and distributing online 16/44.1 data. However, if you have an easily-accessible archive of 24/96 or 24/192, it allows post production people doing remixing work or mastering work a LOT more precision to do their magic, and the end result of the signal processed with filters at 24/192 that is then bounced down to 16/44.1 sounds significantly better (to my ear, at least) than info that was bounced down then processed at 16/44.1.

I am not sure whether or not you were being sarcastic, but this cracked me up! I am a taper, a hobbyist. My "post production people" are, well, me. I assume that is true for most of us here; it definitely is for the people I have discussed this with. I think, in general, tapers are pretty minimalist with respect to post-production. Many trim the ends and normalize and that's about it. Others (I am in this group) do some (typically light) EQ'ing, maybe some limiting for stray peaks, some noise removal, and normalization. A few go further, but, even in that group, we are often constrained by the limitations of the type of recording we do.

I would wager that for me (and probably many other tapers) the audible benefits of processing at 24/96 versus 24/28 are pretty marginal. In my case, I doubt it is worth doubling the file sizes. The standard rejoinder: "Storage these days is cheap!" Well, maybe, but I have a couple of drives in my tower and several more in my NAS. I upgrade (and add more) every couple of years and it is not inexpensive in my mind. On top of that, I have a cloud backup option that also costs quite a bit per year and will cost more as I increase the capacity.

I would be curious to hear your comments on rates greater than 96 kHz. Lavry, for example, says that there is a trade-off between speed and accuracy and that the optimal sampling frequency is around 60 kHz, with 88.2 or 96 being the closest typically available rates. [Also perhaps of interest to you is his way of describing the time/frequency relationship using sinc instead of Fourier functions.]
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: voltronic on October 15, 2020, 12:38:23 PM
Some of you may be familiar with this website, which compares the performance of many different software sample-rate converters when converting from 96 kHz to 44.1 kHz.  See the Help section for details of the methodology and why they chose those rates to convert.  Some look nearly flawless; others atrocious.  Could you tell the difference in a blind test?

https://src.infinitewave.ca/ (https://src.infinitewave.ca/)
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: wforwumbo on October 16, 2020, 01:38:36 PM
Thanks for dumbing things down a little for me. My comment  was mostly tongue  in cheek. I think I have a decent grasp on things. I was lost when the math equations came up and I don’t think any amount of dumbing down the math will help, as I just don’t have the patience for it.
Cheers

I'm not sure if it's really dumbing it down, so much as it is trying to explain the philosophical purpose and implications of the theories underlying the tech. If anything, it's trying to synthesize a more general notion of the theory, which I would argue is more challenging than trying to understand the math. If I can't level the concept with everyone eye-to-eye and cannot effectively communicate its theory, I don't fully understand the concept. The math is more of a way of trying to objectify the theory and trying to manipulate it to understand its limitations, not the de-facto ground truth from which the theory happens to stem (a common mistake amongst engineers, and the very plague which drives physics and ironically enough drove me away from studying raw physics in college).

It is very nice video. Wforwumbo, I understand. It is good to think of the output analog signal as the original analog signal + frequency bandwidth limitation (due to sampling) + quantization noise (due quantization). And the imagine of ​​stairs may be useful when we are modifying the digital signal in our DAW.

Sure, cheers. Just remember: it's an analogy for what's happening - not what's actually happening under the hood. Shoving samples around is more art than engineering (and in practice you don't actually shove the samples around, but that's outside the scope of this thread), it's a shame it's so much fun once you get it right because getting it wrong is like pulling your own teeth out.

That's a VERY high-level summary of what I've been trying to get at here for recording at 24/96 (or higher). At the end of the day, for playback you are totally fine listening to and distributing online 16/44.1 data. However, if you have an easily-accessible archive of 24/96 or 24/192, it allows post production people doing remixing work or mastering work a LOT more precision to do their magic, and the end result of the signal processed with filters at 24/192 that is then bounced down to 16/44.1 sounds significantly better (to my ear, at least) than info that was bounced down then processed at 16/44.1.

I am not sure whether or not you were being sarcastic, but this cracked me up! I am a taper, a hobbyist. My "post production people" are, well, me. I assume that is true for most of us here; it definitely is for the people I have discussed this with. I think, in general, tapers are pretty minimalist with respect to post-production. Many trim the ends and normalize and that's about it. Others (I am in this group) do some (typically light) EQ'ing, maybe some limiting for stray peaks, some noise removal, and normalization. A few go further, but, even in that group, we are often constrained by the limitations of the type of recording we do.

I would wager that for me (and probably many other tapers) the audible benefits of processing at 24/96 versus 24/28 are pretty marginal. In my case, I doubt it is worth doubling the file sizes. The standard rejoinder: "Storage these days is cheap!" Well, maybe, but I have a couple of drives in my tower and several more in my NAS. I upgrade (and add more) every couple of years and it is not inexpensive in my mind. On top of that, I have a cloud backup option that also costs quite a bit per year and will cost more as I increase the capacity.

I would be curious to hear your comments on rates greater than 96 kHz. Lavry, for example, says that there is a trade-off between speed and accuracy and that the optimal sampling frequency is around 60 kHz, with 88.2 or 96 being the closest typically available rates. [Also perhaps of interest to you is his way of describing the time/frequency relationship using sinc instead of Fourier functions.]

I came into taping as a studio guy - I have been doing studio production since high school, and the notion of applying my knowledge to my favorite tapers' work is how I got hooked. Eventually I got looped in by the raw sound of tapes and how to perfect that, since the studio adage goes that it's better to spend 20 times getting the right take and applying gentle (at most) filtering sounds infinitely better than spending hours trying to fix a subpar recording. I can say that I've learned LOADS from taping that has made me a much better studio engineer, but I'd like to flip that coin on its head and mention that lots of tapers could learn a thing or two from post production guys to make a better-sounding end product. Even if it's not for "cleaning up a recording," post production requires a careful ear that helps you tune your mic setup to optimize the amount of work you have to do after the fact, and the reason I started (and still continue) doing it is because that final layer of polish can be the difference between "yep this is great music" to "the veil of realism has been drawn, I can sit back and enjoy this recording without the tape getting between me and the music." I would counter-argue that the constraints from limitations of the recording is limited more by taper knowledge and experience toying with post production (followed by frustration of not putting in the time to learn how to equalize and do ear training), rather than a limitation of the medium or method of recording. Just like making a good tape, learning to apply effects in the post chain takes experience and lots of getting it wrong, before you learn how to get it right; and lots of studio guys would stick their noses in the air at how/why we do things, without understanding how it would make their lives way easier.

The difference isn't marginal - it's pretty significant. I really DO need to get sound clips up here, as it'll prove my point better than words. And likewise, it's something I encourage you to try out on your own next time you make a tape, or grab some 24/96 tapes that are uploaded (most of my tapes are up on eTree as 24/96, happy to get you a copy if the torrents are dead). Play with it yourself by doing the filtering at 24/96 and bouncing down to 16/44.1 after processing, versus doing the identical processing at 16/44.1, and see if you really can or can't hear a difference. That does more talking than any amount of typing I can do behind a screen.

I've taped and tracked at 24/192 before, personally it's past the practical limit given wav files can only contain 4 gigs, and given I run 4-6 channels it's just too much bouncing individual tracks/merging them to create sets. 24/96 I can usually squeeze a set of music running 4 channels without needing to splice. The benefits between tracking at 96 and 192 are there to my ear - I've picked them out when double-blinded - but the filter error is low enough to me at 24/96 that I care a little less about the bump up to 192. If there were tech to make my life easier to track and splice/join stereo pairs of tracks at 24/192, I'd do it in a heartbeat and without batting an eye.

I don't think there's a real "optimal" speed/accuracy tradeoff curve for sampling. There are just too many ways to process audio for there to be a magical number for it to boil down to. You can halve your buffer size if you double the sample rate, and have the same effective latency as the same number of samples-per-unit-time are being processed. Seeking an optimal rate is imo a bit of a fallacy - just go as fast as you can accurately and conveniently sample once you are past Nyquist, which is the point of my posting here. If that's 24/48, so be it.

Curious what you mean about using sinc instead of Fourier "functions" - the Fourier transform listed on the last page is a synthesis function that decomposes a signal into a summation of cosine and sine waves that exist within the time signal. The sinc function [a sine divided by the index of the sine, so sinc(x) = sin(x) / x] pops up a whole bunch in filter design engineering with good reason, but I've never seen it used as a domain transform variable.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 16, 2020, 02:15:51 PM
I would counter-argue that the constraints from limitations of the recording is limited more by taper knowledge and experience toying with post production (followed by frustration of not putting in the time to learn how to equalize and do ear training), rather than a limitation of the medium or method of recording. Just like making a good tape, learning to apply effects in the post chain takes experience and lots of getting it wrong, before you learn how to get it right; and lots of studio guys would stick their noses in the air at how/why we do things, without understanding how it would make their lives way easier.

Fair enough, but, again, I do this as a hobby. I have a (more than) full-time job, a family, and other hobbies. I make recordings that make me happy and that's pretty much my only goal. I feel like my post-production skills have gotten much better over the years; I could definitely go back and improve on some of my previous efforts, but I also know I will likely never find the time to do that. Maybe when I retire and the boy is in college, by which point my hearing will probably be shot anyway...

Play with it yourself by doing the filtering at 24/96 and bouncing down to 16/44.1 after processing, versus doing the identical processing at 16/44.1, and see if you really can or can't hear a difference.

I don't think anybody is actually doing that, though. I process at the depth/rate with which the recording was made. The conversion to 16/44.1 is basically the last step (except for tracking).

I don't think there's a real "optimal" speed/accuracy tradeoff curve for sampling.

Curious what you mean about using sinc instead of Fourier "functions"

Like I said, those things are from Lavry, an engineer who makes ADCs. The trade-off, as I recall, is due to physical limitations of the analog circuitry. Again from memory, the sinc/Fourier thing had something to do with periodicity in the Fourier that wasn't in his sinc explanation. I am not 100% certain anymore, as I read his papers quite some time ago...
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: Gutbucket on October 16, 2020, 05:29:38 PM
Some of you may be familiar with this website, which compares the performance of many different software sample-rate converters when converting from 96 kHz to 44.1 kHz.  See the Help section for details of the methodology and why they chose those rates to convert.  Some look nearly flawless; others atrocious.  Could you tell the difference in a blind test?

https://src.infinitewave.ca/ (https://src.infinitewave.ca/)

It is Halloween season.  Those scary looking sweep images may appear more frightful than they perhaps should be without an appropriate frame of reference.  Which is not to say there is no reason to use the better performing resampling routines, only that it can be challenging to equate them in a meaningful way with perception and appropriately rank the importance of all this in the hierarchy of all the other stuff we need be concerned with.  Be aware, but don't fret too much and frighten oneself into neuroses.  Make sure your batteries are charged and the SD card has enough space.

From the FAQ (https://src.infinitewave.ca/faq.html) at that site:

"Are most SRCs really that bad?
No. If you look at the decibel scale to the right from the graphs, you can see that the range of these graphs is very wide: down to -180 dB. The distortions generated by most properly designed SRCs are below -100 dB and can hardly create audible artifacts. However SRCs differ in the transition band of the low-pass filter and in the amount of pre-/post-echo and aliasing. The bottom line is that most tested SRCs range from fairly good to excellent, but the graphs are very sensitive to emphasize the differences."


Interesting to revisit that site again years later.  The thing most interesting to me now is looking at how many earlier versions of the same software performed considerably more poorly than their more current releases.

Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: Gutbucket on October 16, 2020, 05:30:09 PM
Great discussion in this thread.  Thanks to all participating.

All the resampling talk gets me thinking of fancy dither routines.  OT to sample rate conversion, but not by much as it tends to go hand in hand with bit-depth reduction.  Has anyone here ever noticed the difference between the use of noise-shaped dither verses standard triangular dither in any actual real world listening situation? That is to say, without first amplifying to totally non-realistic listening levels at which no one could actually listen?  I feel good using standard triangular dither not only because I can't hear any difference in real world playback, but also because of those future generation mastering remixers who will be mining and combining all of our current stone age pulls.  ;)
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: kuba e on October 17, 2020, 07:05:00 AM
I don't think anybody is actually doing that, though. I process at the depth/rate with which the recording was made. The conversion to 16/44.1 is basically the last step (except for tracking).
I was doing this  ;D. I was saving my recordings 24/44.1 as 16/44.1. Thanks to our friends who explained the quantization noise and the dither, I think I can say that my approach was not wrong with a little tolerance. Yes, this is not a pure procedure but the range of our audience recordings does not exceed 96db and the background noise is also high. By the way, my Nak300 have listed S/N as "higher than 50db weighted".

I would counter-argue that the constraints from limitations of the recording is limited more by taper knowledge and experience toying with post production (followed by frustration of not putting in the time to learn how to equalize and do ear training), rather than a limitation of the medium or method of recording.
I think this is a misunderstanding that Aaronji didn't mean it that way. My view is that most people here do audience recordings and these recordings are little bit rough. It's more difficult to learn to distinguish subtle details on rough recordings than on fine studio recordings. The learning is easier in the studio. We record as a hobby for joy. But this does not mean, that we don't appreciate any advice regarding theory and practice. I am grateful for your advices. They are valuable to me and broaden my horizons.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: voltronic on October 17, 2020, 07:43:00 AM
Some of you may be familiar with this website, which compares the performance of many different software sample-rate converters when converting from 96 kHz to 44.1 kHz.  See the Help section for details of the methodology and why they chose those rates to convert.  Some look nearly flawless; others atrocious.  Could you tell the difference in a blind test?

https://src.infinitewave.ca/ (https://src.infinitewave.ca/)

It is Halloween season.  Those scary looking sweep images may appear more frightful than they perhaps should be without an appropriate frame of reference.  Which is not to say there is no reason to use the better performing resampling routines, only that it can be challenging to equate them in a meaningful way with perception and appropriately rank the importance of all this in the hierarchy of all the other stuff we need be concerned with.  Be aware, but don't fret too much and frighten oneself into neuroses.  Make sure your batteries are charged and the SD card has enough space.

From the FAQ (https://src.infinitewave.ca/faq.html) at that site:

"Are most SRCs really that bad?
No. If you look at the decibel scale to the right from the graphs, you can see that the range of these graphs is very wide: down to -180 dB. The distortions generated by most properly designed SRCs are below -100 dB and can hardly create audible artifacts. However SRCs differ in the transition band of the low-pass filter and in the amount of pre-/post-echo and aliasing. The bottom line is that most tested SRCs range from fairly good to excellent, but the graphs are very sensitive to emphasize the differences."


Interesting to revisit that site again years later.  The thing most interesting to me now is looking at how many earlier versions of the same software performed considerably more poorly than their more current releases.

Yes, I noticed the same thing about older versions of the same software.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: jerryfreak on October 17, 2020, 05:44:39 PM
wow thats quite illuminating. ive been using the izotope plugins in soundforge instead of the default SRCs and im glad i did!
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: DSatz on October 28, 2020, 10:46:52 AM
kuba e, thanks for quoting the signal-to-noise specification from your Nakamichi deck. Cassette tape was invented by Philips to improve the convenience and performance of dictation machines. As a teenager I carried around my Norelco Carry-Corder 150 onto which I had recorded (via the microphone) some classical records that kept me relatively sane, and I held the speaker up to my ear to listen while traveling, walking around, etc.; when people started recording music on the system, I think a lot of people at Philips were pleased, but also a bit shocked.

So much for "analog has infinite resolution" ...

The attached screen shots are from an AES paper from the current (virtual) convention. The main prior study was criticized (rightly in my opinion) for using source material that hadn't all been natively recorded 96/24. In this study, the author used pure 96/24 recordings of various genres vs. 44.1/16 versions of the same recordings. The participants included musicians, audiophiles, professional recording engineers and producers, as well as people who simply enjoy getting laid, I mean lay people. They were allowed to listen as long as they wanted, play as loud as they wanted, use their own or other equipment for playback, use headphones, loudspeakers or wax paper and a comb if they wanted--any listening kind of thing that they wanted to do, except open the file on a computer to look at its waveforms or properties. The results weren't significantly better than what you'd expect from a coin toss. More experienced listeners (including the professionals) were no better on average at identifying the differences than less experienced ones. Age made no difference, either.

As far as the experimenter's possible bias is concerned, he is the owner of a recording company that makes and sells 96/24 recordings; this study makes him "look good" only in that he is willing to publish results that don't cater to his commercial interests. (In my book, that truly does make him look good.)
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: guitard on October 28, 2020, 12:23:52 PM
Cassette tape was invented by Philips to improve the convenience and performance of dictation machines.  As a teenager I carried around my Norelco Carry-Corder 150

I forget the details of the recorder, but back in the mid '70s, I used to borrow my dad's cassette tape recorder that he used for dictation at his office.  I recorded anything and everything with it:  Don Kirchner's Rock Concert and The Midnight Special (by putting the recorder next to the tv's mono speaker).  I took it to school and recorded conversations with friends.  I pretended to be a basketball game announcer and recorded neighborhood basketball games.  I used it to record my first ever concert recording in 1977 when I recorded one-hit-wonder Ram Jam (Black Betty was their hit song at the time) by holding the cassette player above my head about 15 rows dead center FOB (that recording actually turned out surprisingly well).

The sad part is ... I didn't preserve any of those recordings ... typically I just kept recording over the same cassette over and over.  At the time, I didn't have the foresight to think about how fun it would be to listen to those recordings 40+ years later.
Title: Re: 48 kHz vs 44.1 kHz sample rate
Post by: aaronji on October 28, 2020, 01:04:15 PM
The attached screen shots are from an AES paper from the current (virtual) convention.

I would love to read the whole paper; is it available anywhere under an open access license? AES wants $33 for a download.