The Depths Of My Bits And The Rates Of My Samples...
From the original Analog Industries archive. Some links and images from that era no longer resolve.
Once I got past this dude’s beard and the ImageLine logo that they splashed on someone else’s video (the original isn’t embeddable), I quickly realized that this is the single best explanation of bit depth, sample rate, and dither that anyone has bothered to make.
I’ve always thought that 48kHz (which puts the Nyquist frequency above 100% of the human race’s hearing ability) and 24 bits (which puts the noise floor below that of most any instrument, recording method, and reproduction method) were perfectly acceptable values. This video pretty much says that’s fine, although not for the reasons I thought.
Anyhow, if you’re a musician that records digital audio, you owe it to yourself to spend a half hour watching this.
38 comments
Good video. Although he's totally trying to gloss over the fact that my new speaker cables, made of copper that has never felt a human gaze, open up the soundstage by adding more analog to the stair steps.
Besides the incredibly distracting beard, I thought he was saying "Ziff", as in "Ziff-Davis" (which makes me an old computer geek) instead of Xiph. But those people seem to having a general naming problem; Ogg Vorbis?!? That name is probably the single biggest reason no one ever used that format.
And while I get the whole "supporting open codecs" thing, putting a video up webm or ogg video is ensuring large swaths of internets will never watch it.
Thanks for posting this. As a quantum of the above mentioned "large swaths", I'd never have seen it otherwise.
I've been dithering tracks for years without ever having had anything other than a vague idea that it's something I should be doing. It makes me wonder what else I've been doing based on partly-absorbed information. Not just in music, in general.
And the article he wrote (which he references) seals the deal on 48/24 for capture/mix/master and 44/16 for playback, forever. Read it, understand it, bookmark it.
https://xiph.org/~xiphmont/demo/neil-young.html
Also, Gearslutz should just redirect to the article and video anytime anybody posts something remotely related to this topic.
Awesome video. It's nice to see this companion to his fantastic article from last year on the silliness of 192/24 downloads.
THE BEARD WANTS ME TO SURRENDER!
- c
Between the beard and the peculiar shirt, I got out of focues....I think to need to watch it again
I've seen that video before but watched it again because for a nerdy-type, he's an entertaining presenter and actually knows what he's talking about (I say this as one of the nerdy-type, too).
I've recently had this conversation with a couple of other engineer friends and we've all come to the conclusion that 24/48 is absolutely fine for all of our recording and mixing needs. We aren't mouthbreathers hanging out on Gearslutz debating which dither shape "sounds better", but guys that actually make a living doing this crazy audio business and getting paid by labels and artists. 96kHz/192kHz/DSD - none of it means ANYTHING if no one wants to listen to your song.
Watch eBay prices on analog test equipment soar. Thanks, BeardDude.
A lot of the reason that high-sample-rate dudes crack me up (especially the ones on the ass end, talking about SACD and FLAC and this and that) is this: the Capricorn is the best board Neve ever made.
On paper.
In real life, when you see "Neve room," it's never a fucking Capricorn, is it? People'd rather use some janky old board that had whiskey spilled in it, and all the faders are scratchy from the cigarette ashes they're full of. For "vintage warmth." The Capricorn was Neve's biggest commercial failure, and it almost took the company down.
And that old janky board is what was used to mix that SACD CD.
-CR
Beard knows best.
he's totally awesome . i saw this a while back - being somewhat more geek-oriented than at least a few posters here, it was nice to see/hear Monty in the flesh. Xiph has pioneered a lot of things, and if you're not a Terry Pratchett fan, you won't get the Vorbis reference, but beyond that is FLAC, Theora/WebM, and XviD. and everything is totally open source. he's a hero in my book, and OggVorbis is actually used a lot in game audio because encoding is license free.
the thing i found the most eye-opening was when he described A/D conversion as not being like the stairstep/chiclet thing that i'd been taught and seen in diagrams countless times. but i'm still not totally sure what he is referring to, being not that much of a math or EE geek.
oh yeah - Gearslutz - where dudes with literally Zero musical ability will wax on and on about their handmade custom mic pre with nickel-wound transformers while the 192KHz sound samples they provide show that they are unable to even play an instrument properly to demonstrate the soul-searing beauty of their open, ballsy, yet transparent sound...
>In real life, when you see "Neve room," it's never a fucking Capricorn, is it?<
Because they're bankrupt. Those things were expensive. And plastic-y. Like most things, it was probably a lot better than I thought at the time, and perversely, now I kind of think it would be cool to have one.
you're in luck - seems to be a lot cheaper now...
http://www.ebay.com/itm/AMS-Neve-Capricorn-Digital-Mixing-Console-72-Faders-/360388576869
Great video and, yeah, a different facial-hairstyle would probably flatter this guy's face more.
Are there advantages for using higher sampling rates if you're going to manipulate the recordings?
Like, I've been mangling sounds by pushing them down two or three octaves and figure that recording at 96k will expose some of the detail that I wouldn't normally hear.
@metaphysician: Ironically, that is almost certainly the very Capricorn I mixed my 5th album on. With a Mitsubishi 32-track digital deck. State-of-the-art at the time. And he's almost certainly actually the second owner of it, so a bit of a fib in his description. I'm not personally aware of there being more than one Capricorn in Chicago. Unless he had it in his house and didn't take clients, I'd be shocked if it isn't the one from Chicago Trax.
@bassling: Too many variables. My initial response is "no." I mean, if you're dropping something 3 octaves and "manipulating" it, crystal clarity and perfect reproduction are hardly your aims, are they?
-CR
Thanks for the links. I read the article (faster than watching a video), and what he says about intermodulation distortion of ultrasonic frequencies is right on the money. I ran into that the very first time I tried to sample klang-y metallic objects at 96 kHZ. At first I was confounded, thinking I was hearing aliasing that Nyquist couldn't explain, but eventually I realized I was trying to shove ultrasonic frequency content through various devices and software algorithms that didn't deal with ultrasonics well. It's good to see somebody dropping science to indicate I wasn't crazy hearing those distortion components.
Yeah I got the same distortion problems when dropping the pitches, via slowing the sample rate, on quite a few samples done at 96khz.
2 things helped. Not using my normal field recorder Zoom H4 (duh!), but going with my RME Babyface and some DPA 4061 omnis.
Also, strangely, recording with a large-ish ambience reverb. Like a stairwell or school hall seems to make the dropped sample smoother.
@rollmottle: I've read that article a couple of times and I 75% agree with you, but I don't understand with why 48 for capture/mix/master instead of 44.1. I've gone back and forth between 44.1 and 48 a couple of times since CR first posted that article (reading it each time), but I landed at 44.1 because a) I think it probably doesn't matter much, and b) assuming it's going to end up at 44.1 for playback, why fuck around with something else?
I'm not trying to argue; I want to understand. I totally accept that there could be a reason for sticking with 48, I just didn't get it from that article.
@speak_onion: a couple of reasons - 48k is the standard for broadcast so there could be one less conversion you'll have to do. Pushing the brickwall filter up a little more to 24k gives a hair more top-end headroom and since most mastering guys are going to process through an analog chain and capture the results before making the final 44.1 master, the "stay at 44.1 until the end" argument is rather moot.
44.1 is kind of arbitrary; it puts Nyquist at 20.5, which is technically 500Hz above the nominal threshold of human hearing. The thing is that quite a few people, especially young people, can hear over 20K. 48kHz puts Nyquist at 24kHz, which is _definitely_ above what anyone can hear, even the most Golden-Eared Gearslut. So you might as well just do 48K and be done with it.
Regarding playback, are you talking about a situation where someone bought a CD, put it in a CD player, and listened to that playback directly? That's... quaint.
-CR
48 kHz gives a big more transition width to the antialiasing filter that you need in D/A conversion. A less steep filter can have less ringing.
This is assuming the "standard" antialiasing filter design, my understanding of which is probably up to the level of "Musical Applications for Microprocessors" which was published in 1980. I know that oversampling A/D and D/A convertors are a different issue, but I don't really know what is going on there.
The video is very good - I think it gives reassurance that the theory will always be valid but maybe not the implementation - eg. jitter wasn't really covered or the fine details of the reconstruction.
I think it's worth considering some of the info in this video:
http://www.youtube.com/watch?v=1CkyrDIGzOE#t=24m05s
"48kHz puts Nyquist at 24kHz, which is _definitely_ above what anyone can hear, even the most Golden-Eared Gearslut"
So what happens when I sample a musical sounding prop, and pitch it down just one octave (which really isn't a massive shift - i know plenty of musicians who detune instruments to make kick drums etc)
Surely that 24kHz Nyquist frequency becomes 12kHz, which is well within even an aging rock stars hearing range...
Maybe it doesn't matter if your mics don't capture anything above 20kHz, but plenty of props resonate above 20kHz (eg anything metal) and there are mics available that capture frequencies above 20kHz (MKH80X0, Schoeps etc)
Imho the correct sample rate more depends on what you intend to do in terms of processing and manipulation, rather than making hard rules about the delivery sample rate.....
Assuming your DAW is set to 48k, Nyquist is _still_ 24k, no matter what you're doing with your pitching-down-fun-and-games. Which is why you hear no aliasing "pitching down" but almost instantly "pitching up."
All that said, if you're trying to impress dogs or a TEDx conference, knock yourself out. Otherwise, go write a good hook and some witty lyrics and quit worrying about it.
-CR
Only a typo I'm sure, but isn't Nyquist for 44.1 22.05?
And something nice I've learned from this topic-
44100 is 2 squared times 3 squared times 5 squared times 7 squared
I've always worked at 44.1. There were always instances where we had to make a CD, or import something from a CD, or something, and now it's habit. For the last 20 something years people have debated this. How much of your favorite music from that time was recorded at 44.1k and how much at 48k?
That was great. Thanks for sharing.
@bassling At the risk of sounding like a gearslutz beard stroker I read a comparison of that method of pitchshifting a long time ago, and the dude that wrote it got a high res file to null against the same thing done from 44.1 using some expensive pitchshift plugins... whether there's any detail left three octaves down is another question, but at least yer saving money on plugins!
Well, as for the beard, maybe he does a bit of RAD (reverb algorithm development) in his spare time?
Sean?
This is a great video though. I remember seeing it a couple of years ago and it did indeed confirm my prejudices and confound my beliefs. I watched it a few times since and probably need to watch it a good few times more. But I got the gist of it. I think.
Shame, most noobs, when pointed in this direction won't take the time to even consider. If 16 bits is good, well 24 bits is better. And 32 bits is best of all.
And worst of it. The bits they thought were bits, weren't really the bits they thought were bits at all. You see, it's like an information super-highway. The bit-rate is your speed (max speed) and your bit-depth is your bandwidth (lanes on the highway).
Never mind. Forget I said anything. I'm sure I got something wrong somewhere. At least I know there is a difference, and in reality it makes little difference to me. And at least I know I don't know...
Jeez, this could go on all night.
Enough for a first post, I think.
:-)
Was it KMFDM that wanted or did record a record all at 22k?
I remember that same damn Neve board at Trax.... I think they spent a lot of money on that... Plastickey is the right term. Did a mix or two on it, but I was just producing, not engineering, but I remember there being some debate about why it was supposed to be so great, and others trying to justify the expense.
I have tried to stick to 24bit/44.1, but a lot of scoring gigs/commericals/games/tv etc peeps want all deliverables in 48k, as it is their standard. So depending on the project will set my SR accordingly (until i forget, and then I have to use some freeware SRC which probably does a horrible job, but at the point i'm doing that ((against the clock)) i rarely give a rats behind..)...
obviously in the past src abuse yielded some of the more interesting dsp that would leave people scratching their heads... Like the ol' trick of sampling a vocal into an emax at 22k, resampling it an octave higher, or double speed, then playing it back at it's intended time gave the best warbling aliasing due to the fact that the emax filters were pretty brutal in a good way... I'm sure the same effect could be handled in software in like 7 seconds now, but that was always good fun.
now the question I have for Chris and Adam is: with 24 bit, i like the fact that the added headroom allows me to not record as hot, and maintain decent s to n etc... and I am sick of getting tracks from colloborators etc that basically look like they are pinned to the red like you would have to do with an black face adat to avoid noise floor.... so long explanation unneccessary aside: When you code your plug ins, is it true that most plugs "react" better with lower levels feeding them.... i know use your ears etc, but with something like Rough Rider or any other compressor you would use, or EQ, is there real truth to the assertion that a lower level entering the plug gives the plug more headroom to do it's thing correctly? I have found personally that working with lower levels on channels, and making it up at the end of the 2 bus seem to give me more control, ... but since you gentleman actually MAKE and THINK about these things, just truly how important is gain staging into your products?
sorry for the long winded question.
On the subject of recording at lower levels, I think the answer is a qualified maybe. Proper gain staging still applies in a digital environment. So if you're trying to hit 0dBFS on every track you're probably turning the track faders down significantly at mix time, leaving things kind of backwards from where they ideally should be. The best practice would be to record everything at as close it needs to be in the final mix. Granted, that would be the practice for making an open, dynamic recording. That is not necessarily what people are trying to do now days, so you can do whatever you want to get the sound you're trying for.
As far a plugins go, every one has the potential for handling differently, but erroring on the side of additional headroom is more likely to give you the chance to get what you want than slamming everything is. and in a 24bit system there is rarely any reason you can't leave 6 or maybe even 12dB of headroom on an individual track.
You can always add gain. Better to leave the headroom, in my opinion. I personally record and mix with peaks at -6. I don't squish the living daylights out of my tracks, as I don't make radio or club music that has to compete, and I like them to be a bit more naturally dynamic, but that's personal taste. (Also, in the unlikely event that I get radio play, they have their own limiters.)
-CR
I prefer the open dynamic sound generally, and try and avoid overcompressing stuff on the 2 bus as much as I can possibly, with the exception being certain commercial gigs etc where you unfortuntately HAVE to to even be heard, ... I try and leave -6dB at least before going to mastering, and pay attention to RMS, something the kids these days seem to like a little a ratio as possible...:)
but thanks for your response gentleman, and I hear what you are saying re:taste Chris, and I know there were days where you had to crush the living mud out of stuff to stand out, .... I was just wondering if you find your plugins "behave" better, or in a manner more expected when feeding them "less" so to speak.... have read all sorts of dissertations about -18dBfs etc at 24 bit actually being optimum etc... and deal with a lot of mopes unfortunately who work off their laptops all the time, so tend to have everything in the red all the time so they can hear it out of their headphone jacks, which is a joke, .... cause it makes more work for me to "re-gainstage" stuff so it isn't all distorted to hell over the 2 bus, which they can't hear over headphones or care...
again, long explanation for a simple question: when you and adam design your stuff, and test it to see if it meets your desired expectations, do you find that lower levels fed yield more what you would expect or intend in your design process? (obviously regardless of the ancillary benefits of improper use of said tools that we all probably love to do:) ).
Well, if there's soft saturation in the plug-in it is gonna be a little crunchier the harder you hit it. And we use soft sat any time two signals are summed. As it happens, I think 100% of our plugs have at least one summing stage; Adam would have to speak to that more directly.
That said, our stuff is designed and tested at a nominal (to me, anyhow) level of -6dBfs. That's where I write the presets, and where we do most of the design work.
-CR
thanks for the answer Chris! That's just what I was wondering, what your optimal preset and intended design gain staging was... appreciate it.
I wanted to post this in the Man of La Mancha thread, but it is closed:
http://www.synthtopia.com/content/2013/08/16/the-magnetist-experimental-tape-music-documentary/
- djp
Comments are closed on this post.