|
|
Post by Randge on Oct 11, 2014 12:06:33 GMT -6
I track either 24 bit 96K or 192K. I am always thinking about the future and if anyone needs to go back and get the files, they will at least be recorded very high res for the time period. Imagine someone going back and singing with their deceased father like Hank Jr did a few years ago. That kind of thing happens and I like to be prepared for it. Not to mention remastering records 20-50 years from now into a new format when it comes.
R
|
|
|
|
Post by formatcyes on Oct 11, 2014 14:46:34 GMT -6
Your TV doesent produce ultraviolet or infra-red. They don't have this race to have a wider spectrum than we cannot see Xrays anyone. They do want more pixels but that is so the screens can be made bigger without loosing resolution. Because its digital they want more color chooses shades of white red etc..
If you can hear a difference going higher sample rates it's a problem with the convertors and/or the plugin's. The math is sound we cannot here above 20khz. 44.1k will give us all the information bellow 20khz if you want to reproduce dog whistles you require higher sample rates. If you want to attract bees to your TV you will have to get one that has an increased spectrum..
|
|
|
|
Post by popmann on Oct 11, 2014 18:09:20 GMT -6
I think everyone else should track at 44.1. Tracking above that is obviously some conspiracy created to make engineers' jobs easier....who cares how hard engineers have to work to make stuff sound right?
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 11, 2014 20:50:27 GMT -6
 Yeah popmann, we all stay at 44.1khz while you start to record in 384kHz. (~"11"). (Just kidding...  ) BR, Martin
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 11, 2014 22:40:04 GMT -6
Most of my comments on this forum are related to reverb (for fairly obvious reasons). But I do record classical music when time allows. And I've got to say that the higher sample rates make a positive difference--even if you're eventually knocking the end-result down to 44.1 or 48K. This is something I resisted for a long time--perhaps because I considered it only on the basis of perceivable frequencies. But intellectual honesty requires doing the actual experiment. And that experiment means that I'll be recording at higher sample rates.
It's not really hard to determine why this makes a difference. In this signal-capturing area, the A/D converters can be designed with more forgiving anti-aliasing filters. We're long past the days of brickwalls and resistor-ladder converters, but there still must be filters of this nature. If the slope of these filters can be more gentle, there are fewer effects in the passband. In the signal-processing side, filter design and performance are strongly affected both by word size and by sample rate (this is the reason I used double-precision math in the majority of my filters). Strangely enough, high sample rates actually have a greater effect on filters at low frequencies (at least filters of a particular type). It's also arguable that high sample rates give greater precision in timing (this could affect spatial perception). I'm less-convinced of that, since down-conversion would remove that advantage.
But the fact remains-even with a pair of ears that are decades older than many of yours--that 88.2/96K sounds way better than 44.1. Some of that improvement is lost when I downconvert, but the result still sounds better than recordings that originate at 44.1. I attribute most of that to filter behavior, but there are factors I don't yet understand. It's clear to me that there's a lot more to the puzzle than simply saying we don't hear above 20K.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 0:16:20 GMT -6
Interesting!
So, at which sample rate will be no further quality gain regarding the mentioned positive effects on filters? How far do higher sample rates matter and make a perceivable difference? Is 192kHz really a good ratio of effort/quality gain, do even higher sample rates make sense?
(The 384kHz were not as much of a joke as it might have sounded...i know there are already discussions about this sample rate for audio...)
At which sample rate would we gain more quality, if we increase bit depth instead?
I mean, these are the questions for a forseeable future of digital audio...
|
|
|
|
Post by tonycamphd on Oct 12, 2014 0:51:14 GMT -6
Your TV doesent produce ultraviolet or infra-red. They don't have this race to have a wider spectrum than we cannot see Xrays anyone. They do want more pixels but that is so the screens can be made bigger without loosing resolution. Because its digital they want more color chooses shades of white red etc.. If you can hear a difference going higher sample rates it's a problem with the convertors and/or the plugin's. The math is sound we cannot here above 20khz. 44.1k will give us all the information bellow 20khz if you want to reproduce dog whistles you require higher sample rates. If you want to attract bees to your TV you will have to get one that has an increased spectrum.. This strikes me a little odd, sample rate has zero to do with frequency's pertaining to sound, it's about sample captures per second, conversion and clock quality/ accuracy not withstanding, higher resolutions(more captures per second) should sound smoother/better, it's also generally accepted that plugs function/sound better at 96k. So I'm not really sure what ur saying here? For what it's worth, @96k, I personally hear a significant improvement in Q over 44.1 |
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 4:20:41 GMT -6
As far as i understand formatcyes, he refers to the Nyquist-Shannon theorem, which in easy words considers, that a sample rate of twice the maximum frequency of the signal you want to sample is sufficient to recreate the signal perfectly in an ideal world....and if you sample in a lower sample rate than twice the maximum signal frequency, you will have aliasing effects that can not be filtered out anyhow. Please have attention to the words "ideal world". This is referring to the (e.g. the required lowpass) filter quality, that Michael from Exponential Audio explained, which is not ideal most of the time. (Still, this does by no means render the Nyquist-Shannon theorem useless...) Therefore, e.g. Lavry considered the ball park of 60kHz could be an ideal frequency for audio AD conversion, considering this leaves enough room for the requirements of real world filters to have less side effects in the audible range. Hope this makes any sense for you in explaining the correlations between required signal bandwidth and sample rate formatcyes referred to... Best regards, Martin |
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 4:33:22 GMT -6
Aaand - yes, there are several factors that lead to the fact, that you can have better signal conversion quality AND signal processing quality with nowadays real world converters and software in 96kHz.
Some reasons for this were already mentioned....
Well, we live in a real world, so the observations made with 96kHz sounding better do not contradict Nyquist-Shannon.
Sometimes i think, this might be a point that could relax this regularly re-occuring sample rate discussion...?
Best regards and no offense to anyone...
Martin
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 4:43:47 GMT -6
As for plugs...
they do not automatically sound better at 96kHz, this is also depending on the types of filters used in the plugs. To overcome filter shortcomings, plug-in developers used oversampling in the past. If they didn't, it might be, that the same plug sounds better in higher sample rate...
(I just hope i do not make a fool out of me trying to bring owls to Athens in this topic...if so, i beg your pardon in advance.)
Best regards,
Martin
|
|
|
|
Post by formatcyes on Oct 12, 2014 4:59:35 GMT -6
Thanks smallbutfine. The Audio Engineering Society recommends recording at 48khz this give a bit more room for the low pass filter and in theory allows for perfect reproduction below 20khz. Sample rates above this is just a waste. I just wish all plugins where optimised to run at this sample rate. Sampling rates above this do not improve resolution below 20khz it only extends the range why I referred to a dog whistle which is around 26Khz.
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 5:22:13 GMT -6
Formatcyes,
i do not totally agree with the point that samplerates over 48kHz are *complete* waste. I DO use 96kHz myself in the mean time, because i cannot ignore the fact, that my gear also sounds better at 96kHz, and this is what i use.
Sure - we have to pay for this safety margin or whatever this can be called.
We pay for higher sample rate and word length in needed processing power, needed memory, needed transfer bandwidth etc..
I can totally live with double speed sample rates.
In my opinion it is totally legit to use 96kHz to get the most out of the available equipment. IMHO it is even legit to use a 48kHz AD conversion, as long as it sounds good and even upsample it for further processing ITB.
Nothing wrong with all of this.
The analog side of things may have a larger impact anyway.
But nowadays we face quad speed sample rates and even more for audio, where the benefit is IMHO more than questionable. To my understanding, this is way overdosed - but if everyone is going this route technically, what choice do we have?
Best regards,
Martin
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 5:40:01 GMT -6
Also, the Nyquist-Shannon theorem does not say anything about timing but about perfect spectral reproduction.
This is, what Michael referred to when describing, how higher sample rates might result in better stereo image...
In analogy to TV and images, you can have an absolutely perfect color reproduction, but a *moving* picture would still look better at a higher samplerate...
Just one of the factors that could speak for using double speed sample rate.
(OK, sorry if i overstress the topic. Enough for today...thanks for your patience everyone...)
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 12, 2014 9:26:47 GMT -6
Interesting! So, at which sample rate will be no further quality gain regarding the mentioned positive effects on filters? How far do higher sample rates matter and make a perceivable difference? Is 192kHz really a good ratio of effort/quality gain, do even higher sample rates make sense? This is a lively thread, and I'm afraid I won't be able to tag along fully (I'm at AES, so it's away from the web for a while). I would love to try 192K. My in-studio interface supports it, but the interfaces I use for location top out at 96K. I can say this much: one of my users--a very fine score mixer with his share of statues--routinely mixes at 192K. That was a great early stress test for my plugs. Some of his reasons may be archival, but I think it has mainly to do with overall sonic quality. I do have other guys which record, edit and mix at 8x (DSD). The edge those extreme rates give them might be very small, but I can't argue with the quality of the end result. |
|
|
|
Post by popmann on Oct 12, 2014 13:32:34 GMT -6
Here's the thing about "whatever above 96" you want to discuss....I call you deaf if you can't hear the difference in 44 and 96. 48 and 96, I wouldn't be SO blunt--you'd need coaching (or at least I did) on "quit trying to hear the sample rate and just pick the one you think sounds better"--in which case, the ABX software says I rank above the 95% accuracy. BUT....make it 96 vs 192 and I'll ride the 50% give or take ten all day--which for those not a statistics professor, means I can't yell a lick of difference. NOW...make it DSD, and I'll rarely veer from 99-100-%....but due to the nature--there are gain shifts involved in the transients that are a dead give away...and make the source 96khz PCM (with mastering for DSD) and I will go back to the 50% line.
So, I am an will be a believer that SACD was king--because whether you tracked or mixed analog, 48-96khz PCM, the single little silver disc would deliver 100% of the master quality to the consumer--who I might add could play that fidelity back with a CHEAP player and still sound amazing. But, once you're talking about selling people files, you literally DO need to factor in size to performance. And both 192 and DSD (sonically unrelated-only relate by file size) fall down.
Anyone using Michael's plug in is likely working with DXD....which is short for 384 linear PCM. DSD natively offers very little editing (beyond what a tape machine could do)...no mixing....and certainly I've never seen it incorporate PCM plug ins. Because it literally has to BE converted to be processed at which point, it's just not DSD anymore...on a geek note, should be of interest that DSD has the frequency response I believe right between 88.2 and 96. But with a time domain equivalent to 384khz--so, since I can't tell the 96 from 192....but can from DSD, extending the frequency response doesn't appear beneficial. But SOMETHING does....so....potential time domain holds the answer.
That said, that's tech geek BS at this point. If you have the client budget to buy $50-70k tracking decks and the all analog support to do it, feel free to let us know what you think. But, I often think that dangling what are for most situations functionally distant carrots have less than positive results--if one guy stays at 44.1 because they're "waiting for DXD DAWS..."--which means waiting for the daw, the conversion, the plug in support, etc...you're gonna raise kids before this becomes what a 24/96 PCM daw is in terms of maturity and cost.
|
|
|
|
Post by wiz on Oct 12, 2014 14:08:57 GMT -6
Thanks smallbutfine. The Audio Engineering Society recommends recording at 48khz this give a bit more room for the low pass filter and in theory allows for perfect reproduction below 20khz. Sample rates above this is just a waste. I just wish all plugins where optimised to run at this sample rate. Sampling rates above this do not improve resolution below 20khz it only extends the range why I referred to a dog whistle which is around 26Khz. theory is very cool. I love theory. Theory helps a lot. There is also practice. Some interfaces, I have personally owned and used, sounded better to me at 44.1 Khz than 96Khz. Others have sounded neither better nor worse. Some have sounded better at 96 than 44.1 In the case of the one that i liked better, at 96Khz , I wouldn't say its a waste recording at that setting... I liked it better, I felt better about it, I have the CPU power and disk space, so why wouldn't I use it at that sample rate? Should I not use it, because someone, at some time said above 48Khz is a waste? I have never understood the "black and white" attitudes about this stuff. Unless you can hear the box at the different sample rates how can one say its a waste? Latency alone might be a good enough reason. Because Math, says something...(and I don't for one second disagree with the math) doesn't mean it translates through the manufacturing process to the end user (the design maybe compromised at one sample rate compared to the other and perform better in the physical realm at that sample rate, whatever that might be) YMMV cheers Wiz |
|
|
|
Post by unit7 on Oct 12, 2014 14:36:22 GMT -6
After doing 96kHz for the first time in 2005 that's what I've kept doing when I'm in charge. It just sounds a bit better overall. I'm 50 and my hearing stops at 12k. When mixing (and mastering) analog, to me it just makes sense to stay at higher resolution as long as possible, because all processing sound better.
|
|
|
|
Post by formatcyes on Oct 12, 2014 15:03:08 GMT -6
There is no scientific evidence that sampling above 48khz has any benefit. This is why AES (yep they have a show on now) recommend 48khz. The math and the science is clear.
However here's my get out of jail card some chips have a sweet spot 44.1, 48 ,96 Blind testing required because preconceived ideas influence us a lot in audio (all of us have adjusted an eq thinking we improved the sound only to find it was bypassed).
My problem is the push for higher sample rates may infact be detrimental to good conversion instead of just focussing on 30hz to 20khz. With a perfect system 44.1khz will reproduce this band exactly. We have a push on for higher sampling rate's with no scientific examples of benefits. But because the converter chips are not optimised for 30hz to 20khz we have to go looking for a sweet spot. Having a band width of 30hz to 48khz is fine if we are bats.
Rejecting interfaces just because they don't record above 48khz is very foolish IMHO.
|
|
|
|
Post by popmann on Oct 12, 2014 18:40:57 GMT -6
Sure there's scientific evidence. Not only evidence....PROOF.
|
|
|
|
Post by tonycamphd on Oct 12, 2014 20:36:47 GMT -6
I have demo'd a few hi end interfaces over the past year, and IMO none of them have stood up very well against my current limited I/O rig(so what i'm saying is my current conversion is i/o limited, but kickass), that said, 44.1 sounds very good on my rig, but the difference in quality between 96k and 44.1 is not small. So I say "science" be damned, honestly, the diffs that i've heard here, you would have to have a problem with your monitoring to NOT hear?
I might also add that the HD tracks i've listened to at 96k/24 bit make the 44.1/16 bit tracks of the same sound weak by comparison, just sayin...
|
|
|
|
Post by Martin John Butler on Oct 12, 2014 21:22:29 GMT -6
I always use 48 instead of 44.1 in Logic, it just seemed a little more transparent to me. My computer's long in the tooth, but next time I have a session that doesn't have too many tracks, I'll give 96k a try. If I can hear something from 44.1 to 48, I'd bet I can hear something bumping up to 96. Sometimes I think "feel" might be a better word than "hear". These high frequencies that some people love to claim we can't hear anyway are probably something our nervous system perceives quite well.
|
|
|
|
Post by popmann on Oct 13, 2014 7:54:47 GMT -6
I, for the record, do not say "science be damned".....I typed a bit longer reply and deleted it so as not to offend the planet--but, assumptions made from stringing together various scientifically sound principles into one I can literally scientifically disprove in minutes....is not solid understanding of scientific process.
Just because the sampling science doesn't have an explanation, doesn't mean that it WON'T....nor does a lack of scientific explanation of that which is scientifically observable disprove what was just observed.
Tony, if you want to do that comparison fairly--you know you need to take the HDtracks files and REDUCE them yourself, right? You can't compare it to the CD (necessarily)--because they're very often different masters which DOES make more difference than the resolution. You want to really test you ears--reduce it to 16/48 (not 44.1) and download an ABX'r....
|
|
Deleted
Deleted Member
Posts: 0
|
Post by Deleted on Oct 13, 2014 9:11:15 GMT -6
OK,
so we get a good portion nearer to a common sense that does not neglect scientific methods.
1. Common sense is that a *perfect* theoretic converter could convert the human hearable audio band at 44.1kHz spectrally perfect. (Nyquist-Shannon)
This said, there are very few people able to hear 22.05kHz.
Side Note:
I was able to hear 22kHz at the age of 19, and the doctor made the test three times because he couldn't believe the result. (Today i'm around 16kHz...at the age of 46 last year and this is already very uncommon...)
2. 48kHz was/is for a long time the recommendation to give enough bandwith to shift negative real world filter problems out of the audio band.
This should give enough bandwidth for spectrally neutral/perfect conversion in the real world.
3. Double speed conversion can be tested to sound better.
To do this in a scientific way, you need a 96kHz conversion like one of those from HDtracks or a well produced DVD-A or SACD that you reduce with an unhearable SRC down to a 48kHz and compare ABX.
Now - at this point we may have the problem, that, if you use the same DAC to listen to the both files, you may already hear the optimization of the DAC to the higher sample rate. But anyway - Take a consumer DAC. This is your customers conversion. Windows spits out 48kHz by default into a cheap chip soundcard which nowadays always is able to do double speed files as well.
I am still able to pick the 96kHz conversion. And the HQ SoX SRC i use to downconvert measures unhearable for me.
Thanks, popmann, for this test method. Yes, this is a scientific method to measure by ear. Do it for a statistically relevant number of samples and you are in for a scientifical proof within the determinded environment you use.
So, the verdict is there are perceivable differences in the time domain between 48kHz and 88.2/96kHz going on, that might have to do with stereo field or transient behaviour - not spectral behaviour. (?)
Just trying to break down what was said to this point of the discussion.
Please comment or prove me wrong if there is some misunderstanding from my side...
BTW, thanks everyone for the reasonable and factual discussion. I know this is an often discussed topic, that can easily escalate into something personal...
Best regards,
Martin
|
|
|
|
Post by popmann on Oct 13, 2014 10:55:52 GMT -6
Yeah, what it comes down to--is there's a scientifically perceivable difference--both in spectral analysis and in blind ABX listening tests. That is SCIENCE. ...now, the "why"....the scientific explanation is not there (yet)....there are some implementation based hypothesis--see Larvy's paper on why in the real world conversion 88.2 is most accurate (of current standards)--which isn't far off from the original CD proposition being what 18bit by 50khz? Wasn't the theory 40 years ago what we'd need in the low 50's and about 18-20bit (of actual resolution, which is different than tracking resolution)--now, Larvy, who has spent his adult life making converters says 60khz is what it takes? Ok. Low 50s....60....six of one....half dozen of another....Nyquist theorem speaks of ideal world theory--perfect linear phase decimation filters and such--this too is "science"--but, it's ideal theory. You can "prove" it (or disprove it)--but, it's not related to the perception of difference. You could prove that some perfect lienar phase on chip ADA will perfectly reproduce up to 22.5 khz (at 44.1)--AND you can prove I can't hear a 25khz test tone (this is an assumption on my part, having never tested it)--but, that does NOT prove I can't hear the difference in 44.1 and 96khz. You don't get to do this kind of relation. That can give you basis for a THEORY that I can't tell the difference--but, when you put the ABX up and I get 99%, your theory is disproven--while NEITHER of the factors that LED to your theory is disproven. I still can't hear 25khz....and that perfect converter STILL reproduces frequencies up to 22.5khz. But, the ABX test says I can hear the difference, not only a statistically relevant number of times, but almost without fail. You can't ignore THAT science because it means you don't fully understand why. Show me AES doing any research into this. They have not--at best you can point to some show floor side show thing they did with a handful of people a decade back. Man, that really seems thorough....not. I don't like when this gets cast as some kind of anti science thing. It's not. Just because science doesn't currently have a wonderful and full understanding enough to explain an observation doesn't make it voodoo. Also....48khz is fine. For anyone reading this--on an old machine...or using ADAT connections--48khz is a perfectly fine rate for multi tracking. It's not ideal--but, as an engineer, I'm rarely finding myself able to deal in an ideal scenario.  Leave 44.1 alone. Reasoning for 44.1 TRACKING is technically a flawed one, IME/O. A better sounding recording sounds better in any format-and 48khz will produce a better sounding recording. And, in a "file as delivery"--there is not an iThing that was ever made or computer sound card (that I"m aware of) that won't play back 48khz. Streaming services are simply "who cares"....because they don't have a model to pay for recordings, why do I give a shit about the nuances of sound of their stream? I think Larvy's recommendation is a solid one for those not wanting to experiment. 88.2 if the system can....48 if it can't. How much simpler can it be put? And since I can mix 88.2 audio on my little low rent MacBookAir at this point....for AUDIO production, I simply don't see the argument for 48--other than a system relying on ADAT connections or someone using old legacy hardware recorders--Roland, Akai, IZ Radar, etc....any interface made since like the year 2000 can double the rate....and the older they are the bigger the differential in sound ,IME. |
|
|
|
Post by svart on Oct 13, 2014 11:16:45 GMT -6
So let me throw this out there.
I record 44.1. I do this because my old system couldn't really handle 88.2K in the 24-48 tracks that I usually end up with during a session.
I also do hear a difference, although I don't think 88.2K is 2x better than 44.1K.
As mentioned before, I won't use anything that isn't evenly divisible by 44.1K mainly so that the original tracks can later be used with ITB mixing with minimal mathematical operations. Those who need to do bit depth or sample rate conversions with non-integers end up truncating too much actual signal data and going higher adds nothing.
Anyway, I'll be moving to 88.2K on the new system, but I doubt I'll ever go higher than that as it's useless to do so while having to dither back to 16/44.1k
But as an engineer who works with high speed (2Ghz) signal acquisition on a daily basis, it's fairly rare that anyone complains that there is a loss of signal fidelity without oversampling. We generally shoot for 2.5x sampling rate to allow for some changes in the Nyquist filters.
That being said, filters are signal modifiers. They modify the signal in the band-stop region but then they also modify the signal in the band-pass regions. They modify phase and therefor phase-delay and group-delay. They can have ripple and transient recovery differences over frequency into the passband and then a whole mess of other non-linearity issues that marketing specs like high sample rates and bit depths don't take into account.
Honestly if you ask any analog signal integrity engineer, like me, they'll tell you that the integrity of the filter, the components and the power system are absolutely without a doubt more important to the sonic quality of a converter system than the sample rate or the bit depth.
I would take a well designed converter running at 16/44.1K over a M-box doohickey running at 24/192K any day of the week.
Also, the DAC output converter filters are especially important because no matter how many points you have digitized, if the filter has poor response to the various signal attributes (transients, DC, large signal response, etc) then it won't matter if it's high or low resolution output from the DAC, the filter will react the same and you'll have no difference.
|
|
)
Leave 44.1 alone. Reasoning for 44.1 TRACKING is technically a flawed one, IME/O. A better sounding recording sounds better in any format-and 48khz will produce a better sounding recording. And, in a "file as delivery"--there is not an iThing that was ever made or computer sound card (that I"m aware of) that won't play back 48khz. Streaming services are simply "who cares"....because they don't have a model to pay for recordings, why do I give a shit about the nuances of sound of their stream? I think Larvy's recommendation is a solid one for those not wanting to experiment. 88.2 if the system can....48 if it can't. How much simpler can it be put? And since I can mix 88.2 audio on my little low rent MacBookAir at this point....for AUDIO production, I simply don't see the argument for 48--other than a system relying on ADAT connections or someone using old legacy hardware recorders--Roland, Akai, IZ Radar, etc....any interface made since like the year 2000 can double the rate....and the older they are the bigger the differential in sound ,IME. 