Yes John I agree with you too. It has been an eye opener for me as well. I have always assumed that virtual instruments running at any sample rate would give a very similar result and so Craig thought and started out too and found out different. I am quite surprised myself in what it has shown up.
 

Anderton
One panelist said that to have a reference point for good sound, you should listen to vinyl. I disagreed - I said the reference point for good sound is live music in a room with good acoustics

Yes I tend to agree with you Craig in this eg an orchestra or a live totally un amplified Jazz ensemble for example. But I do have some fond memories of how good vinyl can sound and in some cases it can be an amazing reference point. For example in my Hi Fi heyday in the 70's the finest system I have every heard in my life was a premium turntable and arm fitted with an Ortofon SL15Q pickup (freq response flat to 50 kHz!) feeding a super high end preamp and driving Williamson class A valve amps feeding Quad Electrostatic speakers. (flat to over 23kHz) Also Sheffield Lab records where the studio is connected direct to the cutting lathe with nothing in between! Tape does not compare to this.
 
This has to be heard to be believed and it is what I still have in my mind as a reference point. Pretty hard to beat. There is vinyl and then there is vinyl!
 
PS I owned the above setup for a long time so I know what it sounds like. I also added sub woofer to the Quads. EVERY genre of music sounds unbelievable on this setup. Electronic music sounds ridiculous on this as well.
 
 
 

John
Putting aside the issues here I just want to say this has been one of the finest epic threads I can remember. All the participants deserve a pat on the back. 

What I'm hoping is that it will inspire others to conduct their own tests and report back on the results. I do keep hearing from people I trust that they prefer recording with 96kHz. They don't really want to because their system gets more squirrely and the file size gets bigger, but they do it regardless and not because of client requirements or fashion. 
 
I'd particularly like to hear from more people who don't use distortion or virtual instruments as to whether they hear a difference.

.

I record at 96/24 not because individual tracks sound any better, but when mixing everything together, the end result is more open than when I mix at 44.1/24. The fact that some VST's render better at 96k may also have something to do with it, whether it is due to poor programming, bad drivers or whatever, I'll take it. Anything that makes getting a mix where I want it to be is a plus.
 
Regards,
Dan

mike_mccue
How do we find out if a dsp programmer has incorporated CPU saving short cuts and or round offs to their products' calculations? How do we find out what oversampling options are specifically doing with those calculations. For example; Is "2x" oversampling in a dsp residing in a 44.1kHz project actually using all the increments available in 88.2kHz or does that just bump it up to twice the detail of the actual calculations happening in the dsp?
 
For example; How do we know that a synthesizer isn't just grabbing every other sample from a stream and mashing stuff together into the "synthesis"? I guess we'd have to ask.

I would judge whether they took short cuts by ear. If something sounds good in the way that I use it, curiosity aside I don't really care why. And if it doesn't sound good, or something else sounds better, I wouldn't use it. And the "something else sounds better" is a big reason why it makes sense for plugin authors not to take shortcuts. 
 
"Detail" isn't the right word. Higher sampling rates allow for higher frequencies and this drives the frequencies that will cause aliasing higher as well.
 
In an ideal world, CPU wouldn't be a limiting factor (and today it is much less than it was a few years back), and the plugins' authors would choose an appropriate sample rate to eliminate audible problems for the type of processing that they're doing.
 
Where it gets tricky is some things will only really require a higher rate with certain settings - i.e. a compressor may be absolutely fine unless you push it to really fast attack/release times and set the threshold so it's constantly crossing back and forth. So do you always oversample it and waste CPU for the 90% of uses where it will be fine, or never oversample and potentially get some unpleasant distortion with specific settings, or do you give the user a choice - knowing that most of them will have no idea when it needs to be turned on and when it will have no benefit?

Geoff there is something else going on there. Looks like some sort of HF noise.

How do we find out if a dsp programmer has incorporated CPU saving short cuts and or round offs to their products' calculations? How do we find out what oversampling options are specifically doing with those calculations. For example; Is "2x" oversampling in a dsp residing in a 44.1kHz project actually using all the increments available in 88.2kHz or does that just bump it up to twice the detail of the actual calculations happening in the dsp?
 
For example; How do we know that a synthesizer isn't just grabbing every other sample from a stream and mashing stuff together into the "synthesis"? I guess we'd have to ask.
 
Is it possible that some programmers use the same strategies that audio compression codecs use? Is it possible that some dsp selectively omits the use of information that is thought to be pyscho acoustically masked by other info so as to optimize calculation efficiency? If something did that then scaling the sample rate might help with adding more detail.

 
Mike, I think this is something that Bitflipper would be eminently qualified to discuss. I hope I didn't scare him off because he caught me when I was dealing with some really difficult issues unrelated to Cakewalk or my gig...
 

Where it gets tricky is some things will only really require a higher rate with certain settings - i.e. a compressor may be absolutely fine unless you push it to really fast attack/release times and set the threshold so it's constantly crossing back and forth. So do you always oversample it and waste CPU for the 90% of uses where it will be fine, or never oversample and potentially get some unpleasant distortion with specific settings, or do you give the user a choice - knowing that most of them will have no idea when it needs to be turned on and when it will have no benefit?

 
Drew, I think this is another discussion that while a moving target (e.g., not relevant for 12-core monster, relevant for older laptop) is very important. There are definite economic tradeoffs involved which will ultimately influence how this all resolves.

Ok I did a quick test on a new project using TruePianos, NI B4II and LoungeLizard3. I have other Virtual instruments as well but I use these most often.
 
I recorded a simple midi track on all three and exported the audio output of the tracks at 96k with my converters turned off so they would not cause any problems with he export.
 
then I switched sonar back to 48khz (I did not use 44.1 since that is not the normal setting I use) Imported the resulting audio files into the project making sure they routed to the same bus that the virtual instruments went to.
 
Turned on exclusive solo and began seeing if there was any difference.
 
I can report that for both NI B4II and LoungeLizard3 there was no difference that I could hear. However Truepianos had a very noticeable difference. The 96khz file I imported had a smoothness and a pleasantly much more mellow tone to it. More rounded like a real acoustic piano would sound in a nice listening room.
 
I did not have time to test Jamstix and I don't know what sample rate its samples were recorded at. I also have Session Horns as well that I could test.
 
I may need to add this to my work flow for instruments that benefit from the higher sample rate.
 

mike_mccue
That's a lot of high frequency goodness. :-)
 

Thanks for looking, guys! Notice that the gain is cranked all the way up so that a soft Shhing sound clips the track. Even there, the noise when the mic is disconnected is at -60. So if you turned the gain down reasonably so you could sing into the mic, you would suppress this noise by a lot.
 
I thought the 60 cycle hum was interesting to note. Very quiet, but present. I'm not sure if this is proof of an electrical problem with the tangle of cables and power-strips I have for my gear.
 
When I compare the images with the mic unplugged to the images with the mic plugged in and the DBs, you can see the Mic is picking up lots of high frequency sound.
 
It's important to note the range of data in the images in the two posts are not the same. DigiCheck automatically shows you relevant data (if you set it to). In the image of the 'shh' recording, the range is -50 to 0 where in the images of the mic unplugged recording, the range is -100 to -50.
 
In the images with the mic disconnected, you can see that the average level is around -65 dB and in the one with the mic connected where I'm making a 'shhh' sound, the average level is louder than -20 dB and that the mic is clipping even at this setting. In all images the gain is at 65 (the loudest it goes to). 
 
So if I brought the gain down to 30 or less to make a recording of a singer, It would suppress the noise to -90 dB or something compared to the vocal which might average around -20.

.

I may well give recording at 96 another crack, since if it becomes dodgy, I can always revert to 48 k, which is what I have been running.
 
I have a pretty powerful machine, and lots of available storage, so it is really a matter of running with it for a while at 96 k and deciding if the extra processing time and increased storage needed is worth doing.
 
I would be doing this with the thinking that even if I cannot hear it generally, if the computer is beefy enough to handle it, maybe certain combinations of things will see slight improvements at the higher rate.
 
Most of my work is midi, with either myself or a single other guitar player, and sometimes a bass player, so no much risk to anything even if there are problems.
 
Bob Bone
 

Thanks, Mike. As you know, I record 2-3 nights a week, so since I can only see the problem in the picture (can't hear it), I'm just not going to worry about it. It was interesting that as soon as I touched the loose end of the mic cable the 60 cycle hum increases noticeably. This is a good reason to discourage singers from holding the mic - it looks like.
 
This is the cable...
http://www.sweetwater.com/store/detail/DualCable20?utm_source=Google&utm_medium=PPC&utm_campaign=audio&utm_term=adwords_labelsCables&device=c&network=g&matchtype=&gclid=CjgKEAjwwuqcBRCSuoivmIPnkwQSJACpqj3k-ktQfKxdgV_t5tte7Ov4ZclbPU5Dt1wY6HutUrCSbPD_BwE

Best, G

.

Hi! I'm totally new to this forum, but I've been for almost 32 years in the recording industry. I'd love to share with you some of the ideas other engineers, physics scientists and audiophiles have been discussing for over ten years about hi-def audio. 
 
Let's go to the very basic thing: the only way to have a great recording is recording it well. So... mikes, venue, artists, placing of people, instruments and equipment are important. The recording method and the type of system is paramount at this stage. It's not the same if you're using a 96ch. SSL Board with all the bells and whistles and dumping the recording to a 1" tape than using the same analogue board signals into a ProTools HD rig at 192/24. Here comes the first step: if I have a tape for, let's say, summed tracks for drums on a two-track, I can play with that all I want, til I go to the whole mix. In fact, what we do is mixing in groups to be able to use all the capabilities and maximum fidelity on tape. Teh dynamic range of tape is hard-limited by the medium, so, translated to digital, it can go from 19-20Hz to 22kHz depending on a dozen of factors (azimuth alignment, demagnetization, head wear, and so on).
 
If I capture the performance in ProTools HD 192/24 I still have all the analogue characteristics of the hardware I used (the board itself and all the other processors we love... preamps, outboard EQ's, etc.). But... I have ALL the tracks, up to 96, let's say, on their own and at a very high resolution. The huge difference here is this: I can play with these tracks almost fearlessly. I can compress, EQ, gate, limit, push, colour... whatever I want -obviously, within the limits of musical perfection, without destroying the timbre, tonality, and, most importantly, dynamic range. Most microphones, even the most sophisticated hand-made German ones have serious physical limitations, as our own hearing. I love my U87AI and it "just" goes from 20Hz to 20kHz!!!! So... I'm recording at 192kHz/24bit resolution, but ALL I have to work with is 20 to 20K. If I were recording direct to 44.1/16 I'd be seriously limited in "tweaking" the tracks, since ANY change would alter or terminally destroy its contents. This is due to the Nyquist frequency which has been explained before. Now, I have three times (in 192) or twice (in 96) the leeway to work that on a 44.1 or 48 track.
 
So this is provenance 1: the recording session. I do it in the digital domain in high resolution. A must if I want my product to be Hi Def.
 
Then, provenance 2: My finished mix, let's say in 96/24, will go to the mastering phase. Here, more damage (tweaking... Oh, God! what we do is making things sound better through destroying them?... hell, no!) to the original signal will be done. The practice today is to compress things to absurdity (0dB anyone?), so there's no real silence where silence should be, no mellow or really subtle passages... everything is slammed to the red -without clipping! If the mastering house knows they're going HD, they should keep things where they are... smooth sine forms would show, instead of bricks. And... that's it for provenance. You can still have your 96/24 stereo mix in that sacred realm.
 
Now, we have to go public: distribution is where things get lazy. Let's bump the whole master to -2 and downgrade it to 44.1/16 for distribution. Most of the love that went into the process is turned into a red rage of "normalisation" or whatever name they use for the process... the brights are ultrabright and the shadows are ultrashadowy! 
 
If we stop at stage two of provenance, we could listen to music for much longer and without fatigue. This may not be scientific at all, but it's a fact. There's less fatigue when listening to HD music than to CD-standard music. 
 
We don't understand the hearing process as we do with sight. We don't have golden ears (maybe there are some folks who do, but they are pretty scance. With age, we lose our ability to hear some frequencies. We're "blind" to them.
 
So... what makes HD music better than standard Redbook CD-quality? The depth of the recorded material. Sometimes you can "feel" the silence. The lights are not hot and the shadows are not that dark, so it's like having a picture with detail in both shadows (lower frequencies down to 30Hz, which you feel) and light (like the details of a white cloud hit by the sun, where 20kHz is the limit for our ears to listen to). It's smoother.
 
Is all this scientifically based? Yes, absolutely. Can we hear the difference from a very good 44.1/16 to a 96/24 recording? It can't be demonstrated yet. If you work in a studio environment yes, you would "feel" much better and relaxed working with HD tracks (oh... files!). 
 
Since the topic's been going on for ages... well... it can go on forever until the day science can measure objectively our hearing perception and make it universal.
 
 

I can't agree with that. Hi Def means nothing. Heck my on board sound chip calls itself Hi def. When you increase the sample rate you do not increase the resolution but only the bandwidth. 
 
As far as being able to process a higher sample rate audio file and not impact its dynamic range while using a compressor is absurd.