Remember that 96K TH2 thread? I Just had my mind blown, big-time

I said I'd check it out further, and I have. Now there is NO DOUBT in my mind that recording at 96kHz is a big deal for anything generated electronically inside the computer. For signals coming in from the outside, the input filters and converters pretty much keep things under control. But amp sims, virtual instruments, etc. can easily generate signals that go above the clock, and fold back into the audio range.
 
I created MIDI sequences for Addictive Drums and a the Z3TA+ "Melody Maker" patch. For the latter, I played block chords and transposed a copy up an octave so there would be plenty of high frequencies. I then ran the MIDI sequences at 44.1 and rendered, closed, opened Sonar at 96k, opened the same patches, ran the MIDI sequences, and rendered the resulting audio to 96k.
 
I then opened up the original 44.1kHz project and imported the 96kHz files. But wait, you say...don't you lose the 96kHz goodness because you're bringing them back into 44.1?
 
NO!!
 
That's what blows my mind. The difference with AD was subtle, but noticeable. The cymbals were less harsh and more melodic, and the attack on the drums seemed more natural. But the difference with Z3TA+ was startling. It wasn't subtle, it wasn't something where you had to switch back and forth and listen really carefully on headphones...it was the kind of difference where if you can't hear the difference, you need to pursue a career that doesn't involve audio. There were high frequencies that simply weren't there at 44.1, because they were reproduced instead of turned into aliasing.
 
As far as I can tell, this has nothing to do with 96kHz having an extended response as much as it is being able to reproduce sound cleanly within our usual audio range. When the harmonics fold back from hitting the clock, they fold back out of the audio range. As 44.1 can reproduce sound within the audio range, it was able to reproduce what was captured at 96kHz.
 
That settles it. I'm going to start recording at 96/24 starting now, and see what I can get away with in terms of track count and latency. If I can record at that sample rate, it's worth it. The fact that the advantages survive even when brought back to 44.1kHz is the icing on the...cake.

Oh. and the self-serving part: I will be doing a presentation on this as part of a panel at the New Music Seminar in New York on June 10. I originally joined the panel to be the voice of reason about why I didn't think it was worth going with 96kHz, so I did some research to collect real-world examples that would justify my position. Instead, I found out my position was wrong.
 
The key point here is I think there are two separate issues - recording at 96, and playback at 96. I'm still not convinced 96kHz is necessary as a consumer playback medium, but for recording electronic instruments and processors, I just became a convert. I better get another hard drive...

Interesting,  
I record at 88.2k as rendering the files is divisible by half and supposedly less prone to downsample rendering errors. Where 96k is a non integer sample rate.
 
It could be considered the conversion algorithms in DAWs are much better nowadays and it is not an issue.
 
So apart from not much less HDD space is there any benefit going to 96k?
Or rather a better question does 88.2 give a similar result?
 

It's the maths ;-)

Bjn, some folks here do video and hence 96 since the video distributed at 48 often. IMHO, no reason to use 88.2 over 96 other than disk space and processor consumption.

Also 88.2 isn't as well supported because it's less used. For example the Roland Octa-Capture and Quad-Capture don't support 88.2 but do support 96.
 

I have been using 96 for a while now. At least two yrs. When I started doing it, I just noticed just a better sound. I could/can not describe it at the time, nor can I put it into words today. But I hear something.
 
About 2 months ago, I started a project and decided to do everything in 48k. So some of the songs I had at 96k, I redid them in 48k. About 2 wks ago I had to open an older version of one of the songs to grab some info out of it. It was in 96. So I reset my VS100 to 96, opened the project and listened. I could hear the difference. And I thought it was just my mind playing tricks on me. But my wife was sitting there with me and without me saying/asking anything, she said,"Those guitars sound really good."
 
I made the decision to go back to 96 then.
 
Then Craig started posting about 96k.

I'm not sure what "fold back" means. 

mixmkr
I'm not going to try and recreate this, have never had any problems at 44.1, and this sounds like a 'clock' problem...to this limited, technical person.  Can frequencies really "bounce" off the upper limit of the clock setting and be captured...like sound waves bouncing off a wall?

It seems to me that they would be filtered out. As stated it would imply that the higher frequencies are somehow morphing into lower frequencies. I have heard variations on this as an explanation why it sounds better to the person. I just wonder if a double blind test would give the same results.    

...it would imply that the higher frequencies are somehow morphing into lower frequencies.

That's exactly what's happening.
 
Frequencies above half the sample rate cannot be accurately reconstructed and are misinterpreted as lower, legal frequencies. The result will be a frequency that's the difference between the real frequency and the Nyquist frequency. 
 
At 44.1 KHz, the highest legal frequency is just under 22.05 KHz. If you try to encode a 23 KHz signal, what you'll get is 23 minus 22.05, or a 950 Hz tone. This is what people are referring to when they say it "folds back".

I thought folding back was addressed with a filter.

bitflipper
Sorry, Craig, but  you're inadvertently propagating well-intentioned misinformation.
 

... amp sims, virtual instruments, etc. can easily generate signals that go above the clock, and fold back into the audio range.

This is a true statement, but anti-aliasing will be handled internally within a well-designed synth or distortion processor. If you have to increase your sample rate to make some plugins work better, then you need better plugins.

 
Saying I need better plug-ins is all well and good, but that's sort of like saying all my plug-ins should be 64-bit, which they should be. But some aren't, so I have to use bridging, which is a flawed technology but serves its purpose.
 

If AD sounds harsh at 44.1KHz it's a design flaw in AD. What is the sample rate for AD's samples? 44.1 KHz. Playing them back at the same rate they were recorded should yield the best fidelity. If it doesn't, and the reason is aliasing, then there is distortion happening within AD that either shouldn't be there or that should have been handled internally with upsampling and filtering.

 
The samples sound fine at 44.1 until you start adding synthetic processes like boosting the treble, adding saturation, etc. It doesn't matter to me why the results are an improvement, because I need to make the best-sounding music I can with the tools I have today. I agree upsampling and oversampling helps, but both rely on interpolation and in the case of oversampling, stuffing in zeroes and interpolating on playback because attempting to interpolate while recording creates its own issues. Running at 96kHz for sounds that are generated synthetically provides "real" data for each sample.
 

Consider the most common scenario for generating "illegal" frequencies within your project: harmonic distortion. You might call it an exciter, an amp sim, a tape sim, a revitalizer, a tube emulator, or a saturator - they're all adding harmonics that can potentially include frequencies above Nyquist.

 
As can harmonic-rich waveforms from synthesizers like Z3TA et al; it's not just distortion.
 

Such processors most often add odd-order harmonics. For example, a distorted 10 KHz signal's third harmonic of 30 KHz would exceed Nyquist at 44.1 KHz but not at 96 KHz. However, the fifth, seventh and ninth harmonics still exceed Nyquist, even at 96 KHz. IOW, raising your sample rate is only a partially-effective band-aid for mitigating problems in your plugins that shouldn't be there in the first place.

 
But wishing the problems weren't there in the first place doesn't make them go away; raising the sample rate does. Most of the foldover with 96kHz bounces back into a range that's above 20kHz so we don't hear it.
 
But I also don't think it's just about aliasing. One result that made so little sense I re-did the experiment to make sure was that the imaging of TH2 with its included reverb at 44.1kHz "wandered" compared to the same preset at 96kHz, where the image was rock-steady. The sound quality was the same. I'm theorizing that reverb is sufficiently complex that doubling the processing rate somehow tightened up the calculations and reduced variations between the left and right channels.
 
I checked with designers at Native Instruments and IK Multimedia when I first started running amp sims at 96kHz and I thought they sounded better, but didn't trust my ears. I asked if i was hearing things or whether there was an actual reason why they sounded better. I don't have their responses on this computer, but I can look it up. Independently, both of them mentioned improved computational precision as the main reason why, not distortion. Perhaps the experience with the TH2 reverb supports that.
 
The bottom line is I got incontrovertibly better sounds out of virtual instruments and plug-ins by running the project at 96kHz, even when sample rate converted to 44.1kHz and played back through a 44.1 audio engine.
 
To hear what I mean, here's a link (expires in six days) to download two files produced by the Z3TA+ 2. One was recorded with the project running at 44.1kHz, the other with the project running at 96kHz and converted back down to 44.1kHz. I'm not even going to say which is which, because it's audibly obvious which file reproduces high frequencies more cleanly and accurately. I recommend that anyone who wonders whether running a project at 96kHz can improve the sound listen to these two files.
 
P.S. I also read on the web that digital filters aren't perfect, which may be part of the story as well.
 
 

Craig,
The Z3TA-2 has a render resolution option. Could this be affecting the results?
 
Michael

microapp
Craig,
The Z3TA-2 has a render resolution option. Could this be affecting the results?
 
Michael

Synths and amp sims with oversampling do sound better with oversamplin engaged. Although theoretically running something at 96k should sound better than running at 48k with 2X oversampling, I don't hear a difference. But the bigger point is not everything has an oversampling option (conversely, Live's EQ8 has oversampling always turned on and you can't turn it off), so running at 96kHz is a one-size-fits all solution compared to hunting down the oversampling options and enabling them.
 
This is a new world for me so I'm still looking for answers. But I don't "have a dog in this fight." A lot of people come up with theories, then look for evidence that supports the theory. I was looking for evidence that 96kHz didn't make a difference, but found out I was wrong.

Craig,
Quantization noise and jitter would only apply when going from analog to digital I would think. I would expect the synth output to be sync'ed with sonar's clock which in a freeze or bounce is not in real time. I suspect something else is going on here. The sample Z3ta-2 files you posted sound to me like what happens when Z3ta is switched from low to high resolution. (Similar to a fast bounce vs a fully rendered off-line bounce).
Perhaps there is an interaction between Z3ta output rate and sonar sample rate. Check the output resolution of the Z3ta. I am not in my studio right now but I will attempt to repro your results later tonight.
Michael