[Explained] Why should you use higher Sample Rates while recording

[pipelineaudio]

I think I linked this in here before but will put it here again
Running your software synths at higher sample rates - Gearslutz.com

Pretty interesting read and explanations/examples of aliasing occuring in synthesis.

And compressors...big time

Thats why we oversample in so many dynamic range plugins. Look at scott stillwell's stuff...put it on a cymbal with a fast release and try it with and without the oversampling on

But this is a whole different game than recording at a high sample rate

 

[xeonblade]

Big time bump

 

[QueeZeR]

I found this link a while ago:
http://people.xiph.org/~xiphmont/demo/neil-young.html

It disagrees about the whole sample rate thing (but agrees about 24 bit, during mixing).
I don't know how accurate the info is, but found it an interesting read.

 

[ElRay]

TL; DR: You're right about oversampling in the final product, but for the wrong reason.

The sampling theorem as expressed by Shannon, based on Nyquist's work says precisely: "If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart."

That may be "adequate" for mere recording of the signal, but it's painfully obvious to anybody that has real experience in discrete signal processing that there's an infinite number of analog signals that will produce the exact same two discrete data points, yet sound completely different. The Nyquist frequency is a sufficient condition on the MINIMAL sampling rate to avoid aliasing, nothing more.

In addition, that's just regarding the basic digital samples, that says nothing about the psychoacoustics of hearing. Here's a simple example: Modulate the envelope of a high frequency sine wave with a low frequency signal and then high-pass the result with a cut-off that will guarantee that none of the modulating signal is present in the output. If you listen to that single sine wave, you will hear the modulating signal, even though it is not present. Now, if you were to sample that sine wave at merely the Nyquist rate, you won't hear the modulation. Sample it at 8x the Nyquist rate, and you'll perceive most of the modulating signal.

Finally, if there is any processing that will need to be performed before the final "mix down", the information loss by sampling merely at the Nyquist Frequency is tremendous. Remember, the sampling frequency affects not just the resolution of the input signal, but also the resolution of the filters. As another example, my PhD dissertation was on auditory processing of hearing in noise. If I ran my simulations at anything less than about 160 kHz, they were unstable and/or did not replicate observed phenomena.

You can read a textbook and mis-apply a theory all you want, but that doesn't change reality. The value of the Nyquist frequency is merely as (A) The lower bound on the sampling rate, given the frequency components of a signal, or (B) The upper bound on the frequency component of a signal, given the sampling rate, in order to prevent aliasing in a discrete signal. That's in, nothing more. It says nothing about the quality of the sample, nor the accuracy of the analog signal recreated from the discrete samples.

The reason you can get by with lower sampling rates for audio is not due to the Nyquist theory, it's due to the signal components of music. The 3rd harmonic of the highest note on an 88-key piano (C8) is only 12 kHz. Sampling at 48 kHz gives you four samples per cycle. Considering that the highest note on a 24-fret, standard tuned, guitar is E6, you're getting about 37 samples per cycle at standard CD sampling rates. Take that out to the 5th harmonic, and you're still getting 5 samples per cycle.

Ray

 

[All_¥our_Bass]

A really good analogy is framerate count in animation.
More frames = more fluid movement, less frames is jerky.
Works the same way for sound. If you keep downsampling even from the best recorded tone ever you get this really murky lo-fi kind of tone.

You can try this in your DAW record something or import some really clear hi-fi recording you know very well, select the whole thing and get your daw to downsample it from 44k to 22k, listen, then go to 11k and listen, 5.5k, etc. keep going and notice how the quality degrades, there's less treble and bass and even the midrange gets progressively more blurry and fuzzy.

 

[thraxil]

Finally, if there is any processing that will need to be performed before the final "mix down", the information loss by sampling merely at the Nyquist Frequency is tremendous. Remember, the sampling frequency affects not just the resolution of the input signal, but also the resolution of the filters.

Oh, you mean like in my first post in this thread where I say the same thing?:

Which is not to say that 44KHz is perfectly adequate for all (or even most) recording purposes. Each stage of mixing or layer of effects processing potentially loses information. The higher the sampling rate you start at and work with, the better the results will be overall. The catch, of course, is that higher sampling rates mean larger files, more memory used, and more work for the CPU. So the higher the sampling rate you want to work with, the beefier the computer you'll need.

In a nutshell: record at as high a sampling rate as you can practically achieve (it will be limited by your audio interface or your computer's ability to handle the workload), keep it at a high sample rate through mixing, but when you do a final mix-down (to a .wav, etc) don't bother generating anything higher than 44KHz (unless it's to hand off to someone else who will be doing more mixing and processing on the track).

 

[tr0n]

That means 40KHz is enough to guarantee that any human alive will hear a perfect reproduction. CD quality audio pads that a little more just to be safe and so we get 44KHz as the standard. Unless you're a freak of nature, you shouldn't be able to tell any difference between tracks with sampling rates above that.

This is from the first page of the thread.

CDs are at 44.1kHz not 'to be on the safe side' but because decades ago when engineers were looking for the best way to archive data digitally they decided on VHS cassette tapes, by recording an audio signal as if it was a video signal.

The below information I have re-written from: http://www.cs.columbia.edu/~hgs/audio/44.1.html (this also briefly explains why we have 48kHz and implies why higher sample rates are necessary).

Because there were two video standards back then (PAL & NTSC), if data needed to be shared between continents that used different VHS playback devices then there needed to be a common multiple that allowed the audio to be read on any machine.

The PAL standard is 625 lines per 25 frames (2 frames per field = 50Hz), and with 37 blanking lines per field:

294 lines per field x 50 fields per second x 3 samples per line (RGB) = 44.1kHz

The NTSC standard is 525 lines per 30 frames (60Hz) with 35 blanked lines:

245 lines per field x 60 fields per second x 3 samples per line (RGB) = 44.1kHz

So 44.1kHz was therefore the lowest common multiple between both standards and happened to be a little above the 40kHz required sample frequency.

As the article says, CDs were recorded using the same technology, so that is why they became 44.1kHz.

I have a ton of articles from when I studied audio engineering a couple of years ago. I'd suggest anyone genuinely interested in sample rates read this essay: http://www.lavryengineering.com/documents/Sampling_Theory.pdf

And for DSP in general: The Scientist and Engineer's Guide to Digital Signal Processing

Edit: As for 192kHz I recall reading something that suggested it's down to the brain's ability to distinguish between separate events. 1/192kHz = ~5x10^-6s. But I'm not entirely sure about this.

 

[thraxil]

Nice. I didn't know the history behind exactly why 44.1 was chosen. Though I probably should have since my computer engineering degree is from Columbia and Henning Schulzrinne (who wrote that page) was one of my profs.