Sample rates and bit depths

96k is twice as detailed as 48k. The article doesn't cover it, but what we're talking about is 48,000 (or 96,000) samples per second. The simplest analogy is with film's frames per second. When you record sound in the digital domain, you're taking a series of frozen 'snapshots' of a moving, organic thing (a sound wave), same as you do when you film something. Play the snapshots back in sequence and you end up with a representation of what you were recording. More snapshots/samples/frames per second equals a more accurate representation.

I don't know what you're using them for, but 96k is almost certainly overkill, unless you're doing something specialised like recording source material to be slowed right down for sound design purposes. To put it in perspective, 'CD quality' is 44,100 samples per second, so 96,000 is more than double the resolution of a CD.
 
96k is twice as detailed as 48k. The article doesn't cover it, but what we're talking about is 48,000 (or 96,000) samples per second. The simplest analogy is with film's frames per second. When you record sound in the digital domain, you're taking a series of frozen 'snapshots' of a moving, organic thing (a sound wave), same as you do when you film something. Play the snapshots back in sequence and you end up with a representation of what you were recording. More snapshots/samples/frames per second equals a more accurate representation.

Sorry, this is completely inaccurate!! 96k is exactly the same "detail" (or "resolution") as 48k but simply extends that "resolution" over twice the audio frequency range. A 48k sample rate will record up to an audio frequency of around 24kHz, while 96k extends that audio range to around 48kHz. As human hearing does not respond to audio frequencies above about 20kHz, a sample rate of 96k is therefore pointless for anything except a few specialist applications.

Furthermore, you seem to be confusing digital audio with analogue film frames. Your analogy is incorrect because the digital audio samples ("snapshots") cannot be, and never are, played! In order to reproduce ("play") and hear digital audio it needs to be converted to analogue and this reconstruction process is effectively perfect at any sample rate (providing the sampling rate is at least twice the audio bandwidth). A sample rate of say 2k will therefore reproduce an audio signal (with no content above 1kHz) exactly as accurately as would a sample rate of 96k. In other words, "more snapshots/samples/frames per second" DOES NOT "equal more accurate representation"!

OP: Always aim for 24bit 48k, which is the standard for all TV/Film/DCP and other A/V distribution formats. If you find/source material which is at 96k, somewhere in your software you'll find an option to convert it to 48k.

G
 
Thanks APE. I'm not confused. Rather, I was trying to offer an extremely broad-brush-stroke analogy to someone who seemed not to have a grasp of the basic concepts involved (with no disrespect to him intended). I picked the 'film' analogy because this is a film-making forum, so I hoped this would be a meaningful comparison in his mind - I wasn't trying to suggest that film frames and digital audio are literally the same.

Clearly your explanation is more accurate than my broad analogy, but I'm not sure it's any more helpful to someone who doesn't have an existing understanding of the fundamentals of digital audio.

If the op is confused further as a result of my posts here, my apologies to him.
 
No problem mike. I didn't intend to come across as being so harsh on you personally, it's just that there's a lot of misinformation out there regarding digital audio and it's being deliberately perpetrated for no other reason than to con consumers. I first got into recording and mixing in higher than 16/44.1 digital audio 20 years ago and some of what I've seen done by equipment manufacturers, retailers and record labels is shocking and borderline illegal! I didn't intent to take any of that out on you, my apologies if I came across that way.

G
 
Mike: Rather than using snapshots/film frames as an analogy, which is very misleading, I prefer to use use the circle analogy, which is much more accurate and can be used to demonstrate both the deliberately misleading/propaganda view of how digital audio works and the way it actually works.

In this analogy, imagine that we want to digitally store and recreate circles:

A. Let's say we measure 12 different points on our circle and store that data. When we want to recreate the circle we plot out those 12 points and then draw a line between each of them. What we would end up with is a vague, jagged edged circle shape made up of 12 straight lines (in reality it's actually a dodecahedron). To make the circle smoother and more like a perfect circle we would need to increase the number of points we measure and store. 120 points instead of 12 would logically result in a recreated circle which is ten times more accurate and looks much more like our original circle. However many points we use though, the circle is never going to be perfectly accurate because the coordinates of a perfect circle constantly vary. To get an absolutely perfect recreation of our circle would require an infinite number of points, which of course is both impossible to measure or to store. So, we're never going to get a perfect circle but the basic principle is that the more points we measure and store, the higher the resolution/definition/accuracy of our recreated circle.

This is all very simple to understand and all perfectly logical (which is why it's such effective propaganda). However, it's also complete horsesh*t because this is not how digital audio works!! This is a much more accurate analogy of how it really works:

B. We start off by designing a computer program which only does one thing, it draws perfect circles. Next, we measure two points on our particular circle and store those two points. When we want to recreate our circle, we feed those two stored points into our computer program which draws a perfect circle bisecting those two points. We've now got a perfect recreation of our original circle! What's important here is that two points are all that are required to perfectly recreate our circle. If we were to measure and store 10 points or 10,000,000 points and feed them into our computer program the resultant circle would be no more accurate than the circle recreated with just 2 points, if anything, there's a very small chance that it would be less accurate because there's a great deal more data to process and therefore more opportunity for error!

Although massively simplified, analogy B is fundamentally how digital audio works, although of course we measure and recreate sine waves rather than circles. The fundamental principle underlying digital audio is that it's predicated on measuring and recreating ONLY perfect sine waves and that's why it works so accurately. Analogy B also explains why the sample rate has to be at least double the audio frequency, because you need at least 2 measuring points (samples) per sine wave to perfectly recreate it.

G
 
Last edited:
Extremley high freq aliasing noise is the only problem that I ever encounter in the digital realm with regards to sample rate.

I don't know that it's ever really a major concern for most film people working with recorded sounds, but it crops up from time to time when synth patch building for instance. Overtones and harmonics can be crazy horses sometimes. :P
 
Extremley high freq aliasing noise is the only problem that I ever encounter in the digital realm with regards to sample rate.

What I've described above is a hugely simplified analogy of digital audio theory. How digital audio theory is realised in practise, by the manufacturers of ADCs, DACs and the programmers/makers of digital audio processing software is another matter! Anti-alias (and anti-image) filters were certainly one of the biggest problems/issues of the practical implementation of the theory but today (and for quite a few years) even relatively cheap consumer products achieve 120dB of attenuation in the stopband (audio frequencies above the Nyquist Point), which puts any alias artefacts below the system noise floor and renders aliasing a non-issue, with two exceptions: 1. With very high sample rates (sample rates above about 110k) we start running into the limits of the laws of physics and there is simply not enough time to adequately process the amount of data. With a sample rate of 192k the best performance available is reduced to about 80dB of attenuation in the stopband and therefore the potential for audible alias artefacts. 2. While hardware designs have effectively eliminated the aliasing problem (except at very high samples rates), this isn't necessarily true of all digital audio processing software; it all depends on the complexity of the processing, the knowledge/skill of the programmers and the inevitable compromises required in the design of any commercial product.

In other words, if you're operating your synth at 44.1/48k or even 88.2/96k and are hearing aliasing artefacts it is due to either the incompetence of the programmers or a deliberate design compromise rather than being due to any inherent, un-solveable problem with sampling/sample rates. Potential solutions if you are experiencing alias artefacts are to try running the synth at a different sample rate (maybe the programmers implemented more efficient anti-alias filters at a different sample rate) or buy a better designed/programmed synth.

G
 
I should've added that my post wasn't a request for help, just pointing out that in decades of audio production I've only ever come across it very occasionaly when building extreme synth patches - and that it's something that the OP and most filmmakers will likely never encounter. :)
 
Last edited:
Just to cause confusion and disharmony...
if we did the circle analogy with MIDI then your your limited to only 2 digits of precision for the value of Pi and R when using Circumference = Pi * R squared when playing back the performance information needed to create your circle on the circle synth-o-ploter-thingy

MIDI IS THE LANGUAGE OF THE GODS
 
Just to cause confusion and disharmony...

Well, you certainly are adding to the confusion :), for two reasons:

1. I was talking about sample rates, the number of measurement points, rather than the accuracy of those measurements, which is defined by bit depth. In digital audio, the accuracy of the measurements is irrelevant, which is a basic tenet of the Nyquist/Shannon Sampling Theorem, upon which digital audio is based. It's entirely possible to achieve perfect accuracy in our digital audio "circle" using only 1 bit of data for each measurement. In fact, the highest fidelity audio format ever released to consumers did just this; SACD is a 1 bit format.

2. MIDI is not actually an audio format. Using our "circle" analogy, MIDI does not (and cannot) contain any information pertaining to circles (sine waves). MIDI is effectively a control language, used to store certain settings of a MIDI compatible device, such as a synth or sampler. Using a photography analogy; there are many different digital formats for storing images (jpeg, gif, tiff, etc.), MIDI is not like any of them because it does not contain any image data! Instead, MIDI stores the information about your camera's settings (shutter speed, focus, etc.). Replay this MIDI data and your camera will be reset to those exact same settings. Provided your camera hasn't moved and the subject and lighting hasn't changed, you should be able to recreate a previously taken photo. If anything does change (what you're photographing for example), obviously you'll end up with anything from a slightly different to a completely different photo and MIDI would be none the wiser because it's not storing any information about what you're actually photographing, just data about how you photographed it. Additionally of course, without a camera this (MIDI) file of camera settings is completely useless. Likewise, if you feed this file of camera settings into a different camera, you may or may not get in the vague ball park of the original photo, depending on the characteristics of this different camera (whether it even has the same settings, it's lens, etc.). In effect, MIDI stores some of the physical data about how an instrument has been played (when a note is played, which note is played, how strongly it's been played, etc.) but does not store the actual sound the instrument produces. Without that instrument to recreate the sound, the MIDI file is useless and feed that MIDI file to a different instrument and obviously you'll get a different sound.

G
 
Last edited:
I really was only having fun.. but I must defend the MIDI gods..
midi is more than settings, its actual performance information, how hard you hit the note, how long you held it, what after touch did you do, etc. So in our analogy a CAMERA that took a MIDI "image" would store the information about the scene, not the pixels, so a MIDI image of a circle would include the information on how to MAKE that circle (Pi*R squared) but not a picture of a circle. I suppose more like vector based graphics (adobe illustrator) rather than raster graphics (photo shop)...
That was fun, and thats all it was meant to be :)
 
midi is more than settings, its actual performance information, how hard you hit the note, how long you held it, what after touch did you do, etc.

Nope! A MIDI "note on" command is a setting, as is a MIDI "note off" command. BTW, MIDI doesn't actually know how long you you held the note, it just stores the note on and note off command which when sent at the same time as you recorded them will recreate the same duration. MIDI does not contain any information about "circles" though, what sort of circle/s you get is entirely dependant on how the programmers of a particular synth (or other MIDI instrument) have programmed that synth to respond to those MIDI commands. So not only do you get very different circles with the same MIDI data from synth to synth, you also get very different circles from patch to patch within the same synth. In fact, that's the whole point of synths having banks of different patches!

G
 
WARNING: THIS IS OFF TOPIC TO THE OP.. but I have to answer..


its both..

MIDI note ON ----time passes --- MIDI not off. That is performance data. Not settings.

AND

.. you can use the MIDI protocol to send any value from 0 - 127, including settings data. MIDI is a communications protocol. You can use it to send numeric data. Thus IN OUR ANALOGY, I could use MIDI protocol to transmit the parameters needed for a synth to create a circle, all I would need to transmit on MIDI is the Radius of the circle and the X,Y, coordinates of the center of the circle. The synth would then do the math and PLOT a circle with radius of what ever I sent at the XY coordinates of what ever I sent. In this video a guy actually connected a Plotter to a midi keyboard..

hes not drawing circles but you can see it would be potentially possible.

https://www.youtube.com/watch?v=Eu_ciYIoPtI
 
Its both.. MIDI note ON ----time passes --- MIDI not off. That is performance data. Not settings.

No, it's just a setting! For example, a Note On command followed by a Note Off command does not define the duration of a sound. What defines the duration of a sound completely depends on how the programmer of a particular synth patch has programmed that patch to respond to the Note Off command. The note may end the instant the Note Off command is received or it may continue for a fraction of a second or many seconds after the Note Off command is received, depending entirely on how the programmer has designed the synth patch. MIDI does not contain any data about the duration of the sound produced. Let's use another example you cited: "what after touch did you do". How exactly does the "After Touch" parameter affect the sound created?

Thus IN OUR ANALOGY, I could use MIDI protocol to transmit the parameters needed for a synth to create a circle

No you can't! You can send data using the MIDI protocol to tell the synth when to start drawing a circle (for example) but MIDI contains no parameters or data of any kind which defines the actual circle itself!

.. you can use the MIDI protocol to send any value from 0 - 127, including settings data. MIDI is a communications protocol ...

I started programming with MIDI nearly 25 years ago and a decade ago I was teaching MIDI at university, so I really don't need any instruction on the basics of what MIDI is!

G
 
APE,
I want to give you the benefit of the doubt here, you normally know exactly what you’re talking about and I would defer to you on any topic regarding audio. We must just be failing to communicate at some level
For example your comment:
How exactly does the "After Touch" parameter affect the sound created
Suggests that we might be using different terms for something that I'm sure you must be aware of. Perhaps you know it as "Polyphonic Key Pressure" which can be found in the MIDI spec here: http://www.midi.org/techspecs/midimessages.php
Also this comment:
a Note On command followed by a Note Off command does not define the duration of a sound.
Though correct in relation to the SOUND, does not relate to the issue of MIDI transmitting performance data. The data transmitting from a MIDI instrument, AKA Controller, IS time based and thus the “Note On” followed by a “Note Off” does indeed denote how long the note was held on the MIDI controller. A synth can do anything it wants with the information just as you described, but it is FACT that the MIDI controller is transmitting MIDI performance data, which has NOTHING to do with the actual sounds generated.

I KNOW digital communication protocols, of which MIDI is a very basic example. MIDI is a time based network protocol, wherein each midi event, which is composed of numeric data, is transmitted serially, one event after the other. Regardless of how nodes in the network respond, or what the data being transmitted is, the only thing going over those wires is data. I can send any data with the limit being numeric values from 0 – 127 (0-255 for status bits), however I can send any number of them that I like, thus any data you can imagine can be sent over MIDI. The MIDI protocol allows for custom data to be sent over MIDI, using SYSTEM EXCLUSIVE operations (also on the doc linked above)
In response to:
… MIDI contains no parameters or data of any kind which defines the actual circle itself!
This is true, but for that matter neither does MIDI contain any parameters for any sound, it only contains numbers that represent the performance data. There is NOTHING about the actual SOUND in MIDI data.
For example: say we had this MIDI data in our synth buffer..
Code:
144 060 120
This is just data… the MIDI protocol only requires the synth to accept this value as valid input and forward it out any THRU ports, but beyond that the synth can do what it wants with it.
A piano synth for example might interpret the first byte, 144 as a “channel one note ON”, the second byte 060 as the note “Middle C”, and that the third byte 120 as the velocity at which the “middle c” key was pressed. The synth then takes the performance data and uses it as input parameters for the SOUND it will generate. The mapping between channels\notes and actual sounds is beyond the scope of this example, but sufficient to say that the performer pushed the middle c key on his keyboard and a sound non-unlike a piano middle c was generated by the synth.


No imagine a different synth, our imaginary synth of the analogy. Our new synth would use that same data differently. The imaginary synth would use 144 as the X coordinate on some grid and 060 as the Y coordinate on that same grid, thus defining a “point” and the third byte 120 as the radius value. Our imaginary synth would then instead of making sound, would use those three inputs to plot a circle on a screen or piece of paper at the position defined by points 144, 060 with a radius of 120. How the synth does this is irrelevant to the discussion in the same way the channel\note to sound mapping is irrelevant in the above Piano synth example.

Thanks
 
Back
Top