LOL ^
Equally worth looking at is the partner video, the ‘Digital Media Primer’ video, which is found here:
http://xiph.org/video/vid1.shtml
What is particularly good about this video is that the creators change the bit depth / sampling rate of the audio on the fly during the presentation, which is a very practical audible technique to help answer the original poster’s question (take a look at the video from about 10mins onward). Even though it doesn’t specifically use a 96kHz sample rate, it helps make the link between digital audio bit depth/sample rate and what we hear.
Wow. I really enjoyed that video. Clears up a lot of opinionated Dogs Brown.
So, what are the actual pros and cons of recording at higher sample rates (bearing in mind I can’t hear any difference above 44.1khz)? With Bitrate, I’m more than happy with 24bit but could someone explain what the benefits of using 32bit Float are? I’ve never understood it!
Thanks
Jono
Pro - 96kHz conserves the audio in a higher quality format for future high quality distribution formats (or for mastering requirements where mastering engineers may request the original files in 24-bit/96kHz format).
Pro - 96kHz gives you a slightly better chance of capturing the best possible recording, when compared to 44.1.
Con - uses more computer resources / storage space.
Con - subsequent sample rate conversion of 96kHz when the audio is released on CD (coversion to 44.1kHz) or some other format can damage the audio, particularly if you use poor SRC. Once again not everyone is going to hear the damage.
This is completely unrelated, but for the sake of edification bit depth is more accurate. A word is a fixed length (depending on the CPU architecture) so word length has no meaning since it is fixed and cannot be changed.
Pros:
- In case you have non-oversampling A/D and D/A coverters with poor anti-aliasing/anti-imaging filters the sound quality may improve slightly (flatter freq response near 20kHz and/or less aliasing distortion).
- Some DSP algorithms may introduce aliasing distortion if not implemented correctly. So if you have ill-behaving plugins you can move artifacts they are creating to inaudible frequency range by using higher sample rates (and then removing them completely by downsampling with good quality sample rate converter).
#1 hould be non-issue nowadays (for last 10 - 20 years now depending on if you use pro or consumer grade converters). However #2 may still be valid point today.
Cons:
- Storage space requirements
- Processing power requirements
- Some D/A converters may perform poorly with high sample rates
- Introducing ultrasonic content into your analog audio chain (anything after D/A) may cause unwanted intermodulation distortion.
That sounds very familiar 
How do I convert tape to MIDI?
Thank you for the advice. I use a MOTU HD192. Unfortunately I have not yet been fortunate enough to try anything out of this price bracket (yet). I’m just wondering how the HD 192 fares in quality and whether it would be a good idea to Record at higher sample rates? Also, I’m sorry to ask more questions but you seem very knowledgeable, could you explain what 32bit Float is and why I would use it over 24bit (I apologise if this is hijacking the thread but it’s all important stuff to know about).
Jono
Recording in 32-bit floating point is only benefitial if you are going to do a lot, actually a LOT, of offline processing.
Online processing (real time mixing) is always executed in the 32-bit floatig point realm, no matter the bit-lenght of your source material.
Sorry, I can’t tell for sure. I have no experience on HD 192. But it’s modern pro/semi-pro unit and should be OK and not having problems of early cheap systems. According this article: MOTU 2408 Mk3, 24I/O & HD192 HD 192’s converters are 128x oversampling. This should eliminate all aliasing/imaging problems in A/D/A conversion. It’s probable you won’t gain anything by using higher sample rates (when it comes to A/D/A conversion).
Or LOT of bouncing.
Don’t know if you were disagreeing or just not understanding what I was trying to say.
Sample buffers provide protection against audio glitches while your computer’s processing audio. A 256-sample buffer provides less protection from these glitches at a 48kHz sample rate than it does at 44.1kHz, both because it represents a shorter amount of audio time at the increased rate, and because your computer has to work harder to deliver audio at that higher rate. So even though you may experience less latency while using a sample buffer of a given size at a higher sample rate, you’re also running a greater risk of getting clicks, pops, and dropouts in your audio if you don’t increase the size of the buffer to scale with the increased sample rate.
Neither. I was basically saying, “WTF are you talking about?!?!”
WTF are you talking about?!?! Protection? You mean like Cosa Nostra? That kind of protection?
Maybe you know what you are talking about, but you just don’t know how to say it.
First off, I assume when you say sample buffers, you mean audio buffers. Do you know what an audio buffer is? I’ll explain it. An audio buffer is a reserved segment of memory. It is used to hold audio data to compensate for processing delays, aka latency. When the sample rate is higher, the audio data will pass through the buffer faster, and then the latency will be lower. The smaller the buffer, the less time it takes for audio data to pass through it, and the more processing power/CPU work is needed. So if your processor is not up to the task, you may have to raise your buffer/increase latency, or work at a lower sample rate in order to keep your buffer and latency down. There is no ‘protection’ going on.
Cheers.
I think preserving more detail that we can’t hear (above a sample rate of 44.1K or 44K) is equivalent to recording more ultraviolet and infrared on our videos. Let’s not forget that we can’t hear digital. We can only hear actual sound pressure waves coming out of speakers, which are the result of conversion of digital to analog. They do not come out with stair steps. I’m using 44.1, which takes less space.
I do see an advantage to using 24-bit depth, as it gives you more headroom than you need. It takes a third more disk space, but I can’t imagine that’s a problem for anyone. You can’t make a mistake in input levels (too low I mean) and you can’t record a signal that’s noisy (internally, that is - you can certainly record a signal that has noise in the chain). You just don’t have to think about it. At 16-bit, you have to give it a little thought. At least in my experience. I noticed a difference when I switched from 16 to 24.
So – 44.1K or 48K and 24-bit. There you go. Use dither on mixdown to 16-bit. Or don’t. In truth, nobody will hear the difference.
Digital audio macho guys chest it out.
LOL, too funny! 
First - I didn’t author the article and have no connection to it.
Second - It’s not wrong. Nor garbage to be wholly dismissed as others claimed. It is what I said, a good place to START.
As for the 24bit 32bit float difference:
To sum it up in one line without the pages of technical engineering data:
32 bit float allows for processing or storage of a higher dB range (Dynamic Range) before being subject to waveform truncation (‘clip’).
With 32bit float, using the 6dB per bit rule, you end up with about 1530dB dynamic range.
24 bit is just over 144 dB.
Now how is this useful in processing/Mixing?
SOUND is the word we use for the translation of a waveform (or more appropriately to the topic of mixing MANY INTERACTING waveforms of different frequency)
to illustrate:
Take 10 complex frequency sources at 0dB and sum them to your 2-channel Buss.
At your 2-channel Buss you will be well over the 0dBFS limitation and your 24bit project will be a clipped, aliased and distorted mess.
But this is not the case with 32 bit float!
32 bit float allows for the dynamic range to be extended just beyond 1500dB - more than enough to handle the processing of the sources.
Now you can’t directly translate that 32bit float out through a 24bit D/A process but you can process it and store it digitally at 32bit float.
So if you saved the processing you are doing in a ‘Pulse-Code Modulated’ (xPCM) format like a WAV file at 24 bit you would have a clipped, aliased and distorted mess.
if you saved the processing you are doing in a WAV file at 32 bit float you would have the APPEARANCE of a clipped, aliased and distorted mess! Examination of the visual representation of the wave would look like a big fat sausage or block.
A simple “Normalize to 0dB” would restore the entire stored encoded waveform back to listenable without it being a clipped, aliased and distorted mess.
So where it matters internally is that you can drive the gain of a source , or summed sources, well into the RED (above 0dB) and enjoy all the many benefits and effects it has on your other sources while being summed or processed.
Does that make the concept a little less muddy?
(probably not! but the more you learn about it and experiment with it the clearer it will become!)
Thats actually why I am using 32bit float, it is to easy to accidentally destroy a bounce or mixdown.
It’s a safety net, that IMHO won’t tax your computer that much more than 24bit.
It’s actually more than a safety net!
It is an indisputable fact that 32bit float can more accurately represent a source signal than can 24 bit or 16bit or 8bit.
In fact, 32bit (according to the math) is about 65,536 times better at accurately representing a signal than is 16 bit and 256 times more accurate than 24 bit!
If we were only recording a single source signal and then outputting that unprocessed single recorded signal, we could probably get away with 16 bit or lower and still get a halfway decent representation. Though outside the test lab nobody is recording a 1khz sine wave, exporting it to CD and calling it music!
When we are processing and manipulating the multiple complex frequency signals (as happens during mixing) you want to start with the best possible sources and give it the best possible processing in order to get the best possible end result.
Nobody said you did.
Yes it is. It’s garbage. First of all it plays “connect-the-dots” game, which is fundamentally flawed approach to digital audio. I don’t have to go any further into details. This alone makes it wrong, garbage, flawed, utter cow excrement! It is not good place to start. It’s horrible place to start, because it gives you wrong impression on how digital audio works. It’s as good place to start as ancient greek mythology when you try to understand basic laws of quantum physics.
You are comparing apples to oranges: fixed-point vs floating point. 32-bit float is in matter of fact just as “accurate” as 24-bit fixed, but in much wider dynamic range.
Are you talking about The Good Old “Stacking” Myth? That quatisation errors of multiple tracks will magically accumulate into final mix? Please tell me it isn’t so.
Of course not. But still 44.1kHz/16-bit digital audio can exactly reproduce any incoming music if
- It’s bandwith limited to less than 1/2 of sampling freq
- You don’t count quantisation distortion/dither noise below -96dBFS.
That’s scientific fact. Proved decades before anyone even thought about digital audio.
But what if your “source” (audio moderately isolated room recorded with microphone through pre-amp and converter) already has about 80dB dynamic rage (something you will get in home/project studio if you are lucky)? Would 24-bit recording help you? Of course not, the last 8 to 10 bits are just noise.
Once and for all … just watch and listen to what people who really know has to say…
Monty Montgomery about sampling: https://www.youtube.com/watch?v=cIQ9IXSUzuM
Ethan Winer about audio myths (including bit depth around 46 mins): https://www.youtube.com/watch?v=BYTlN6wjcvQ
Oh boy…
I’ve already watched the video by Monty.
You spoke up. But aren’t really listening (nor reading apparently)!
And until you actually “learn it” and prove it to yourself rather than repeat someone else’s idea (and bad info too) after watching a fixed test on youtube it’s pointless to go further - it’s like arguing with a religious zealot.
I happen to record acoustic sources at the high sample rate and increased bit depth because there is a definite quality difference!!
Whether Monty, Ethan or Jarno say different …
Not only does Physics back it but It is something easily noticeable to my ears doing blind comparison tests!
I would also go with the actual tested results at DOLBY Labs long before Monty or Ethan!
You know them, they’re the real scientists who have spent billions over the years to determine that 96kHz 24bit is the most appropriate for accurate 8 channel reproduction in the world’s movie theaters (and 192kHz 24 bit for 6)!
But I guess despite all the evidence and science you must be right, since you watched a video on youtube.