quote:

Originally posted by DigiNut OK, what? Man, sorry, but you're not making much sense on these forums lately. How could you possibly export a 32-bit float wave file if the sequencer wasn't already keeping the information in floating-point form? Are you actually suggesting that it upconverts when you bounce? Export is export. It's not the same as mixing. No matter what you export, you're dithering from 32-bit float (unless you export in that format). Now if you're talking about the "Record Format" in the Cubase project options, that refers to when you freeze instruments or process audio tracks. Or when you - get this, I know it'll shock some of you that this actually still happens - record a live instrument or voice. And there's a very good reason to use 16-bit or 24-bit instead of 32-bit: space. Most of you probably never touch the audio features in Cubase but some of us do, particularly if doing trendy edits like stutters and glitches. My last project had over a gig of audio files lying around; if the audio had been kept in 32-bit it would have been between 1.5-2x as large. Incidentally, this is with NO freeze files or full bounced tracks; just for effects and edits. Higher bit depths also place a greater strain on system resources, particularly the hard drive. If you're playing LOTS of raw audio at the same time, then using 16-bit instead of 32-bit tracks effectively doubles the amount your system can handle. In practice, it's not quite double, but it's close. Same with the sampling rate; you're cramming twice as much data into 1 second of audio if you double it. So the question of "why would they even give you that option" is a rather silly one. A lot of producers don't seem to realize that higher quality isn't free, and if you can barely hear the difference or can't hear it at all, then you've hit the point of diminishing returns and probably ought to lower your bits and/or rates. Even if you do all your internal recording and final exporting at 16 bits, though, the internal mixing in a software sequencer is still happening at 32-bit float. It just is. EOD.

Alright, that makes sense.

In regards of making sense, in a certain previous thread you didnt make sense at all, but arrogant enough to drop the discussion thinking u were right, while i could easily prove you wrong, but ur never wrong are you ?

___________________


http://soundcloud.com/subtara

Last edited by Subtle on May-15-2008 at 14:49

quote:

Originally posted by Derivative not unless they were created digitally with 32 bits worth of dynamic range to begin with

this is what i meant by freezing and bouncing parts - usually they will have some kind of effect applied. likewise, many (most?) people don't have hardware, so we're talking about soft synths running in 32bit already. of course, if you have an unaffected sample in lower bit rate, there is no benefit from upconverting, hence why i don't do it on import.

quote:

Originally posted by Derivative Besides, the difference at normal listening levels is so unbelievably small that unless your mixdown and sound design is as good as its ever realistically going to get, theres no point even thinking about dither unless you are training to be a mastering engineer or something.

one could say the same about 96k

point taken though, there are upsides and downsides to everything.

quote:

Originally posted by derail you have a completely clean mix which doesn't clip and peaks at -0.1 or one which peaks at -3.0. A mastering studio will be able to use either.

I wouldn't be so sure...I was always taught that you shall give the mastering engineer a bit of headroom...so peaking at -3db is much better...I think

but that's a bit off topic "bit loss" :>

___________________
Help me make soundtrack for my grandpa´s art exhibition:
Check out my topic about it here
Thank you!

Last edited by BOOsTER on May-15-2008 at 20:10

I'd say some mastering engineers would ask for -3dB, just so they know the chance of digital "overs" is very remote. But if they receive a file peaking at -0.1dB, and it has no master compression applied and there are guaranteed no "overs" (no clipping) in the whole song, they'll be just fine using that. If they, for some reason, need to reduce the input signal, they can just reduce the input signal on the first processor in their chain.

People here are saying it is pointless to use any thing higher than 44.1khz but, does no one hear the difference in your softsynths sound? Like smoother filter transitions. Maybe it is me but, I notice a huge difference. That being said, I have actually been running at 44, but only to save cpu cycles since my projects have been growing with my knowledge.

Nyquist for a 44.1khz samplerate recording is 22.05khz. This is the highest bandwidth you can represent before aliasing occurs and is fortunate because it is higher than the entire bandwidth of ideal human hearing.

But...lets say you use an oscillator to generate a square wave at 3khz.

The fundamental will be 3khz. The 3rd harmonic will be at 9khz, the 5th harmonic will be at 15khz, the 7th harmonic will be at 21khz, the 9th harmonic will be 27khz and the 11th harmonic will be at 33khz.

These harmonics linearly drop off in power and the 7th and 9th harmonic will both be inaudible. The 9th harmonic however, will alias at 17.55khz because it is over Nyquist by 4.95khz and the 11th harmonic will alias at 11.1khz because it is over Byquist by 10.95khz.

In order to prevent aliasing of an inaudible frequency in an audible frequency range you need to make Nyquist higher (and thus make more available bandwidth before aliasing). an 88.2khz recording has Nyquist at 44.1khz which is more than double the bandwidth of human hearing and you might think unnecessary as a consequence but in the example above the 9th and 11th harmonic of a 3khz square wave will not alias (nor for that matter will the 13th harmonic).

There are broadly 2 ways of solving aliasing practically:

1) Oversample your AD/DA and oversample your oscillators, stick an anti aliasing low pass filter at the reasonable upper limit of human hearing (20khz roughly) and then downsample back to native samplerate.

2) Do everything natively at a much higher samplerate.

Option 2 is simpler but very heavy handed on your computer as all digital audio recordings take up lots more physical space on hard drive and in RAM seeing as they comprise of more samples. Any realtime post processing will have to work on more samples too so you will find that plugins have an increased strain on CPU load.

Most of the time you can't oversample synth oscillators. Vanguard has a switch on the rear panel which lets you switched between regular oscillators (lots of aliasing) and non aliasing oscillators (oversampled, much less aliasing). You can hear the difference too if you play a single sawtooth/square wave tone in the lower/upper mids sort of range and pay attention to the treble end of the spectrum.

By the way, I run at 44.1khz because my PC is not that fast and increasingly I find that the CPU/Memory load is just too much to bear at 96khz. Literally 2 instances of SIR at 96khz make my CPU load shoot up 60%. Its crazy.

Last edited by Derivative on May-16-2008 at 02:05

I seem to remember someone saying that 88.2 was better than 96 because if you go down to 44.1 it would be more accurate. Is that not right?

quote:

Originally posted by Ray_Chappell I seem to remember someone saying that 88.2 was better than 96 because if you go down to 44.1 it would be more accurate. Is that not right?

Use 88.2khz if you intend to go down to 44.1khz eventually because the samplerate conversion (SRC) will be synchronous.

Similarly use 96khz if you intend to go down to 48khz because again the SRC is synchronous.

Alternatively, don't SRC and stay at the native samplerate (i.e. 96khz which is dvd audio sampling rate anyway)

Synchronous SRC is easy because the ratio is 2:1 or 1:2 or whatever integer value. To go from 88.2 to 44.1 is well easy - you stick an anti aliasing filter at 22.05khz (Nyquist frequency for the destination samplerate) and then you simply throw away every other sample.

If you wanted to get 96khz down to 44.1khz you would need to use a clock frequency with common denominator of 96 and 44.1 (i.e. 14,112khz) otherwise the two clock trains will be running at different periods so you can't just throw away every other sample. Its more complex than I understand so the extent of it is that an asynchronouse SRC has to be well designed otherwise it will be more prone to calculation errors. They are more expensive to build to a high standard of quality. Synchronous conversion is computationally easy and cheaper to build to a high quality.

quote:

Originally posted by Derivative Nyquist for a 44.1khz samplerate recording is 22.05khz. This is the highest bandwidth you can represent before aliasing occurs and is fortunate because it is higher than the entire bandwidth of ideal human hearing.

Not to be a pedant, but I wouldn't call it good fortune, I believe the sample rate was picked very deliberately for precisely that reason.

___________________
My party schedule:
2009-02-21 - DJ Attention @ I'm So Popular
2009-06-18 - DJ Annoying @ People Need To Know Where I'll Be
2012-11-32 - DJ Insufferable ɸ Or At Least the Stalkers I Complain About
2048-06-66 - Spastic & Whocares ¶ Although I'm Actually Flattered
9999-45-81 - Tweaker Gimp ☼ I Probably Won't Even Go To This But I Have To Make Sure I Fill Up All The Available Space Here

quote:

Originally posted by Derivative Use 88.2khz if you intend to go down to 44.1khz eventually because the samplerate conversion (SRC) will be synchronous. Similarly use 96khz if you intend to go down to 48khz because again the SRC is synchronous. Alternatively, don't SRC and stay at the native samplerate (i.e. 96khz which is dvd audio sampling rate anyway) Synchronous SRC is easy because the ratio is 2:1 or 1:2 or whatever integer value. To go from 88.2 to 44.1 is well easy - you stick an anti aliasing filter at 22.05khz (Nyquist frequency for the destination samplerate) and then you simply throw away every other sample. If you wanted to get 96khz down to 44.1khz you would need to use a clock frequency with common denominator of 96 and 44.1 (i.e. 14,112khz) otherwise the two clock trains will be running at different periods so you can't just throw away every other sample. Its more complex than I understand so the extent of it is that an asynchronouse SRC has to be well designed otherwise it will be more prone to calculation errors. They are more expensive to build to a high standard of quality. Synchronous conversion is computationally easy and cheaper to build to a high quality.

Gotcha - thank you.

quote:

Originally posted by Derivative To go from 88.2 to 44.1 is well easy - you stick an anti aliasing filter at 22.05khz (Nyquist frequency for the destination samplerate) and then you simply throw away every other sample.

Ok, this part confused me. I know what a filter is and all but, ok lets say I wanna work in 88.2 then dither down to 44.1. Does this mean I have to use a special filter? Or it automatically does when you dither down?

And also, when dithering down in Ableton Live, you have the 3 POW-r modes plus rectangular and triangular, I don't really understand all the modes. How do I know which is best for me? If it is too much for someone to explain, do you have a link or something that I can read? I very much would like to run natively in the higher rates and dither down, as I CAN hear the difference but, usually when I dither down it sounds worse, and, well I think partly it was because I was going from 96 to 44.1.


So if anyone can explain,----> Filter and POW-r modes, please, and thanks in advance