Hard disk recorder modes, int32 vs. float32?

oskari
Site Admin
Posts: 296
Joined: Mon Nov 21, 2011 2:04 pm

Re: Hard disk recorder modes, int32 vs. float32?

Post by oskari »

Here's a 7-bit float (without special values like zero).

Image
oskari
Site Admin
Posts: 296
Joined: Mon Nov 21, 2011 2:04 pm

Re: Hard disk recorder modes, int32 vs. float32?

Post by oskari »

And here's a 7-bit integer.

Image
Dr3Tri
Posts: 27
Joined: Sat Dec 17, 2011 9:01 pm
Location: Finland
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by Dr3Tri »

Thanks for clearing this up. Your explanations really gave me some light in this question.

So, as my use of wavs is that i export songs to myself instead of some recording studios and so on, i start to use 32bit float from now on. Especially as i like to do further modifications to wavs with external wav editor (i still use good old Cool Edit 2.1).

I use CE to cut silence, amp env for in's and out's, normalizing and limiting. CE has user modifiable keyboard shortcuts for favorite commands, which have made editing really QUICK for me. Especially after using it all this years... i probably wouldn't ever switch to other program.

AFAIC.anU.nderstand it is good to do all those edits when file is still in 32bit float format, and when done, converting it down to 16bit wav and then FLAC.
Or if converting to MP3 or what ever lossy format suites best in situation, directly from 32bit float after all edits are done.
Per Lichtman
Posts: 16
Joined: Wed Jun 13, 2012 2:59 am

Re: Hard disk recorder modes, int32 vs. float32?

Post by Per Lichtman »

@Dr3Tri Yes, if we are talking about editing on your computer, 32-bit float is most direct export you can do because it matches the format that Buzz itself mixes in (as Oskari mentioned).

I would suggest that you also dither when creating lossy compressed files as well. You really want to dither anytime you go to 16-bit or less: that is pretty standard practice, though dithering when going to 24-bits gets argued about more often.

You only want to dither once at the very end and ideally you want to pick your dither algorithm carefully. The majority of audio programs do not use great dither algorithms by default. You could say I'm a bit of a nerd about this stuff - I pay a lot of attention to it if I'm doing final output, but I'm willing to make do with less if I'm doing drafts along the way.

I spent a lot of time bending George Massenburg's ear on the topic of that and sample rate conversion the first time I met him. He did a good job of emphasizing the importance of doing it well and I researched the topic a lot on my own. I found two very good resources for making up your own mind.

Here's the one for dithering. It gives a blind test of several different algorithms. One of my favorite traditional ones is MBit+ with Ultra shaping, but it's a blind test so you can make-up your own mind. As you can hear, there IS a difference that you can clearly hear when you amplify the results. These differences can sub-consciously affect your listener even when they may be quiet enough that the listener isn't really focused on the sound.
http://www.24-96.net/dither/

One of the only ones it doesn't compare is Schwa Psycho Dither. I don't think I tested it but I might do so soon since they have a free trial.
http://www.stillwellaudio.com/?page_id=27

And here's the resource I mentioned for sample-rate conversion (a separate but related topic).
http://src.infinitewave.ca/

One of the reasons that 24-bit has been the most common delivery format for mastering engineers (other than the things mentioned above) is the high percentage of people that work in Pro Tools. It is only within the last year or two that Pro Tools has started supporting 32-bit floating-point files (even though so many of the competitors have had over a decade). So if you work with anyone using an older version of Pro Tools, you have to deliver them a 24-bit integer file (which has very widespread support beyond Pro Tools).

When you're exporting/importing files, it's helpful to be aware of the native bit-depth of the DAW/sequencer/tracker/editor you are working with. Hosts like Reaper and Sonar support so called "double-precision" floating-point 64-bit. Most people don't feel they need that precision during export/bouncing/etc. so Reaper is often setup to use 32-bit floating-point during bounces/gluing/etc. by default (but you can switch it to 64-bit if you want).

DSP gear varies in the precision. Focurite Liquid Mix products used 40-bit precision if memory serves correctly (but I don't remember what format it output to the host) but I remember that UAD-1 cards output to the host using dithered 24-bit output.

As mentioned earlier, modern analog to digital and digital to analog converters normally support 24-bit integer (except in the complicated case of the Sony DSD format, which is overlooked for the purposes of this discussion). But when 24-bit integer files are edited in 32-bit or 64-bit floating-point programs, they can produce values that fall outside the 24-bit integer range as mentioned earlier. So sticking with floating-point lets you avoid worrying about any of that until something requires that you downsample.
User avatar
UNZ
Posts: 808
Joined: Mon Nov 21, 2011 9:42 pm
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by UNZ »

or you could trust empirical results on questions like dithering etc :)
they bust the myth that dithering to 16bit is really neccessary (altough it obviously is for 8bit etc)

http://www.youtube.com/watch?feature=pl ... YTlN6wjcvQ

Jump to around 32 mins. And at around 34 for dither.

oh and float 32 ftw :)
Per Lichtman
Posts: 16
Joined: Wed Jun 13, 2012 2:59 am

Re: Hard disk recorder modes, int32 vs. float32?

Post by Per Lichtman »

@UNZ Great link, but actually your implication differs from what that YouTube author said. :)

Here's a quote from Ethan Winer's follow-up article called "Does Dither Really Matter" that he links to on the AES webpage he references in the YouTube video you linked to. Bold emphasis was added by me.

"To be clear, using dither is never a bad thing, and it can sometimes help on soft material recorded at very low levels. So I'm not arguing against using dither!"

http://www.ethanwiner.com/dither.html

If someone is going to evaluate dithering, they need to use the uncompressed files on that page or the ones I linked to earlier- not the YouTube video or MP3s. In addition, the YouTube video (unlike the later version he posted in the link above) does not provide a good empirical reference because the noise pulses in and out, preventing the listener from being able to accurately compare the same section both with and without dithering. Why deal with trying to extrapolate as the result of an extra variable when you don't have to? :)

What the tests I linked to earlier did (and I recreated similar results for myself several times) was emphasize the effect in worst case scenarios by boosting the dithered and non-dithered results back up in level after the processing was performed. Sometimes this would be clearly audible but this also helps give a sense for what the listener is actually going to be responding to in less obvious situations.

It's been extensively documented that humans respond to sounds outside the dynamic and frequency ranges that are thought of as "audible." This is something I've had a chance to discuss personally with engineers like George Massenburg (experienced recording engineer, inventor of the parametric EQ and designer of the GML series of gear), Rupert Neve (the man behind original Neve consoles - some of the earliest and most desired mixers, and the more recent Rupert Neve line of gear) and various others.

If your main concern is "will this dither be extremely obvious in a loud mix" then you shouldn't worry about it too much. I'm addressing the question of "if you're really trying to get the best sound you can under the most demanding circumstances, what's the way to do it". The artists I've worked collaborated with or engineered for in the past are accustomed to the best (featured soloist and co-composer with Hans Zimmer, Julliard graduates, Grammy winners, etc.) and I am accustomed to finding to finding out everything I can about providing that.

My take on it is, the effort required to find out what dithering approach might suit your engineering style best is minimal. So why not make the effort, even if there are some conditions where it might be less noticeable than others? And if you're going to do that, why not test with the most demanding material you can find, as opposed to material that is designed to be "normal" in its demands? :)

Also, the 24-96 test I linked to earlier provides material that is equally useful for empirical testing (as well the blind preferences from hundreds of people that took the test which can be useful to check against your own preferences. :)
User avatar
UNZ
Posts: 808
Joined: Mon Nov 21, 2011 9:42 pm
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by UNZ »

they don't differ imho. everyone seems to hear what he wants to hear ;) He doesn't argue against it but also doesn't see a reason to really use it either (for 16 bit), so i let you decide what you make of that :) if something is 90db down, seriously i really dont give a fuck :)

and yes, if you mess with the levels obviously it makes a difference because you're not even using the whole 16 bit range then... the video says this too, "for properly normalized signals" and "I've never heard dither make any difference when applied to typical pop music recorded at sensible levels" from your link (and assuming that its not just 1 peak that goes to 0 and the rest hovers at -60db or something i guess).

blind tests: i think you'd be hard pressed to find people who constantly can tell which dither is which in a double blind test, but if you have studies about this post them... Myself, i fail the blind test, with very high end gear (rme fireface / akg 702) and (i believe) still good ears. It's not even documented that people can reliably tell the difference between a 320kbps mp and a wav file... where does that put dither...

I guess what i want to say is: use common sense, get the important stuff right before worrying about "pointless" details (which ARE pointless if you do the important stuff right).

i can see why you strive for excellence, and by all means, you should. Your clients deserve it and there's money to be made with it hehe. But i'm just saying that some of these things that people are trying to sell as the best thing to ever happen to your music are borderline snake-oil and just don't deliver much in therms of sound quality. Unfortuntately most people's time is limited, and they should focus on what really makes an impact on their music in that time.

keep it practical :)
User avatar
Joachip
Posts: 352
Joined: Mon Nov 21, 2011 7:21 pm
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by Joachip »

There's theory and there's real life. Let me stick to the latter, because I can't be assed to read this wall of text right now, but I want to throw in an experiment that'll give some answers, and then leave the conclusions up to you guys.

Dithering 24 bits when converting to 16 bits will cause a chance somewhere around -93 dB. You won't be able to hear it unless you first kill all life on earth to make earth quiet, or turn your volume up to a nonsense level. Basically noone can tell the difference between dither on/off if they're just listening directly to the result.

Dithering when converting down to 8 bits or even 6 bits make an audible difference, though! I highly recommend converting various kinds of sound (something with a lot of treble, and then something with very little treble and much bass) into 8 bits, with and without dithering, and then listen. You will be able to tell the difference! The tendency is for the noise to move to a higher frequency area when dithering is enabled.

But please, go ahead and try, instead of wasting your time on forums! ;)
cyanphase
Posts: 3
Joined: Wed Nov 30, 2011 12:01 pm
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by cyanphase »

I don't have much to add about dithering, I just needed to get my post count up so I can PM someone. :mrgreen:
Myself I like to keep things as 32-bit floats or higher when its possible.
Dr3Tri
Posts: 27
Joined: Sat Dec 17, 2011 9:01 pm
Location: Finland
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by Dr3Tri »

How about those dithering settings? It would take ages to just test them all with different combinations.
Could someone just recommend good all around settings (in CEP 2.1)?

My only use would probably be to convert 44.1khz float32bit files to 44.1khz 16bit files.
How about defaults (triangular, no noise shaping)?

Sorry about selfish question to general discussion. :D Of course it would benefit others too, but i don't know how used Cool Edit Pro is nowadays. But maybe those good overall settings would be same in different editors? If they have similar options in dither settings..
Attachments
Noise shaping
Noise shaping
p.d.f.
p.d.f.
Last edited by Dr3Tri on Wed Nov 14, 2012 11:58 pm, edited 1 time in total.
User avatar
UNZ
Posts: 808
Joined: Mon Nov 21, 2011 9:42 pm
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by UNZ »

even if dither really made a difference for 16 bit, nobody could tell you THE GOLDEN settings for your input...

i don't know, sometimes i wonder why i bother to post here, its like nobody even reads what people write, are we a write-only forum hehe ?
Dr3Tri
Posts: 27
Joined: Sat Dec 17, 2011 9:01 pm
Location: Finland
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by Dr3Tri »

UNZ wrote:even if dither really made a difference for 16 bit, nobody could tell you THE GOLDEN settings for your input...
i don't know, sometimes i wonder why i bother to post here, its like nobody even reads what people write, are we a write-only forum hehe ?
I have read this thread from start to end. The fact that i didn't quoted anything doesn't mean that i didn't read anything. There was that linked opinion (using dither is never a bad thing). So that's why i GUESS that using just about any settings would be beneficial. Some more, some less.

Besides i just disagree on most what you have said. Like this for example "I guess what i want to say is: use common sense, get the important stuff right before worrying about "pointless" details "
It doesn't do anything harm to do all those "pointless" little things as well as they can be done, before finalizing most important things, music itself. Who told us which is only acceptable order to do things? I bought my first CD before i even had player couple decades ago.

Earlier i have even converted mp3 files back to .wav's, did some editing and encoded it back to mp3. How rude is that? Even that can be done if files purpose, where you are gonna use it, allows it. Still, it doesn't do any harm to make this whole process all the way from bmx to FLAC/mp3 purposely. And after making it habit, it wouldn't take too much effort.

I believe that things that you can not directly hear can have impact on ones experience about music. Just like we can not see all the light that exists, but we surely get tanned by UV-B and have some serious health damages if being exposed to UV-C. We can't see IR over certain wavelengths but we start to feel it as warmth..0.

What if our soul would have some kind of more developed "ears"? What if our aura could sense something about music that our ears can not, and give us slightly different feeling about the content whether there is something or not. It doesn't bother me to make phases in the process that i can not actually hear. I would just be happy knowing that it even that small thing is taken into account in sound quality.
Then, after that process is in good condition, i might have better feeling to compose better music. That's my personal approach.

BTW. IF you even looked those pics I attached, there is 12 options for noise shaping and 5 for that p.d.f.. Now, how much does that do? I guess that there is no human being in this planet who has really compared all those possibilities against each others. That would take life time.

So, there just MUST be some "OK settings" where ppl. settle for. I am (not) sorry that i asked.
Last edited by Dr3Tri on Thu Nov 15, 2012 2:09 pm, edited 3 times in total.
User avatar
UNZ
Posts: 808
Joined: Mon Nov 21, 2011 9:42 pm
Contact:

Re: Hard disk recorder modes, int32 vs. float32?

Post by UNZ »

Dr3Tri wrote:So, there just MUST be some "OK settings" where ppl. settle for. I am not sorry that i asked.
yes, ANY setting is good, it just doesn't matter at all... (that's my point, and if you don't like it, just use ANY OTHER setting).

but don't take my word for it, try it: try a few options (that look the most different to you) and then try to hear a difference. Do a null test and try to see a difference..
Per Lichtman
Posts: 16
Joined: Wed Jun 13, 2012 2:59 am

Re: Hard disk recorder modes, int32 vs. float32?

Post by Per Lichtman »

@De3Tri I wasn't subscribed to the thread so I didn't see your post.

So what did you end up finding out over the last month in your own tests?
Per Lichtman
Posts: 16
Joined: Wed Jun 13, 2012 2:59 am

Re: Hard disk recorder modes, int32 vs. float32?

Post by Per Lichtman »

cyanphase wrote:I don't have much to add about dithering, I just needed to get my post count up so I can PM someone. :mrgreen:
Myself I like to keep things as 32-bit floats or higher when its possible.
You needed to up your post count? Seems weird when we've all been using your machines for what seems like a decade. :)
User avatar
mantratronic
Posts: 296
Joined: Mon Nov 21, 2011 7:23 pm

Re: Hard disk recorder modes, int32 vs. float32?

Post by mantratronic »

Per Lichtman wrote:You needed to up your post count? Seems weird when we've all been using your machines for what seems like a decade. :)
It's a conspiracy to get cyan to make more machines ;) Though wasnt there some rumour he had one ready that he'd release if we all annoyed him enough? :twisted:
Per Lichtman
Posts: 16
Joined: Wed Jun 13, 2012 2:59 am

Re: Hard disk recorder modes, int32 vs. float32?

Post by Per Lichtman »

I didn't hear about that rumour...

(Gets out poking stick).
;)
User avatar
mantratronic
Posts: 296
Joined: Mon Nov 21, 2011 7:23 pm

Re: Hard disk recorder modes, int32 vs. float32?

Post by mantratronic »

<cyanph[> oh joachip, if you'd like to update buzzmachines.com with a new machine theres an updated basstrk on my twitter, i noticed the massive pack had a basstrk from 2001 while the last updates were from 2005
<mtronic> lol, u mean my made up rumour was true?
<cyanph[> mtronic: yep
:D good work poking PL!
Post Reply