Ignorance is bliss

[Rusty K]

Hello,

Well as I've stated in other recent posts, I seem to have reached a new plateau on the learning curve and am now trying to go back and remedy old mistakes.

One such mistake is having a session where many of the tracks were extracted from CD's (44100/16bit) yet having the session properties be (44100/32bit float). I now know that upsampling of the CD tracks to 32bit float is unproductive.

My question is what to do about the bitrate knowing that I still have to master? Can I resample the whole session (without dithering) or should I just start a new session at 16bit/44,100 and transfer all the tracks? Of course the tracks I've recorded would still be at 44,100/32bit. I'm not sure this can be fixed...

Thanks,
Rusty K

 

[chrisharris]

One such mistake is having a session where many of the tracks were extracted from CD's (44100/16bit) yet having the session properties be (44100/32bit float). I now know that upsampling of the CD tracks to 32bit float is unproductive.

Hey Russ...I don't have an answer to your question, but that part of your post intrigues me. I was under the impression that it DOES help to transfer the whole session to 32 bit.

We're not talking about turning an mp3 into a wave file (where information has already been lost and cannot be "put back" into the file just by turning it into a wave file). My impression has always been that if you're going to be editing individual files in CEP, that it works internally in 32 bit, no matter what bitrate your session is at. So if your files are 16 bit, it converts them first, then edits the file, then has to convert the file BACK to 16 bit. This extra 2 conversions for every edit seems like it would cause problems.

Bottom line is that from the time I started doing EVERYTHING in 32 bit right up to the point of burning a CD, my crap has sounded better. What am I missing?

 

[Rusty K]

chrisharris,

Well I started this session so long ago that I've forgotten just what I did in the first place. I do have some tracks from CD that I kept at 44,100/16bit (as stated in the edit view) but the ones I recorded into the session are at 44,100/32bit Float. Obviously I've got mixed bit rates in my session. What the program my be doing in the background I don't know but would like to...ha!

From other threads I've read on the subject of dithering/readying a file for CD, if one starts out at 16bit he should stay there. The upsampling of information that is not there in the first place and the incremental improvement in quality of processing at a higher bitrate is a wash when you add the digital noise/dither to downsample again for CD.

The argument makes sense to me, if I have it right, so..... if you don't have the option in session properties to choose the bitrate you want isn't that a drawback of the automatic conversion to 32bit float internal?

In the Option/Settings/Data Tab there is a check box for "automatically convert all files to 32bit float on opening". I probably had that box checked (default I think) when I started this session. I have since unchecked that box.

Rusty K

 

[Jim Y]

Actually, I don't think it can harm to use 32bit float from 16bit integer sources. The conversion is reasonably straightforward, unlike changing sample-rates.

The point is that subsequent tranforms are of better quality than they would if you stayed at the lower bit depth. The harm that can be done is probably because your CD source already is dithered, but you cannot un-dither so you're never going to get over that one.

If you're using CD material as-is and limiting to mixing with other sounds, then stay at 16bit, but if you're applying EQ or modifying in any such a destructive way, my common sense says convert to 32bit floats first. If all the souces were off CDA, then it might probably be a good idea to save without dither, if it's already been done on most of the source material.

 

[chrisharris]

If you're using CD material as-is and limiting to mixing with other sounds, then stay at 16bit, but if you're applying EQ or modifying in any such a destructive way, my common sense says convert to 32bit floats first. If all the souces were off CDA, then it might probably be a good idea to save without dither, if it's already been done on most of the source material.

Yeah, see THAT makes sense to me (if I understand what you're saying). I think what you're saying is that if you're not doing any destructive processing of the 16bit file, leave it alone. Mix with it, whatever. Otherwise, work in 32 bit.

Here's how my tiny brain understands the concept. WHEN I DO A BUNCHA' STUFF TO A 16BIT FILE WORKING WITH IT AS A 16BIT FILE, IT SOUNDS LIKE DOGSHIT EVENTUALLY, lol.

In fact, you've given me an idea. I'm going to take 2 identical 16bit files, convert one of them to 32 bit, and then run each one through about 10 processes and see if I can hear a difference in the final version. My experience tells me I will, but I'll post back.

 

[Rusty K]

Although I don't really know if my CD extraced tracks are dithered or not, I'm betting they were not. Just for argument let's say they were dithered. If they are upsampled to be processed then they will have to be down sampled again and then you have dither on top of dither...this is turning into gobble-d-gook. Are we splitting hairs here?

I was under the impression that dither was more critical for bitrate conversions not sample rate.

Rusty K

 

[Jim Y]

Rusty, this is making my brain hurt

Not long ago, I would have dismissed all this as irrelevent. Now though, since I got some proper monitors and applied some acoustic treatment, I now notice the quality of low-level detail, particularly on fade outs and reverb tails so this discussion is quite pertinent to my own interests.

As the point of dither is to randomise the lowest bit, that random nature ought to survive bitwise conversion. Sample rate conversion could (I think) alter this because a computer can only do "psuedo random" and the introduction of another rate could introduce a pitched nature into the existing dither noise. Re-sampling is, effectively, re-modulating, so there are possible frequency based artifacts to consider.
Then again, perhaps if you ensure that any exposed silences really are digital silence, then you will get away with applying another dose of dither. On audible material, re-dither is only randomising that which is already random and shouldn't be heard as anything new - no pitched artifacts added, but that's only theory from me.
Your practical test is the way to go.

 

[chrisharris]

Your practical test is the way to go.

I was doing the test, but I got bored after about 5 edits...and like an idiot, I was using a ton of effects for each test and...well, both the 16bit and the 32bit sounded like shyte. Then my "ADD" kicked into overdrive and I just quit altogether.

 

[Rusty K]

This is not an argument about 16bit vs 24bit. If I'm generating the wave files I will always use 24bit/48000 the best my soundcard can do. If however I'm extracting tracks from CD's the "die is cast" so to speak, 16bit is already established.

I believe that we can agree on one thing and that is....dither when down sampling bitrate. I'm not techical enough to have followed Jim Y's theory completely but dither adds noise. If I were preparing a track to send to a friend via CD I would record at 16bit/44,100 to avoid the dither issue all together but again for arguments sake if the CD tracks were dithered then to once again to add more noise to your tracks just to be able to process at a higher bit/sample rate seems counter productive. If the CD tracks weren't dithered (as in my case) then I could maybe go either way on the upsampling thing though others have said in forum that it's a wash.

I hope we can get some others to weigh in on this discussion. In fact I bet a similar discussion has occured on more than one occasion somewhere here at HomeRecording. This issue is important because I want my work to be the best quality I am able generate within the limitations of my gear. Frankly my ears are pretty good but sometimes I must rely on the proper digital techniques (as learned from those more experienced than myself) to process my tracks.

I would still like some clarification on my original issues like having mixed bit/sample rates within the same session and how critical the dither issue in reference to sample rates only. My impression is one would not dither if sample rate is all that's being down sampled, in fact Nero will burn a CD at 16bit/48000. I don't know if Nero converted the sample rate automatically or just ingnored the sample rate but I forgot to convert before burning and didn't realize it till after a successful burn.

Rusty K

 

[Jim Y]

Nero automatically converts to 16/44 stereo in a CD compilation. I do not know how its conversions and dither compare to anything else for quality.

Dithering is habitually applied for commercial CD, whatever the original source. The lowest bit is impossible to accurately reproduce (it must be 1 or 0, but what if it should be halfway?), so it is dithered in order that it is heard as noise, never a tone. A noise level in the file that should only be worth 1/2 of the least significant bit would, without dither, get a whole value which would in fact amplify it. Not only that, but it could get (say) 5x 1's followed by 5x 0's which gives a tone of 1/10th the sample rate.
Noise shaping is (I think) an attempt to shift the frequencies of the dither noise into bands that humans notice the least.

Sample-rate conversion is actually quite complicated. Simply changing the rate makes the file faster or slower. A rate change that preserves the time of the file (such as 48K to 44.1K for CD) requires a software algorithm that interpolates sample values from those that are either side - calculating new ones (if upsampling) or throwing away (if downsampling). Tricky. I think it would be impossible to describe with just words what this might do to an existing dither signal. For sure, it cannot be done without some distortion - how much depends on the method used

If CD were 24bit, dither would probably not be needed as few, if any, could hear a 1bit tone as it would be well below normal hearing.

 

[Jim Y]

Well, I seem to have killed this thread with my technobabble.

So, I'd advise a read of this PDF document...
http://www.izotope.com/products/audio/ozone/OzoneDitheringGuide.pdf

Or, if you invest in a copy of the book 'Using Audition' by Ron Dabbs (recommended), you'll find the pdf on it's examples CD.

Although the document is for the Ozone software, it's examples and text are mostly general. You can download some small(ish) .wavs of some examples, but unless you have very good equipment, you may not hear the low level detail involved.

In addition to the points I made (getting rid of low-level quantise noise with a 1bit signal), it explains very well how dither effectively increases the dynamic range together with the different flavors of noise shaping.

It does not go into what happens when using pre-dithered 16bit material, just reiterates the advise that you shouldn't dither twice.

However, I'll restate a point I tried to make earlier....
Converting the bit depth up from 16bit simply adds zeroed bits to the rightside (least significant) end of the number. The volume with respect to 0dbFS remains the same.
Converting the bit depth down to 16bit simply discards (truncates) the rightside (least significant) bits, the volume remains the same - and THEN you add dither.
Dither has nothing to do with sample-rate conversion - it is entirely aimed at masking the quantise noise of the lowest bit when the bit depth is truncated.

If you own a soundcard that only actually works at 48Khz such as a Creative Live! or Audigy - either get rid of it or stop worrying. Is there any reason to persevere with these things now that the E-mu 0404 is so cheap? Sorry Creative owners, but even torture won't change my mind on that issue.

So, it depends on what your edit processing did (eq, compression etc) what happens to the dither, bearing in mind it could have been "noise shaped" and predominate only in certain frequency bands. Have a look at a fade out of one of the CD tracks using the Frequency analyser - if the noise floor tips up at the HF end, it's probably shaped dither.

As to the question "is it worth working at 32bit float with 16bit in and 16bit out?" - I say YES. You are always going to throw bits (= quality) when editing. A higher editing resolution will definately limit the damage.
However, if all you're doing is sequencing CD audio for compilation purposes (there is no EQing, normalising or the like) and the only mixing is during crossfades, you might just as well stay in 16bit all the way, turn dither off and thus preserve everything 1to1.

If your material came from commercial CD, be assured they ARE dithered. Either it's an early digitisation from analog tape, where the A/D converter provided the dither; or with recent digital productions, it was a higher bit depth and dither was added last of all at the mastering stage. Pro Tools can't do better than 24bit integers, so with 32bit floats to work with, what worries have you got?

And don't forget, Dither adds signal. So ensure there is headroom in the file for it, that is, NEVER normalise to 0dbFS, -1db will do the trick and nobody will notice as the dither will raise it a bit. If you dither a 0dbFS file, you will (must!) cause some clipping.

You'll be pleased to know I have no more to say. I'm sorry it's technical, but - this is a technical business.

Cheers

 

[Rusty K]

Jim Y,

I must admit that I constantly have to balance spending time understanding the technical side with my need to get new songs written...so I trudge along as best I can. Thanks so much for your help. I've saved the pdf to fav's and will read it.

Ok since I'm 99% sure the CD tracks I'm using (only two tracks in the session)were not dithered, shouldn't I upsample to 32bit float in CE to match them to my recorded tracks , sample rates being the same at 44100. Seems like it should be ok but I don't want to loose all my session work. I'm not sure if I want to start over completely which would probably be the best thing to do.

One more technical question....can you think of any reason why my batch conversion function would not be working (it's not). Edit window conversions are ok.

Thanks again,
Rusty K

 

[Jim Y]

Yes, by all means upsample the bits - it can only help.

Can't help with batch issues, I put on a hair shirt and do everything the hard way. Actually, I do composition in Sonar (need midi) and only master and/or edit with Audition. 24bit recording in Sonar (32bit internal) and export 24/44.1 without dither then into Audition.

I tiptoe cautiously around people who can run batch edits and hope they don't bite me. It helps if you've made up your mind what to do, and I never can.


That pdf is very good, I think you'll enjoy it.