16-Bit Project 32-Bit DAW 24-Bit Plug-Ins...what am I hearing?

[mark4man]

I am preparing tracks for mastering; & the majority of the projects (including the capturing of the original source recordings) are 16-Bit (don't ask.) These 16-Bit tracks are mixed in a 32-Bit DAW (SONAR.)

 

Now...I had been under the impression that the totality of what I was hearing during playback was the internal precision of the DAW...32-Bits. So, I had been bouncing stems to 24-Bit files so as to capture as much of what I thought I was hearing as I possibly could...the extra ambience/atmosphere of the plug-in processing running in the 32-Bit environment.

 

But...I happened to notice that the 24-Bit bounces sounded "richer" than the 16-Bit project running in real-time. The 16-Bit project on the other hand sounded "livelier".

 

So, I picked a song that had a strong but simple rhythm section...piano, bass & drums...& performed an experiment. I applied my best EQ plug-ins to each of those three tracks; & tweaked 'till they sounded perfect. Then I bounced each of the three tracks as a 24-Bit file; & opened them in a new 24-Bit project, to perform an A/B. As I bounced back & forth, the 24-Bit files sounded a tad thicker; & the 16-Bit project had a little more sizzle. (I preferred the sound of the original 16-Bit project.)

 

So now I'm thinking that the reason the 24-Bit bounces sound warmer is because I'm hearing the residual DSP signal I am NOT hearing in the 16-Bit project...the plug-in ambience & atmosphere that occurs below the 16-Bit floor.

 

Therefore...even tho the internal processing of the DAW operates at a greater depth than that of the original 16-Bit source recordings...because I'm processing 16-Bit files...I am only hearing how the dynamic range of those 16-Bit files are affected by the 32-Bit DSP. I am not hearing 32-Bit files...I am hearing what that 32-Bit DSP makes the 16-Bit files sound like, right?

 

To prove this to myself, I bounced the 16-Bit, plug-in applied tracks, to 16-Bit files; & performed not only a listening test, but another null test.

 

In listening, they now sounded identical.

 

In the null test, I turned the headphone volume all the way up (don't try this at home, folks) & listened for a faint signal...any difference at all between the 16-Bit tracks playing back w/ plug-ins applied; & the bounced 16-Bit files. There was dead silence...a perfect null.

 

If I had been hearing any 32-Bit DSP in the 16-Bit project (beyond that which effects 16-Bits)...there would have been some faint difference in sound during the null test; & there wasn't.

 

Conclusion: If I'm correct in my assumption, I'm now going to bounce to 16-Bit so they sound identical...(the way I want them to sound.)

 

Agree on the theory? Disagree on the theory?

 

 

mark4man

 

 

BTW - As far as AI/DAW config...my AI's internal resolution is 24-Bit, which is reflected in SONAR's global settings. The Project's bit depth is set to 16.

[Kiwiburger]

If your source material is 16 bits, then no matter what you do, the audio can only ever have an effective bit depth of 16 bits.

 

24 or 32 - you are simply adding redundant zeros that add nothing.

 

However - as soon as you manipulate your audio in any way, the greater available resolution means that less damage occurs from rounding errors.

 

When you are doing your null test - both audio streams are in 32 bit floating anyway, so you aren't proving much.

 

Simply use as much resolution as you can afford, and only resample down to 16 bits after all mastering is done and you finally need to burn an audio CD.

 

Start tracking a 24 bits - all this stuff is academic if you are still dicking around with 16 bits.

[Kiwiburger]

Also, consider what damage any resampling is doing in your playback software. Would I be correct is assuming that if you can't track at 24 bits, you also can't playback at 24bits? Usually if you have a 24 bit soundcard, you can track at 24 bits, so i'm guessing you only have a 16 bit soundcard?

 

If that is the case - you are wasting your time even comparing them. You are forcing your playback software to do resampling on the fly, and chances are it's doing a bad job.

 

Resampling is a big potential for screwing your sound. I saw some charts the other day, that showed that most major DAWs make a hash of resampling. You can buy good resampling software that makes a better job.

 

Do yourself a huge favour and invest in some quality 24 bits A/D and D/A.

[mark4man]

kiwiburger...

 

What the heck are you talking about, man ?*?*?

 

I'm not resampling anything...I'm simply asking what resolution I'm actually hearing...as opposed to what's occurring inside the DAW.

 

I know all the typical platitudes concerning tracking at 24 & dither at the end & all that...that's not the question. I'm not dicking around with 16-Bit...my lead guitarist recorded his music at 16-Bit before he knew what he was doing; but the takes were so fabulous we decided to go with them anyway. Of course I track in 24-Bit now...this was old stuff I was asking about.

 

Listen to the question...what am I hearing?

 

mark4man

[Kiwiburger]

Well thanks for clearing that up ... so you do have 24 bit D/A.

 

Well what you are HEARING is audio waves - not binary bits at all.

 

Maybe you should rephrase the question - I have no idea what you are talking about.

[Kiwiburger]

 

I'm not resampling anything...

I think if you get to the stage where you realise that you are actually doing a lot of resampling, you might understand this better.

 

Specifically you are doing a lot of bit depth resampling. You may also be doing sample rate conversion, but you haven't mentioned sampling rates yet.

[mark4man]

Well, let's see...

 

The only thing I can think of that might be outside my realm of understanding would be that, since the DAW operates @ 32-Bits internally, I might need to dither my 16-Bit project when I bounce to a 16-Bit .wav, so as to grab as much of the extra precision as possible.

 

Is this what you mean by doing "a lot of resampling"?

 

mark4man

 

 

BTW - No real resampling going on...the projects are 44.1 & the bounces remain @ 44.1.

[Will Chen]

Just go with what sounds better to you. Forget all the technical stuff unless you want to get into mathematical representations of sound which often has little to do with what you actually hear.

It's the difference between painting and building the chemical compound to make the paint.

 

I've never heard anyone say "That song would be so good if only it were 24bit"....

[Jamplified]

 

Originally posted by wbcsound I've never heard anyone say "That song would be so good if only it were 24bit"....

 

 

 

Well said!

[alfonso]

 

Originally posted by Jamplified Well said!

 

 

Ok, but since we have tools it's better to use them in the best way.

 

A guitar recorded at 16 bits can sound incredibly well if the player and all the subsequent chain have been used well.

 

But if you want to add reverb or some delays or any other processing, you are generating new data, they will match the calculation depth of your daw, in SX you will end with a 32float file anyway, and that bitdepth will be a very nice thing in regard to dynamics and details.

 

The mentioned zeroes will not be zeroes anymore, because, just to make the easier example, the reverb tails will fill them with new data.

 

If you want to take advantage of what your system can offer to make things sound at best (different of what you can do when you play), you have to treat these data as they deserve, because if you ignore them and just truncate them in the final exported file, you will just have a worst result. If you're fine with it, ok! If you are working professionally you should care, also because you don't have to fatigue so much...

[Will Chen]

 

Originally posted by alfonso Ok, but since we have tools it's better to use them in the best way. A guitar recorded at 16 bits can sound incredibly well if the player and all the subsequent chain have been used well. But if you want to add reverb or some delays or any other processing, you are generating new data, they will match the calculation depth of your daw, in SX you will end with a 32float file anyway, and that bitdepth will be a very nice thing in regard to dynamics and details. The mentioned zeroes will not be zeroes anymore, because, just to make the easier example, the reverb tails will fill them with new data. If you want to take advantage of what your system can offer to make things sound at best (different of what you can do when you play), you have to treat these data as they deserve, because if you ignore them and just truncate them in the final exported file, you will just have a worst result. If you're fine with it, ok! If you are working professionally you should care, also because you don't have to fatigue so much...

 

 

I don't believe the "filling of zeros" is an accurate description of how audio is rendered in a DAW playback engine. That is the assumption that your raw signal is increasing in volume with processing to a point that greater bit depth is required. In reality, the reverb tail example doesn't need any additional bit depth beyond the bit depth of the sound naturally decaying. Any additional decay would be quieter than the actual sound at 0. That doesn't make any sense to me.

 

In order to fully hear the noise floor difference between 16 bit and 32 bit, you would need monitors with a signal to noise ratio of 192db. You won't find a monitoring system with a signal to noise ratio anywhere near that. Even if you subtract around 20db for an extremely quiet (I'm talking pro designed studio) room , you're still talking about 172db range.

 

The biggest problem with digital audio is not "warmth", it's the ability to make mathematical sound measurements at a precision greater than the human ear is capable of hearing and marketing the crap out of it.

 

All that being said...yes, I track at 24bit even if it is (perhaps) a placebo. However, the moment I worry over my songs emotional impact in a 16 vs 24 vs 32 bit render is the moment I will quit calling myself a musician and start calling myself a mathematician...

[alfonso]

 

Originally posted by wbcsound I don't believe the "filling of zeros" is an accurate description of how audio is rendered in a DAW playback engine. That is the assumption that your raw signal is increasing in volume with processing to a point that greater bit depth is required. ...................................................................................... However, the moment I worry over my songs emotional impact in a 16 vs 24 vs 32 bit render is the moment I will quit calling myself a musician and start calling myself a mathematician...

 

 

The "zeroes" that someone mentioned were used to say that if you take a 16 bit signal and you record or export it as a 24 or 32 bits file, you wont detect any difference and that the added bitdepth will be just a larger size that won't carry any new data, so "zeroes"

 

I took the "zeroes to be filled" as a simple way to express the concept that if you add some processing in the 24-32 bits format, those previous "zeroes" or new places in the number, existing but not carrying any additional information in regards to the 16bits file, will carry some new information generated by the new 24-32 bits processing.

 

When you add bitdepth you add dynamics in the sense of the lower levels, not in the sense of an higher volume, you add detail.

While that detail might be not so much relevant for your ears in terms of level, it's very important in terms of calculation precision in digital processing, because the effect of calculation errors is very much reduced at increased bitrates, and that affects very much the quality of the final result.

 

To be honest and direct, I think that if you want to enter in a discussion about bitdepth you are talking about math, not about music. You might want to be only bothered by the quality of your music, but if you bring the fruit of your inspiration to a mastering facility and spend the big bucks to have a good sound, you won't give a damn if the guy is a great musician or not, he is payed for his technical abilities and his knowledge of these math issues.

 

Coming into a discussion about bitdepth and saying that afterall it doesn't matter because music is what does matter...well is not helping the discussion and is not very respectful

for those that are possibly wrong or right, but interested on bitrates anyway.

[Will Chen]

 

Originally posted by alfonso The "zeroes" that someone mentioned were used to say that if you take a 16 bit signal and you record or export it as a 24 or 32 bits file, you wont detect any difference and that the added bitdepth will be just a larger size that won't carry any new data, so "zeroes" I took the "zeroes to be filled" as a simple way to express the concept that if you add some processing in the 24-32 bits format, those previous "zeroes" or new places in the number, existing but not carrying any additional information in regards to the 16bits file, will carry some new information generated by the new 24-32 bits processing.

 

 

If you'd like to talk math, I'll talk math. I've never coded any audio applications, however I'm assuming when you upsample bit depth from 16 to 24 (bits is bits) I do not believe you are leaving empty numbers. 0db is 0db so the 16 bit audio interpretation of null is resampled to the 24 bit null. However, the 16th bit "on" is resampled as the 24th bit "on. Basically min and max are converted to the new base number. This has the result of redundant data in the upsample in order to simulate the perecision difference between the bit depth rates. If it were done in some other fashion, 16 bit audio would be quieter, and that's just not the case. Therefore, a reverb tail can not be less than the null of the original sampled sound.

 

Now, surely we want the maximum bit depth for processing purposes for the greatest simulation of reality. However I would seriously like to put this to a blind listening test to see if anyone can hear the differrence between upsampled 16 to 24 bit reverb tail versus 16 bit native reverb tail.

 

Now, if source is 24 bit there is a clear noise floor difference in the specs in addition to a smoother ratio from null to max of which a small population will be able to hear the difference in a quiet listening environment with very dynamic music (not top 40).

 

 

Originally posted by alfonso When you add bitdepth you add dynamics in the sense of the lower levels, not in the sense of an higher volume, you add detail. While that detail might be not so much relevant for your ears in terms of level, it's very important in terms of calculation precision in digital processing, because the effect of calculation errors is very much reduced at increased bitrates, and that affects very much the quality of the final result. To be honest and direct, I think that if you want to enter in a discussion about bitdepth you are talking about math, not about music. You might want to be only bothered by the quality of your music, but if you bring the fruit of your inspiration to a mastering facility and spend the big bucks to have a good sound, you won't give a damn if the guy is a great musician or not, he is payed for his technical abilities and his knowledge of these math issues. Coming into a discussion about bitdepth and saying that afterall it doesn't matter because music is what does matter...well is not helping the discussion and is not very respectful for those that are possibly wrong or right, but interested on bitrates anyway.

 

 

My point being that this type of information is crutial when designing applications. However, to me those who use intanglible words like "warm" to describe the precision of 1's and 0's in reference to bit depth is just kind of silly and rather oxymoronic.

 

I'm all about trying to get the best quality possible, but I'm a big fan of listening with your eyes closed. Ignore all the marketing and bull{censored} and listen. Does it sound good? Well then all that other stuff just doesn't matter does it?

[alfonso]

 

Originally posted by wbcsound ..... I'm assuming when you upsample bit depth from 16 to 24 (bits is bits) I do not believe you are leaving empty numbers

 

 

That's not upsampling, the signal remains intact, I only make place for further decimals, if I rewrite the file in the new format.

 

 

Originally posted by wbcsound However, the 16th bit "on" is resampled as the 24th bit "on.

 

 

Wrong, you add bits (not resample) for the lower levels, the 16th bit remains the 16th bit.

 

 

Originally posted by wbcsound Therefore, a reverb tail can not be less than the null of the original sampled sound.

 

 

For what I objected before, this sentence doesn't make any sense. A 32 bit reverb tail could be represented until such low levels not available at 16 bits. More bits is less noise, that's more detail, not more volume.

 

 

Originally posted by wbcsound However, to me those who use intanglible words like "warm" to describe the precision of 1's and 0's in reference to bit depth is just kind of silly and rather oxymoronic.

 

 

I didn't do that, but "warm" is commonly used for an aesthetical evaluation of the result of an audio process, it has not so much to do with 1's and 0's as it refers to the analog sound and could maybe indicate the success in the digital emulation of an analog sound.

On the other side, the fact that deeper bitrates allow for better analog emulations provided the quality of the algos, that's common knowledge.

 

My 0.03

[Zooey]

Once you process (like with a plug-in) or sum the individual 16 bit tracks, you've changed them. The resulting mix could be better captured with a 24 bit rendering. But you don't gain anything by converting unsummed, unprocessed tracks to 24 bit if they were natively recorded in 16 bits.

[Will Chen]

 

Originally posted by alfonso That's not upsampling, the signal remains intact, I only make place for further decimals, if I rewrite the file in the new format. Wrong, you add bits (not resample) for the lower levels, the 16th bit remains the 16th bit.

 

 

You're probably right here. I was thinking more of how one might represent white versus black in a 16 versus 24 bit image in which upsampling would be more comparible to changing the sample rate instead of bit depth.

 

 

Originally posted by alfonso For what I objected before, this sentence doesn't make any sense. A 32 bit reverb tail could be represented until such low levels not available at 16 bits. More bits is less noise, that's more detail, not more volume.

 

 

Still believe this statement to be incorrect within the constaints of the original poster's situation. When you rewrite the 16 bit data as 24 bit data the noise floor of the 16 bit data remains (approx -96). Based on your above explanation of "zero padding" which would be the additional dynamic range afforded by greater bit depth the 16 bit "noise" remains as the data has simply been null padded. Therefore, recording in 16 bit and processing in a greater bit depth will still suffer from the signal to noise ratio of the original 16 bit signal. Hence nothing can be quieter than 16 bit "silence" as it will be masked by the noise floor of the original bit rate conversion.

 

Burn CD of silence and listen to it. 24 bit audio offers an additional 48 db of range below what you're hearing. The 10 db or so below 96 will be able to be heard by few on the very few monitors which offer 100 to 110 dynamic range in an extremely quiet environment. But no modern monitors offer the full dynamic range of 24 bit data.

 

 

Originally posted by alfonso I didn't do that, but "warm" is commonly used for an aesthetical evaluation of the result of an audio process, it has not so much to do with 1's and 0's as it refers to the analog sound and could maybe indicate the success in the digital emulation of an analog sound. On the other side, the fact that deeper bitrates allow for better analog emulations provided the quality of the algos, that's common knowledge.   My 0.03