Skip to main content
Pro Audio Community

32 bit integer for interface, useful?

The new Steinberg interface supports 32 bit integer recording. I am curious if this can be expected to be a bit improvement over 24 bit. Im an the market for a higher end interface to transfer a ton of old tapes and records, as well as record and mix new stuff. I wonder if its worth holding off until there are more interface options, and higher dynamic range numbers, in the realm of 32bit tracking.

Im also wondering if im confusing what a 32bit integer interface means, and the audio isn't actually tracked at 32bit.

Comments

cyrano Tue, 07/23/2019 - 21:01

There's one true 32 bit audio interface around. It's been around for a decade or so. Hasn't taken the pro audio world by storm, so...

The one true advantage is that the input is line and mic level at the same time, due to it's immense dynamic range.

Of course, it's also VERY expensive. Around 15.000 € without any options.

I don't know if the Steinberg units are true 32 bit. There are a few other brands that have developed these too. Most are double 24 bitters, overlaying one over the other by level shifting the signal. Has been around for ages, even when only used in one high-end interface in the past.

So my guess is it's more of a marketing thing than real tech innovation.

kmetal Wed, 07/24/2019 - 17:07

cyrano, post: 461574, member: 51139 wrote: There's one true 32 bit audio interface around. It's been around for a decade or so. Hasn't taken the pro audio world by storm, so...

The one true advantage is that the input is line and mic level at the same time, due to it's immense dynamic range.

Of course, it's also VERY expensive. Around 15.000 € without any options.

I don't know if the Steinberg units are true 32 bit. There are a few other brands that have developed these too. Most are double 24 bitters, overlaying one over the other by level shifting the signal. Has been around for ages, even when only used in one high-end interface in the past.

So my guess is it's more of a marketing thing than real tech innovation.

Interesting stuff. Thanks!

kmetal Wed, 07/24/2019 - 18:20

Thats quite the dynamic range, 200db. I am curious about subjective quality as well, since the jump from 16 to 24 bit is noticeable.

I think reaper can track at 32bit.

It seems like alot of conversions going on under the hood and otherwise. Track at 24/96, mix at 32bit float/96k, mix down to 16 bit 44.1, or convert to mp3 or youtube ect.

Im starting to wonder if this is part if the reason standalone digital recorders and/or mixing via a digital mixer seems to have more solidity to the sound than itb in the daw.

bouldersound Wed, 07/24/2019 - 22:30

kmetal, post: 461584, member: 37533 wrote: Thats quite the dynamic range, 200db.

What exactly are you going to do with all that DR? LPs are one third of that. The best tape is less than half that. A good preamp approaches half that. You could run your peaks on the converter at -110dBFS and the analog noise floor would be above the digital noise floor.

kmetal, post: 461584, member: 37533 wrote: I am curious about subjective quality as well, since the jump from 16 to 24 bit is noticeable.

I'd like to see those double blind A/B/X test results.

kmetal Thu, 07/25/2019 - 07:16

bouldersound, post: 461586, member: 38959 wrote: What exactly are you going to do with all that DR?

Me, nothing. Is bit rate limited exclusively to DR? Im not exactly sure if the dynamic range specs relate to better, more accurate tracking of the (micro) dynamics of a ringing snare, or intonation. If it tracks the wave better or not. Is it analogous to throttle response on a car, or is it just a top speed rating. I cant do 200mph on any road practically, but a car that can, generally has a much more dynamic response at any speed, when you touch the throttle, in a very general sense, compared to a regular car.

Sounds like engineers need to improve the analog noise floor.

bouldersound, post: 461586, member: 38959 wrote: I'd like to see those double blind A/B/X test results.

I dont have the data. It is general consensus, but could be myth. I could regularly hear a difference in the 16bit mix down of a song, vs the multi track session in 24bit. Maybe the summing process comes into play. But its being summed in realtime by the master channel to output to the speakers. The mix downs sounded smaller, and flatter front to back than the multitrack. This is something i noticed, it was not pre-conceived. Maybe its just DP7 or maybe a hallucination of mine that occured often?

I was also of the minority that claimed to hear the difference between DAWs. Which Marco demonstrated in his video series years later.

Theres the lavry point of veiw that digital audio cant improve past 24/96. How long can that be argued for? Theres subtleties and timbre that specs cannot depict but exist in listening, and others that go unnoticed. Is 32bit or even 24bit one of them im not sure. When you start going down to 12bit or 8bit, things start getting obvious.

I expect dimishing returns since we use the logarithmic Db scale so often.

Boswell Thu, 07/25/2019 - 07:40

You have to be very careful how to interpret a manufacturer saying their interface is "32-bit". I know of very few audio interface / converter boxes that have an ADC that has usefully more than 24 bits of native conversion. However, a large number of audio interfaces incorporate some form of digital signal processing (DSP), and this is almost always either 32-bit floating point or integer arithmetic of 32-bit or greater.

Having been involved in high-resolution converter designs, I can say that the extra care and effort that needs to go into designing and manufacturing a unit that can achieve (say) 28 bits usually means that it can be very good at 24 bits, at least on paper.

bouldersound Thu, 07/25/2019 - 10:36

kmetal, post: 461590, member: 37533 wrote: Sounds like engineers need to improve the analog noise floor.

That would require amending the laws of physics. The noise floor of analog circuits is determined by thermal effects at the molecular level. The noise floor of an LP is determined by the material it's made of.

And human hearing has its limits as well. Overall you could probably hear 140dB of dynamic range, but you can't hear that all at once due to effects like masking. Exposure to higher volume sounds will temporarily reduce your sensitivity to lower volume sounds. That's why 16 bit is sufficient for a final product. Using 24 bit gives plenty of flexibility for adjustments that need to be made in production.

kmetal, post: 461590, member: 37533 wrote: I dont have the data. It is general consensus, but could be myth. I could regularly hear a difference in the 16bit mix down of a song, vs the multi track session in 24bit. Maybe the summing process comes into play. But its being summed in realtime by the master channel to output to the speakers. The mix downs sounded smaller, and flatter front to back than the multitrack. This is something i noticed, it was not pre-conceived. Maybe its just DP7 or maybe a hallucination of mine that occured often?

There's a lot going on there besides word length (bit depth). The real time processing may simply be different, and not necessarily better in a technical sense, than the mixdown processing. There could be errors that give it a character that you perceive as better. Technically better isn't always subjectively better.

kmetal, post: 461590, member: 37533 wrote: I was also of the minority that claimed to hear the difference between DAWs. Which Marco demonstrated in his video series years later.

I would have to see how that played out in a rigorous scientific test. I wouldn't want to say that you were just imagining it until I had solid data to support the claim.

kmetal, post: 461590, member: 37533 wrote: Theres the lavry point of veiw that digital audio cant improve past 24/96. How long can that be argued for? Theres subtleties and timbre that specs cannot depict but exist in listening, and others that go unnoticed. Is 32bit or even 24bit one of them im not sure. When you start going down to 12bit or 8bit, things start getting obvious.

I expect dimishing returns since we use the logarithmic Db scale so often.

I think the consensus of those who know what they're talking about trends toward 60kHz as the optimum sample frequency. I haven't read too much lately on the subject, but I haven't heard anyone suggest that more than 24 bit is desirable. Of course, going to 32 float inside the DAW does have its benefits.

cyrano Thu, 07/25/2019 - 15:13

bouldersound, post: 461583, member: 38959 wrote: What does 32 bit get you, almost 200dB of dynamic range? That seems like way more than way more than enough.

8 bit = 48 dB
16 bit = 96 dB
24 bit = 144 dB
32 bit = 192 dB

In theory...

Human hearing has around 100 db DR, at best. You'll often see 120 to even 140 dB, but that doesn't take masking into account. Masking means you can't hear a whisper against the background of a rock concert. Your hearing is capable of adjusting it's 100 dB DR up and down. So, if the background noise is 30 dB, your hearing will be more sensitive than when the background is 60 dB. That mechanism can take minutes to adjust, it's not instant.

And these numbers can be greatly enhanced by dithering. Imagine a stepped pyramid. Poring some sand on the steps will make it look like a non-stepped pyramid. That's what dithering does. It smoothes the wave form by adding noise.

That's why 16 bit is more than enough for reproduction...

The only inherent advantage of 32 bit is that gain settings matter far less. It doesn't matter if you're recording a choir and a jet fighter blasts over. The jet will be recorded perfectly...

kmetal Thu, 07/25/2019 - 15:54

bouldersound, post: 461599, member: 38959 wrote: That would require amending the laws of physics. The noise floor of analog circuits is determined by thermal effects at the molecular level. The noise floor of an LP is determined by the material it's made of.

So we have reached the pinnacle of analog equipments noise designs and cannot improve upon it?

bouldersound, post: 461599, member: 38959 wrote: I would have to see how that played out in a rigorous scientific test. I wouldn't want to say that you were just imagining it until I had solid data to support the claim.

Thats easy. No two audio engines are coded the same. How could they possibly sound identical? Identical projects with identical pan law settings, dont null daw to daw. It doesnt even require rigor to demonstrate.

bouldersound, post: 461599, member: 38959 wrote: I think the consensus of those who know what they're talking about trends toward 60kHz as the optimum sample frequency.

Ive read a paper a year or two from the maker of mytek that was supporting the notion that ultra high sample rates sound batter.

I didnt notice a difference when i compared years back on a low end converter, but that doesnt say much.

It would be tough to imagine we reached max digital audio quality in the mid nineties, and all technical improvements from there on out are irrelevant with regard to bit depth and sample rate. I guess it could be possible speaking of PCM, but i remain skeptical.

cyrano, post: 461604, member: 51139 wrote: That's why 16 bit is more than enough for reproduction...

Why is 24 bit the standard for tracking if 16bit is more than enough? Are we just wasting drive space? Im curious as opposed to defending one or the other.

Instead of dither smoothing things, could the steps between samples be reduced so it was smoother?

cyrano, post: 461604, member: 51139 wrote: The only inherent advantage of 32 bit is that gain settings matter far less.

Does this mean you can set a pre amp where its signal to noise is most optimal and potentially acheive superior results than on a lower bit depth?

So overall are the improvements (if any) contrained to the extremes and do not translate to the in-betweens or the useful area?

bouldersound Thu, 07/25/2019 - 16:29

kmetal, post: 461606, member: 37533 wrote: So we have reached the pinnacle of analog equipments noise designs and cannot improve upon it?

As far as noise, I think so. There are other areas where improvements can be made, but good modern gear is just about as good as it gets.

kmetal, post: 461606, member: 37533 wrote: Thats easy. No two audio engines are coded the same. How could they possibly sound identical? Identical projects with identical pan law settings, dont null daw to daw. It doesnt even require rigor to demonstrate.

I could accept that. But I don't think the math of summing is especially different from one program to the next. Pan laws are more complicated because it's not just the amount of difference between center and panned, it's the slope between them.

kmetal, post: 461606, member: 37533 wrote: Ive read a paper a year or two from the maker of mytek that was supporting the notion that ultra high sample rates sound batter.

I didnt notice a difference when i compared years back on a low end converter, but that doesnt say much.

It would be tough to imagine we reached max digital audio quality in the mid nineties, and all technical improvements from there on out are irrelevant with regard to bit depth and sample rate. I guess it could be possible speaking of PCM, but i remain skeptical.

Converter quality is more than just the word length and sample frequency. Oversampling and digital filtering made a huge difference. Before that it was necessary to apply the LPF in the analog domain. Digital filters can do things analog filters never could.

kmetal, post: 461606, member: 37533 wrote: Why is 24 bit the standard for tracking if 16bit is more than enough? Are we just wasting drive space? Im curious as opposed to defending one or the other.

A playback file is generally not going to be manipulated digitally, so it can be locked down so to speak. During the production process a digital file is going to be adjusted in various ways so you need some margin at the top and bottom to allow that manipulation to be done without hitting any limits.

kmetal, post: 461606, member: 37533 wrote: Instead of dither smoothing things, could the steps between samples be reduced so it was smoother?

It's unnecessary. Reconstruction filtering allows the ADC to reproduce the original waveform to a degree of precision that's more than adequate. Pretty much all PCM audio is 6dB per bit.

cyrano Thu, 07/25/2019 - 17:16

kmetal, post: 461606, member: 37533 wrote: Why is 24 bit the standard for tracking if 16bit is more than enough? Are we just wasting drive space? Im curious as opposed to defending one or the other.

I wrote "for reproduction, as in playback...

Once you start working with the recorded audio, 24 bit will yield less errors and if you render out to 16 bit, it will even suppress more artefacts.

Instead of dither smoothing things, could the steps between samples be reduced so it was smoother?

There's no need, as going to analog will smooth out the steps. Dithering just makes them a little bit smoother, if applied with some knowledge about the process. That's why there are usually just a few presets for dithering.

Does this mean you can set a pre amp where its signal to noise is most optimal and potentially acheive superior results than on a lower bit depth?

Yes and no. It makes it easier, but it doesn't mean better results. Less need to ride the faders, if you will.

So overall are the improvements (if any) contrained to the extremes and do not translate to the in-betweens or the useful area?

I'm not sure what you mean. 32 bit could be useful for measurements, for instance, but it's usefulness for audio is very limited.

kmetal Fri, 07/26/2019 - 05:11

bouldersound, post: 461609, member: 38959 wrote: As far as noise, I think so. There are other areas where improvements can be made, but good modern gear is just about as good as it gets.

I could accept that. But I don't think the math of summing is especially different from one program to the next. Pan laws are more complicated because it's not just the amount of difference between center and panned, it's the slope between them.

Converter quality is more than just the word length and sample frequency. Oversampling and digital filtering made a huge difference. Before that it was necessary to apply the LPF in the analog domain. Digital filters can do things analog filters never could.

A playback file is generally not going to be manipulated digitally, so it can be locked down so to speak. During the production process a digital file is going to be adjusted in various ways so you need some margin at the top and bottom to allow that manipulation to be done without hitting any limits.

It's unnecessary. Reconstruction filtering allows the ADC to reproduce the original waveform to a degree of precision that's more than adequate. Pretty much all PCM audio is 6dB per bit.

Cool. Good stuff.

cyrano, post: 461610, member: 51139 wrote: I wrote "for reproduction, as in playback...

Once you start working with the recorded audio, 24 bit will yield less errors and if you render out to 16 bit, it will even suppress more artefacts.

There's no need, as going to analog will smooth out the steps. Dithering just makes them a little bit smoother, if applied with some knowledge about the process. That's why there are usually just a few presets for dithering.

Yes and no. It makes it easier, but it doesn't mean better results. Less need to ride the faders, if you will.

I'm not sure what you mean. 32 bit could be useful for measurements, for instance, but it's usefulness for audio is very limited.

Two things im still not clear on. Why is 24 bit not needed for reproduction vs 16 bit? Ie why is it not better.

To try and clarify a previous question (the last one you were unsure of what i was asking)

Is 32bit just a sort of min/max improvement, or does the increase in dynamic range help the intermediate levels and dynamics. Ie would it track a snare hit then a snare hit slightly louder, more accurately at higher bit rates. Or tiny vocal cord inflections and intonation?

bouldersound Fri, 07/26/2019 - 11:25

kmetal, post: 461613, member: 37533 wrote: Two things im still not clear on. Why is 24 bit not needed for reproduction vs 16 bit? Ie why is it not better.

The dynamic range of 16 bit audio is about equal to the effective dynamic range of human hearing. In other words, you can't hear the difference between a (properly mastered) 16 bit file and a 24 bit file. If there's a perceivable difference between the files there's some other cause than word length.

kmetal, post: 461613, member: 37533 wrote: To try and clarify a previous question (the last one you were unsure of what i was asking)

Is 32bit just a sort of min/max improvement, or does the increase in dynamic range help the intermediate levels and dynamics. Ie would it track a snare hit then a snare hit slightly louder, more accurately at higher bit rates. Or tiny vocal cord inflections and intonation?

As far as I know it's literally just noise floor. If your analog source noise floor is -100dB from the peak level (very optimistic) and your digital stream noise floor is -192dBFS, you could theoretically record with peaks as low as -92dBFS and not add any noise, for whatever that gets you. But with 24 bit you can track with peaks at -12dBFS and your analog source noise floor will be more than 30dB above the digital noise floor.

If you actually used all the dynamic range in 24 bit audio the levels would vary from practically inaudible to painful. In other words, if the peaks are at 0dBFS and you set your volume so the loud parts are loud but not painfully loud, the noise floor would be more than 40dB below the threshold of audibility.

Each bit still represents a 6dB change in volume. But combined with the sample rate and the filters it's quite capable of reconstructing any waveform in the audible range with mathematical precision. The various slopes between each sample and the next, taken together, contain the information needed to perform that reconstruction. Something that illustrates this is inter-sample peaking.

https://www.masteringthemix.com/blogs/learn/inter-sample-and-true-peak-metering

cyrano Fri, 07/26/2019 - 15:57

kmetal, post: 461613, member: 37533 wrote: Two things im still not clear on. Why is 24 bit not needed for reproduction vs 16 bit? Ie why is it not better.

As bouldersound already explained, no human can hear the difference.

To try and clarify a previous question (the last one you were unsure of what i was asking)

Is 32bit just a sort of min/max improvement, or does the increase in dynamic range help the intermediate levels and dynamics. Ie would it track a snare hit then a snare hit slightly louder, more accurately at higher bit rates. Or tiny vocal cord inflections and intonation?

It's just min/max, as you put it. To add precision, you need to go to higher sample rates. But these will create lots of empty space in the wav file, under ideal conditions, or inaudible noise, in real conditions. And that inaudible noise does affect plugins and other audio calculations, like metering.

Analog can't do 120 dB dynamic range. Humans can't do 100 dB. And your average consumer playback system can't even do 80 dB. So the 144 dB from a 24 bit ADC is already more than what we NEED

audiokid Sun, 07/28/2019 - 21:54

Always keep in mind, a well designed 16 bit sounds better than a cheapo 24 bit design. Lots to do with the analog counterparts.

I heard this the first time with Lavry converters. Be sure the converter works well with your DAW.

Example. Lavry sounds awesome but lacks how it interfaces with a DAW at times.