Digital Performer audio engine resolution!

[Tesionman]

Hey ya'll!

I've noticed in the DP specs over at the Motu website that the audio engine resolution is 32 bit float.

I know some other DAWs are now completely 64 bit float, so my question is:
Is that really relative sound-wise?

Just curious to know.

Cheers and Happy New year to all!

[Tritonemusic]

I'm not sure I can answer that accurately for you. However, I do know that the MW Limiter processes audio at 64-bit floating point.

[HCMarkus]

Hey ya'll!

I've noticed in the DP specs over at the Motu website that the audio engine resolution is 32 bit float.

I know some other DAWs are now completely 64 bit float, so my question is:
Is that really relative sound-wise?

Just curious to know.

Cheers and Happy New year to all!

And a Happy New Year to you, too!

To answer you question: No. (imo, considering the resolution of human hearing, AD/DA conversion, microphones, preamps, musical instruments, the s/n of straight wire, and the many other variables present)

[Shooshie]

Well, ultimately the ideal is analog. 32 bit float is a pretty good imitation of analog, if there were a 32 bit floating point D/A converter. I've never seen one. So, 64 bit floating point processing is of course an even better rendition of analog, but it's a long way from the 24 bit D/A converters I have in my audio gear. I'm sure some specialty digital audio company makes such a thing, if it's possible, but who has heard it?

The best any digital audio will sound is the limit of A/D and D/A converters. Of course, the closer to analog in your processing will preserve certain nuance UNTIL it is converted to analog. But at 64 bit float, the discrepancy between achievable nuance and conversion back to analog is such a wide gap that it's debatable whether anything is really preserved at all. I'm sure there would be visible differences in the final wave of the output, but then you have to ask whether, with such a wide discrepancy, those differences are actual preservation of nuance, or new levels of aliasing that have yet to be solved for 64f/24 (or worse, 64f/16) conversion.

Shooshie

[Kubi]

How does 64bit float divvy up the bits? AFAIK, 32bit float is 24bit words with an 8bit "scaler". Resulting dynamic range is pretty spectacular.

So how is 64bit float put together?

[Shooshie]

Wikipedia has an article on double-precision floating point (64 bit fp) format, Here.

My question at this point is whether audio formats operate the same as other applications of digital data formats. If they do, then the following graphic illustrates the way each number is utilized:


The article also says that double precision format, which is essentially two single precision numbers processed as one, requires anywhere from 3 to 24 times as much processing time as single-precision numbers. One would have to assume there is a CPU tradeoff between resolution and speed, but that's a pretty significant hit. I'm sure that in a world where double precision resolution is the norm, Intel would develop double precision CPUs that burn through those numbers as if they weren't there.

Shooshie

[Michael Canavan]

The best example of why this is a non issue is my other DAW Live.
Ableton the makers of Live suffered through years of people attempting to say Live sounded duller and less clear etc. than other DAWs, and as a result of constant attention to the fact that Live had a 32 bit engine compared to Studio One, Pro Tools, Logic etc. they gave it a 64 bit engine. They, the company itself literally said it did nothing to the audio, that it was just placebo, and not much more.

When all is said and done we listen to audio at 44.1khz. 16 bit, but people want there to be some magical way that a DAW makes things sound better. IMO DP sounds better because it's set up for mixing and mastering in a logical and easy to use way, the rest is just hype, and a 64 bit engine on DP won't improve it just like it didn't improve Live.

[James Steele]

The best example of why this is a non issue is my other DAW Live.
Ableton the makers of Live suffered through years of people attempting to say Live sounded duller and less clear etc. than other DAWs, and as a result of constant attention to the fact that Live had a 32 bit engine compared to Studio One, Pro Tools, Logic etc. they gave it a 64 bit engine. They, the company itself literally said it did nothing to the audio, that it was just placebo, and not much more.

When all is said and done we listen to audio at 44.1khz. 16 bit, but people want there to be some magical way that a DAW makes things sound better. IMO DP sounds better because it's set up for mixing and mastering in a logical and easy to use way, the rest is just hype, and a 64 bit engine on DP won't improve it just like it didn't improve Live.

I'm still using a MOTU HD192 interface that I bought YEARS ago. It's capable of recording and playing back at 192k. Ask me how many times I've ran it in 192k mode? And I know this isn't the same thing, because that's about conversion, not the audio engine itself, but my point is that although it might sound better in ideal conditions, I'm not the least bit interested in it. It's geeky esoterica in the context of my needs. My HD192 happily chugs along at 24 bit/44.1k... sometimes 48k. Similarly, I'm not even curious about a 64-bit engine. When the public by and large listens to music as MP3s though cheap headphones, whether a DAW has a 32-bit or 64-bit audio engine really shouldn't be the deciding factor in which DAW one chooses to use. I'd get the one I work fastest in. For me that's DP.

[kurtl]

24 bit linear is overkill. Our analog ears are the limiting factor.
Maybe when we turn into cyborgs will it make any difference.

http://en.m.wikipedia.org/wiki/Dynamic_range

https://courses.physics.illinois.edu/ph ... _Lect5.pdf

[EMRR]

Audio can be recorded at 32 bit float. I have never encountered audio recorded at anything higher. This is about dynamic range, and 32 exceeds anything and everything useful to human perception as it relates to the extremes of theoretical noise floor and the threshold of pain/damage.

64 bit is about processing power. Separate issue. DP8 runs at 64 bit default.

[James Steele]

Right, but isn't this about the "audio engine?" I'm assuming that has to do with the resolution of the processing calculations.

[mikehalloran]

I know some other DAWs are now completely 64 bit float,

Really? Which?

[Shooshie]

Audio can be recorded at 32 bit float. I have never encountered audio recorded at anything higher. This is about dynamic range, and 32 exceeds anything and everything useful to human perception as it relates to the extremes of theoretical noise floor and the threshold of pain/damage.

64 bit is about processing power. Separate issue. DP8 runs at 64 bit default.

Word.

[Shooshie]

Right, but isn't this about the "audio engine?" I'm assuming that has to do with the resolution of the processing calculations.

There are a number of resolution issues that we tend to confuse in digital applications. Each one has to do with the number of digits available for very large or very small numbers, thus the bits, but each application is different. Here are some that I can think of:

1) addressing. Computers must be able to write and retrieve billions of snippets of information, and these are stored in sectors with a number identifying where it is. The more numbers you have, the more sectors you can create. Thus, a hard drive gets twice as many addresses for data with each extra bit of addressing that's available. With 64 bit addressing you have more possible addresses than there are atoms in our solar system!

2) CPUs - CPUs crunch numbers, and numbers can either be a certain length, or a certain range of values. Of course, each number has to have an address to keep track of it until it's needed, unless they all have the same length, in which case the CPU will know when it's reached the end of one and the next one begins. Each of these things is described in bits, as in 32 bit word-length, or 64 bit addressing, etc.

3) Floating Point Processors - They deal more with nuance: the translation of our analog world into a digital one. Thus, they are very much concerned with resolution, big and small. The floating point, or exponent, just moves the decimal place, making the rest of the number significant rather than a bunch of zeroes required to do the same thing. If you use most of your digits just getting to the ballpark, you have fewer digits to describe the park. Let the floating point put you in the ballpark, and then you can use all those other digits to describe what's there wayyyy more accurately.

4) Graphics: graphic processors have to do exactly what I just explained. The more digits they have to describe it, the more real it looks. The floating point not only enables more nuance, but faster processing of that nuance, thus the real-time total-immersion games like World of Warcraft.

Imagine that you're orbiting a planet, and you want to describe its surface. Without the floating point, you might use all your digits just getting through the atmosphere, leaving you only with enough digits to say "big plain, shallow lake, big plain, long hill, plain, mountain, plain, ocean." But with floating point, the first few digits get you down to the surface, and then you have enough digits to say "grain of sand, grain of sand, grain of sand, ½" rock, grain of sand..." Obviously, you're going to get a more realistic picture in the latter, though from a zoomed-out perspective the former still has its place. With FP you can have both, right next to each other.

5) Music: the important stuff along a wave is often very subtle. We have two axes: time and amplitude. To describe those we must first sense them and measure what we sense, and that is the job of a converter and its codec (the software behind the converter). Each axis has its own resolution. Time is resolved to 44.1K, 48K, 88.2K, 96K, etc. Amplitude is resolved to 16 bits, 24 bits, 32 bits fp, etc. It's arguable that beyond 32 bits fp, we humans could not perceive differences, since 64 bits fp resolves amplitude to a level that's far smaller than the molecules that are vibrating within the air, and even smaller than the atoms in a phonograph groove.

There are people who say that we cannot hear the difference between 16/44.1k and 24/48k, but they are wrong. Perhaps they cannot, if their hearing is damaged, but a person with good hearing can certainly tell the difference between the two. When I was first learning digital audio, I was astonished at the difference, and at the fact that we were all accepting 16/44/1k in CDs. It wasn't something I could put into words, but I could easily sense it. I guess you could say that 16/44.1 sounded harsh and unfocused. 24/48 sounded much more clear, luxurious and deep (words that don't have a lot of meaning for stuff like this unless we agree beforehand how we're going to use them). Now, I'm not sure I could still hear it, as tinnitus and age have certainly bitten huge chunks out of my hearing.

Anything that increases accuracy brings us that much closer to analog, but when you are resolving past the molecules of air and the atoms of a groove, you have to question the need for THAT much accuracy. That just slows down your computer, fills up your addressable space with redundant and insignificant data, and provides little more than bragging rights.

So, we have established that there is 64 bit addressing, which makes possible our 4 TB and even 4 petabyte hard drives; 128 bit (or whatever) CPUs which utilize that addressing and/or crunch bits of code that may be 64 or 128 (or whatever) bits long; and there are 32bitFP codecs which do their best to describe the nuance of an analog world through the measurement of geometry, lightwaves, or audio waves. There may be other applications that I'm not thinking of at the moment. They all work together in a digital world, but they are resolving different things.

Shooshie

[Shooshie]

One more thing: when it comes to describing the "bitness" of CPUs, I'm dealing with information I've read over 30 years, and rarely regurgitated in conversation, so I'm on pretty unsure footing without searching Google or Wikipedia (which I should have but didn't).

If that subject is interesting to anyone, here's an actual discussion about it. There are probably better ones. I didn't search very deeply. I think I was trying to establish the difference between register width and word length, both of which are measured in bits.

Shoosh