64Float vs. 39Int

[McFly]

Unclear how Reaper manages clean bit depth of a raw incoming 32-bit (integer) file. If I select 64FP in project settings (Media & Advanced) and set device setting as 32WAV file, and I input a 32-bit file, does Reaper capture to disk as a clean 32-bit file? Or does the 64FP engine reduce incoming 32-bits clean to some smaller file, say 24-bits or 27-bits clean?

Not concerned right now with processing (fades, pan, plug-ins, EQ, etc.). Just considering raw 32-bit-int recording and playback in Reaper. Is there a setting in Reaper (64F, 39I, etc.) that will capture 32-bit int cleanly, and deliver it cleanly? Or does the Reaper 64FP engine capture/deliver a 32-bit-int input at some bit length less than the original 32-int?

I'm pretty sure that any DAW running a 32FP audio path can only guarantee around 24-bits clean. But if there's a 64FP audio path (Reaper), maybe it can manage 32-bits clean? I don't know, I hope someone here does.

(yes, I have 32-bit-int files that need to be captured and delivered clean)

[EvilDragon]

What you're talking about is mixing engine. What you want to set is the recording format, which is on Media tab of Project Settings, on the bottom. And yes, you can set 32 or 64 bit FP here as well. Once you set that, that's what you get when recording - not 24 bits or anything less than what you select.

[McFly]

EvilDragon: Thanks.

Are you 100% certain that an incoming 32-bit int file (say WAV32, or FLAC32, or m4a-32) will remain bit-clean inside reaper? And remain 32 bit clean when exiting Reaper?

[EvilDragon]

Well, for starters, it cannot be 32-bit int, it can only be 32-bit floating point. That's what Reaper supports. PCM formats (integer) are 8-, 16- and 24- bit. Above that is all FP. Or, there's also 4- and 2-bit ADPCM.

[metal_priest]

if you only need to record them and then send them, be sure to record them at 32bit, then simply take the audio files from the folder where you store the recorded files and it's done.

if you need to process them someway and then render them, simply select 32bit as depth for the rendered files in the render window

edit: oh, -int i didn't read it. well, evil said it

[serr]

What hardware 32 bit AD converters were used to make genuine 32 bit raw recordings? Just curious.

I'd like to hear comments on the shootout with high end 24 bit AD converters that prompted this too!

[McFly]

EvilDragon and MetalPriest -- yes, 32-bit original integer files, not IEEE float format. On the file side, WAV, FLAC, m4a, and others can all deliver 32-bit integer data. On the h/w side, AVB, Ravenna, Dante, USB, HDMI, and others all transport 32-bit integer data.

Serr -- ADC's with greater than 24 real bit performance have been around for over 20 years. The Salzbrenner ADC does 26 real bits, and they've shipped over 100,000 channels to-date. Google it. In the coming years, we'll be seeing this same level of performance from microphones, power amps, etc.. I guess you'll just have to trust me on that :-)

Hence, my question remains: can Reaper capture and deliver 32-bit-int clean data?

[EvilDragon]

As above mentioned - 32-bit PCM/int is not supported by Reaper, at least it doesn't seem to be able to record in that format (it CAN open it!). So it'll likely get converted to 64-bit FP at mixdown, and 32-bit or 64-bit FP for rendering, if you set that to be your rendering format.

[Xenakios]

Originally Posted by EvilDragon

Well, for starters, it cannot be 32-bit int, it can only be 32-bit floating point. That's what Reaper supports. PCM formats (integer) are 8-, 16- and 24- bit. Above that is all FP. Or, there's also 4- and 2-bit ADPCM.

Reaper does appear to be able to import 32 bit integer/PCM WAV-files. It does not allow exporting them, though.

[DVDdoug]

There's no advantage to 32-bit integer... Once you "do anything", change the volume, mix, add an effect, your data is no longer "bit perfect" because you've intentionally changed it.

There is a huge advantage with floating-point because it's virtually unlimited headroom and is virtually impossible to clip unless something goes terribly wrong. It also makes DSP "easier" because you're often manipulating values greater than your 0dB reference.

32-bit integer is limited to 0dB and will clip like any other integer format. (The additional dynamic range is all on the low-side.)

It's my understanding that there are no DACs or ADCs that are accurate to 24-bits, so good luck with 32-bits.

"CD quality" (16/44.1) is already better than human hearing,* unless you have very-low digital levels and you then amplify to the point where you can hear the quantization noise. And, that's optimistically assuming that the analog noise is below the quantization noise... In the real world, usable resolution is usually limited by analog noise.




* If you don't believe that, you've probably never done any proper scientific-blind listening tests. ...It can be kind-of surprising how those "night and day" differences become difficult to hear once you start blind listening.

[Justin]

Benefits of 32-bit integer aside: 64FP can exactly represent all 53 bit integers (which includes all 32 bit integers).

[Xenakios]

Originally Posted by DVDdoug

data is no longer "bit perfect"

The original poster may have some esoteric reason to keep the data "bit perfect", but it's not clear if that actually is what he is after.

[EvilDragon]

Originally Posted by Justin

Benefits of 32-bit integer aside: 64FP can exactly represent all 53 bit integers (which includes all 32 bit integers).

Right! Because 53 bits are the base, and 11 bits are the mantissa?

However, increasing the HD space necessary for 64FP just in order to correctly represent 32-bit int, why not just support 32-bit int as recording format as well? Because of the clipping?

[McFly]

Justin - thanks. Will test at 64FP.

Xenakios & DVDDoug - consider metadata stored in bits 25-32, that is, critical "non audio" data which is time-coherent with the audio in bits 1-24. It all needs to remain "bit accurate." Not looking to "process" -- just store and retrive. In our application, once the data is processed, all bets are off.

DVDDoug - We've had ADCs accurate to 26-bits for 20 years (Google Salzbrenner TrueMatch). Note also that "0dB" is arbitrary, dependent upon one's chosen digital reference. I've seen 0dBFS referenced as high as +28dBu, and as low as +4dBu. 16-bits is actually not even close to the limits of safe human hearing. Our ears can detect 0dB SPL (hence, 0dB is defined as the threshold of human hearing), and our upper "safe" hearing limit is around +150dB SPL, but only at low frequencies (think boom cars which routinely generate +155dB SPL inside the car, but only at VLFs). Our ears are specified to roughly 160dB SPL in a limited freq range (27 bits), and 120dB SPL full range (20 bits) for shorter time periods. "If you don't believe that, you've probably never done any proper scientific-blind listening tests." The benefits of higher dynamic range audio capture, storage, and playback go far beyond "noise." But I'll leave those benefits as an exercise for you to discover.

[DVDdoug]

Our ears can detect 0dB SPL (hence, 0dB is defined as the threshold of human hearing),

I've never been in a 0dB SPL environment.

Our ears are specified to roughly 160dB SPL

Yeah? Maybe once, then your hearing is permanently damaged and you never hear anything near 0dB again. Maybe you'll never hear anything again! I've forgotten the exact number, but 160dB is somewhere around the point where air starts clipping in on the negative-side because you're pulling a full-vacuum.

Just listening to 100dB music gives you a temporary threshold shift, so you cant hear quiet sounds for awhile and you may have ringing you your ears. If your music covers such a wide dynamic range you're going to get at least some temporary hearing loss, and you aren't going to hear the quiet parts of the music.

...A couple of days ago the were testing the alarm system in our building at work. It didn't seem super-loud and nobody was complaining that their ears were hurting, but my ears were ringing when they shut it off.

The National Institute for Occupational Safety and Health (NIOSH) has recommended that all worker exposures to noise should be controlled below a level equivalent to 85 dBA for eight hours to minimize occupational noise induced hearing loss.

Xenakios & DVDDoug - consider metadata stored in bits 25-32, that is, critical "non audio" data which is time-coherent with the audio in bits 1-24.

I have no idea what you're talking about. MP3 tags, SMPTE time codes? And what's that got to do with audio quality?

[Bri1]

...A couple of days ago the were testing the alarm system in our building at work. It didn't seem super-loud and nobody was complaining that their ears were hurting, but my ears were ringing when they shut it off.

Yes-I think lots of people take noise levels as a given-pushing people further to deafness- standard hairdryers and household vacuums for eg can reach levels of around 90db ++!! that's right next to the ears in most cases!
Here's 1 hearing firm's (A.H.C) scale->

Painful:
150 dB = fireworks at 3 feet
140 dB = firearms, jet engine
130 dB = jackhammer
120 dB = jet plane takeoff, siren

Extremely Loud:
110 dB = maximum output of some MP3 players, model airplane, chain saw
106 dB = gas lawn mower, snow blower
100 dB = hand drill, pneumatic drill
90 dB = subway, passing motorcycle

Very Loud:
80–90 dB = hair-dryer, kitchen blender, food processor
70 dB = busy traffic, vacuum cleaner, alarm clock

Moderate:
60 dB = typical conversation, dishwasher, clothes dryer
50 dB = moderate rainfall
40 dB = quiet room

Faint:
30 dB = whisper, quiet library

"Sounds that are louder than 85 dB can cause permanent hearing loss.”

If anybody can hear less than -85db in any daw-then they are doing quite well!

[EvilDragon]

Originally Posted by McFly

Our ears are specified to roughly 160dB SPL in a limited freq range (27 bits)

Nonsense. Do that and you lose hearing permanently. Prolonged exposure to anything above 120 dB is always going to take a toll.

Are you crazy, perhaps? Or do you REALLY enjoy standing right next to an airplane engine taking off?


BTW. Google Johnson-Nyquist noise. Showcasing that anything more than 24 bits is really not necessary, as far as ADC/DACs go, because that noise is inherent in electronic components, and when at 1 kHz it sits -144 dB, which is, whaaaat? 24-bit dynamic range.

[Xenakios]

[serr]

Maybe look to the medical device industry if this isn't audio related and audio sampling techniques don't make the cut for whatever you're doing. There are uses for digitizing data with sample depth that goes beyond the needs of audio. Other scientific industry too.

With audio, a 24 bit sample size captures a very large dynamic range with high precision to define very small discreet volume steps throughout the range. But the analog stage circuitry in the AD converters is the big limiting factor here. Garbage in, garbage out, so if you don't deliver a pristine image of the analog signal to the digitizer chip, lost nuance is never digitized. That analog connection is 99% the equation and even lowly 16 bit SD formats don't have near the impact. (And of course the volume war levels and painful treble boosting common on CD versions of albums is done intentionally with portable listening devices in mind rather than this being due to a loss from any format limitation.) I'm one of those that can hear and appreciate 24 bit HD program but I'd be lying if I suggest that the format discussion was more important than room, monitor system, and quality of mix.

So if you are playing with digital data other than audio data with relevant precision that goes far into the decimal dust, I'd look to the medical or other scientific instrumentation industry rather a than boutique AD/DA converters aimed at audio use.

[McFly]

serr - brilliant. thank you. in fact the seismic and vibrational analysis industries have specialized equipment for these levels of resolution.

[karbomusic]

I'm one of those that can hear and appreciate 24 bit HD program but I'd be lying if I suggest that the format discussion was more important than room, monitor system, and quality of mix.

Except each time we ask to prove it someone avoids the blind testing.

[kaikoshru]

Why do we necessarily need 32 bit integer rendering?

The term dynamics in connection with bit depth covers only to a very small extent the aspect that extremely quiet (and for humans inaudible) signals could also be recorded. Of great importance, and of crucial importance in the foreseeable future, is a completely different aspect that is very often overlooked in the audio industry and which is perhaps better called resolution in this context: the representation of very quiet sound nuances.

Such nuances, e.g. a single very quiet room reflection, have as such a level that is far below -100 dB, perhaps even below -200 dB. At the same time as such a reflection, the sound signal also includes direct sound and many other, sometimes much louder room reflections, in which the individual weak room reflection is embedded, so the overall signal has the usual level of maybe -30, -15 or even -1 dB. In appropriate environments, the human ear is easily able to hear such very quiet nuances in a differentiated way; 3D hearing plays a decisive role here.

So, if we are talking about the digital audio signal again, we are talking about the resolution of the signal: how many intermediate stages are between -10 and -11 dB, for example. Here 32 bit integer is far superior to 32 bit FP. Maybe we will need even 40 bit integer in the future.


For which kind of Audio Recordings does it matter?

For stereo recordings that are played back in stereo format, 32 bit integer is almost not advantageous. But already for advanced upmixes from conventional to binaural stereo it can look quite different.

Recordings with the Eigenmike em32 contain high quality recorded 3D audio material. I mix them in Reaper using novel methods that really bring out the spatiality.

A 32-bit integer DAC is connected to the hardware output, of which there are many different devices in the high-end sector. Some smartphones are also capable of reproducing such signals. Reaper reproduces my mix perfectly with this DAC.

But I have an unsolvable huge problem when I want to render or bounce 32 bit integer audio files of my mixes in Reaper. Whatever I try, the files contain only 32 bit FP quality audio. Strangely enough, this also refers to 32 bit integer files (AIFF PCM and WavePack 32 bit integer)! Soundwise it means that an existing great spatiality (and also spatial staging in the sense of a relief) is largely lost.

32 bit FP may be suitable for the reproduction of 3D direct sound signals, but it does not even come close to reproducing 3D raumschall (the German word Raumschall has no equivalent in English, it includes reflections and diffuse sound).


What is needed in Reaper?

After rendering or bouncing, 32 bit integer formats need to contain exactly the 32 bit integer audio which is obviously present in Reaper and which is passed on to the corresponding hardware (DACs). The direct support of WAV 32 bit PCM (integer) would also be appreciated.

Dear Reaper developers: Please upgrade as soon as possible, otherwise I do not know how to deliver a product to my customers.

[mschnell]

Originally Posted by kaikoshru

32 bit integer audio which is obviously present in Reaper and which is passed on to the corresponding hardware (DACs). The direct support of WAV 32 bit PCM (integer) would also be appreciated.

AFAIK, all three is wrong.

Reaper internally uses floating point format, and there are close to no 32 bit (integer) DACs in use, And Reaper does support rendering to 32 BIT Wavepack (i.e. losslessly compressed integer PCM).

-Michael

[kaikoshru]

Originally Posted by mschnell

AFAIK, all three is wrong.

Reaper internally uses floating point format, and there are close to no 32 bit (integer) DACs in use, And Reaper does support rendering to 32 BIT Wavepack (i.e. losslessly compressed integer PCM).

-Michael

1. this is not the point or in question
2. better do some research, I bought mine (S.M.S.L.) 3 years ago and there are almost 100 now
3. you better read more carefully what I have written (Wavepack by Reaper does not deliver what it should do, I suspect the signal is transformed to 32 bit FP before rendering) - anyway often in the forum it is written (also in this thread!!) that Reaper simply is not doing any recording / rendering in 32 bit integer but only in 32 bit FP, and this is what I experienced in week-long trials

-Gernot

[mschnell]

Originally Posted by kaikoshru

1. this is not the point or in question

maybe

Originally Posted by kaikoshru

2. I bought mine (S.M.S.L.) 3 years ago and there are almost 100 now

I did not say 32 bit DACs don't exist. But close to nobody uses them as 24 bit is widely agreed to suffice for close to any audio application.

Originally Posted by kaikoshru

3.Wavepack by Reaper does not deliver what it should do, I suspect the signal is transformed to 32 bit FP before rendering

I suppose you are misled here.

Reaper internally uses 64 bit FP which in any case provides more accuracy than 32 bit Integer. Rendering from and to 32 bit integer PCM hence is lossless. Wavepack also does lossless compression of the 32 bit integer PCM, and easily can be losslessly expended to 32 bit PCM for your DAC. (In fact storing 32 bit PCM without doing a lossless compression algorithm is just silly waste of file space)

I.e 32 (Bit ADC) -> 32 Bit PCM -> Reaper -> 32 BIT Integer Wavepack -> 32 but PCM -> (32 bit DAC) will not twist a single bit (or degrade analog accuracy of 32 bit = ridiculous 192 dB)

-Michael

[kaikoshru]

Originally Posted by mschnell

maybe

I did not say 32 bit DACs don't exist. But close to nobody uses them as 24 bit is widely agreed to suffice for close to any audio application.

My post is about Higher Order Ambisonics (HOA) and Eigenmike recordings - maybe you never heard about this audio technology. What perhaps is widely agreed upon is not valid any more when this technology is used in a way which makes a spacial sensation similar to good concert halls possible. There is a terminus technicus for this kind of change: "disruption".

Originally Posted by mschnell

I suppose you are misled here.

Reaper internally uses 64 bit FP which in any case provides more accuracy than 32 bit Integer. Rendering from and to 32 bit integer PCM hence is lossless. Wavepack also does lossless compression of the 32 bit integer PCM, and easily can be losslessly expended to 32 bit PCM for your DAC. (In fact storing 32 bit PCM without doing a lossless compression algorithm is just silly waste of file space)



I.e 32 (Bit ADC) -> 32 Bit PCM -> Reaper -> 32 BIT Integer Wavepack -> 32 but PCM -> (32 bit DAC) will not twist a single bit (or degrade analog accuracy of 32 bit = ridiculous 192 dB)

-Michael

You write what you guess but not what you know and again you did not read carefully what I have written. I also expected the kind of behaviour you are describing but experienced Reaper to behave in the way I have described. That is the main reason why I have written this post.

-Gernot

[airon]

I'm sure if there is enough interest and good use cases, Reaper will be given the capability to record 32-bit integer or floating point samples.

I first stumbled on to 32-bit ADC in the Sound Devices recorders, but have never used it or researched the benefits.

My use cases for 32-bit files are for mixes in 32-bit floating point files which are then loudness normalized. That way I don't need to keep my look-ahead mastering limiter in my main mix session.

Iirc(have to check as it was a while ago), Harrison let users record 64-bit floating point files in their dubbing recorders from their MPCD5 console range. They use modular PC systems to expand channel capability and recording. Their entire signal chain is 64-bit floating point like Reaper.

So there are some use cases for high bit-depth recording for a minimum of truncation.

Visual effects companies have highly detailed floating point image formats too after all. They're a small but significant use case, just as 32-bit audio seems to be at the moment. If I was archiving very low level stems, I might opt for a 32-bit integer or even 64-bit float audio files too. It just hasn't come up for me yet. Storage is cheap.

[fakemaxwell]

Originally Posted by kaikoshru

My post is about Higher Order Ambisonics (HOA) and Eigenmike recordings - maybe you never heard about this audio technology. What perhaps is widely agreed upon is not valid any more when this technology is used in a way which makes a spacial sensation similar to good concert halls possible. There is a terminus technicus for this kind of change: "disruption".

Just looked, nothing in the Eigenmic documents say anything about 32-bit. Eigenstudio records in 24-bit. Google searching for HOA bit depth did not return anything about requiring 32-bit. So if you can point anybody to what you're talking about, that would be great.

[Justin]

If you need 32-bit PCM, you can use AIFF in this mode.

If you need .WAV -- use 64-bit FP, which is a complete superset of 32-bit PCM (64-bit FP can represent every 32-bit PCM sample exactly, in addition to many billions more).

We'll look at adding 32-bit PCM .WAV writing support, too.

[mschnell]

Originally Posted by kaikoshru

You write what you guess but not what you know

What exactly did I state erroneously ?

- Reaper uses 64 BIT FP
- 64 bit FP is more precise than 32 Bit integer
- Reaper can render to wavepack (and AIFF) 32 bit integer files
- 32 bit wavepack losslessly compressed contains exactly the same information as 32 bit PCM
- with 32 bit input and no processing done to modify the signal by software (including disabling dithering), passing the audio stream though Reaper results in not changing a single bit
- VCL can play 32 bit integer wavepack files

(BTW: I don't know what format is used by VCL to forward the audio stream to the computer OS and then to the hardware. this also might result in degradation.)

Thanks,
-Michael

[McFly]

Originally Posted by DVDdoug

I've never been in a 0dB SPL environment.

Sure you have. You enter a 0dB SPL environment every time you put on a good set of closed-back, over-ear headphones, which can attenuate room noise up to 30-35dB. In a quiet room, this exposes your ears to around -8dBSPL ambient noise, which is the threshold of human hearing at 4kHz.

Originally Posted by DVDdoug

I have no idea what you're talking about.

High SPL. On the other end of real-world human hearing, we have boom cars (etc) that routinely generate 150-160dB SPL at very low freqs (the world record is something like 185dB). Snare drums, trumpets, etc routinely develop 155dB SPL. Recording engineers stick high-SPL microphones on these things every day (Shure SM57, etc). These are real world acoustic sources that go into our recording chain.

Hence, -8dB SPL at the low side and +155dB SPL on the high side. The real world range of audio engineering is around 160dB SPL, or 27-bits. Most of us don't need this kind of performance, which is fine, but that doesn't change sonic reality.

The inevitable next paradigm shift in audio dynamic range will be devices (mics, ADCs, DACs, etc) that can perform at the 27-bit level. Typically, this will be done via multi-path topologies. Of course, we already achieve sufficient high-level headroom with legacy technology, hence all of the additional dynamic range must be achieved via lower-noise hardware and electro-mechanics. To achieve this, we will move from 1uV to 50nV broadband noise conventions (2000-ohm vs. 10-ohm equivalent circuits).

[McFly]

Originally Posted by EvilDragon

Nonsense. Do that and you lose hearing permanently. Prolonged exposure to anything above 120 dB is always going to take a toll. Are you crazy, perhaps?

See my reply to DVDdoug.

I'm not saying that everyone needs to record at 27-bits (162dB). I'm saying that 27-bits is the inevitable evolution of audio capture, process, and delivery. This conversation is simply asking the question: should we always be pushing for lower levels of noise (higher dynamic range) in our recording paths? The answer is, of course, yes. Q.E.D.

[mschnell]

Being the devil's advocate (against decent practical consideration):

A technical "x dB" signal might have "delta-like" peaks using a lot more bits, as "dB" (there are several definitions) use averaging over some time. (Any Mic will smooth out those peaks by shifting phases, though,m without changing the ear's perception.)

-Michaelk

[McFly]

Originally Posted by mschnell

A technical "x dB" signal might have "delta-like" peaks using a lot more bits

When specifying dynamic range, assume broadband, unweighted, RMS measurement at the low side and peak measurement + possible ISOs on the high side.

[mschnell]

BTW.:
Reaper supports reading and writing (close to: you need to set a "noise floor") losslessly compressed 32 Bit floating point files.

Regarding that 32 Bit FP A/D - D/A hardware seems to be taking over, this seems to be a very interesting topic.

[sekim]

Originally Posted by McFly

The inevitable next paradigm shift in audio dynamic range will be devices (mics, ADCs, DACs, etc) that can perform at the 27-bit level.

Meanwhile everyone is trying to figure out how to make their mixes sound good when they compress/limit to 3 bits total dynamic range once the "music" starts...and the really cutting edgers are shooting for 2 bits, lol.

[serr]

Originally Posted by sekim

Meanwhile everyone is trying to figure out how to make their mixes sound good when they compress/limit to 3 bits total dynamic range once the "music" starts...and the really cutting edgers are shooting for 2 bits, lol.

Like the way the kids are working around the max allowed levels on streaming services now?
"This max limit will end the volume wars."

"Hold my beer!"

The move where you brick wall limit to -7 LUFS and then lower the post limit level until the meter reads -14 LUFS again. Gets you 2 to 4db louder than 'proper' mastering would. Sounds fully like ass for it but it's louder!

[sekim]

Originally Posted by serr

Like the way the kids are working around the max allowed levels on streaming services now?
"This max limit will end the volume wars."

"Hold my beer!"

The move where you brick wall limit to -7 LUFS and then lower the post limit level until the meter reads -14 LUFS again. Gets you 2 to 4db louder than 'proper' mastering would. Sounds fully like ass for it but it's louder!

LOL!