[also posted in General Discussion Forum / 64Float vs. 39Int]
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 wanted 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. From my point of view such an upgrade is also necessary to maintain Reaper's large lead in working with Higher Order Ambisonics.
64 bit files cannot be listened to by artists, business partners, customers in the end. They need files for their audio players (e.g. VLC) which match exactly what better DACs are able to play.
32 Bit uncompressed is just a silly format, as it can be losslessly compressed by a factor of 3 (e.g.Wavepack). I don't know if WAV can be coded that way.
I tested that VCL can play 32 Bit integer Wavepack files rendered by Reaper. It also can play 64 bit Floating Point WAV files. It also can play 32 bit Floating Point WAV files (potentially less accuracy than 32 but integer).
Can you provide a detailed list of 32 bit formats any usual player software (like VCL) can handle ?
32 Bit uncompressed is just a silly format, as it can be losslessly compressed by a factor of 3 (e.g.Wavepack). I don't know if WAV can be coded that way.
I tested that VCL can play 32 Bit integer Wavepack files rendered by Reaper. It also can play 64 bit Floating Point WAV files. It also can play 32 bit Floating Point WAV files (potentially less accuracy than 32 but integer).
Can you provide a detailed list of 32 bit formats any usual player software (like VCL) can handle ?
-Michael
I would love to use Wavepack files but as explained above and (from a different view) explained also in my reply to your post in the General Discussion Forum / 64Float vs. 39Int Reaper does not fill the 32 bit integer formats with integer information as one would usually expect.
When using 32 bit DACs 32 bit integer files have to be used because dithering to 32 bit beforehand is essential to pass the best quality to the listener. 64 bit files are played by VCL but then of course all the bits exceeding 32 bit are just truncated.
Reaper does not fill the 32 bit integer formats with integer information as one would usually expect.
I have no idea how you come to that conclusion.
Quote:
Originally Posted by kaikoshru
When using 32 bit DACs 32 bit integer files have to be used because dithering to 32 bit beforehand is essential to pass the best quality to the listener. 64 bit files are played by VCL but then of course all the bits exceeding 32 bit are just truncated.
Supposedly the least significant bit is rounded and not truncated, to be perfect. If the input resolution was 32 bit, all bits are preserved.
When rendering from 64 bit FP to 32 bit integer, no dithering is necessary, as 64 bit FP provides more than 32 bits resolution (some 48 bit AFAIK)
-Michael
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.
MH Acoustic's Eigenmike EM32 outputs 24 bit, 44,1 or 48 kHz over ADAT. The "32" in the name comes from the 32 mic capsules it contains.
I'd like to know which smartphone is capable of outputting 32 bit integer as I've never heard of any audio interface that's able to do so...
__________________
In a time of deceit telling the truth is a revolutionary act.
George Orwell
I'm very interested in finding out why in fact the use of 32 bit integer WAV file format (seemingny now provided for completeness) would yeals any improvement on the issue of feeding a 32 bit DAC (we have not yet been told the brand of such).
(Disregarding, that the whole endeavor is rather fruitless because we now learned that the input signal by the Eigenmike EM32 is only 24 bit, anyway.)
Well, the whole point of high sample rates and bit depths is in mixing. For recording, one can use a simple 16-bit/44KHz, but if there are plenty of channels, like drums or orchestra, for example, then, mixing the whole thing at 64-bit/192KHz does yield a big difference. Of course, even better if you can record at 64-bit/192KHz, but even if you can't, mixing processors (EQs, Reverbs...) will add their magic and expand the sound in all directions when mixing at 64-bit/192KHz.
When using 32 bit DACs 32 bit integer files have to be used because dithering to 32 bit beforehand is essential to pass the best quality to the listener. 64 bit files are played by VCL but then of course all the bits exceeding 32 bit are just truncated.
-Gernot
No, no. What you are truncating are just only zeros!
When recording you should take some facts into account:
Human hearing has a range of about 120 dB. That is the
range from "just audible" to "pain threshold".
The dynamic range possible with digital formats is:
8-bit ---> 48 dB
16-bit (CD) ---> 96 dB
24-bit ---> 144 dB
32-bit ---> 192 dB
From this we see immediately: 24-bit was always sufficient
for recordings! Even the 24-bit still have an overdynamic of
24 dB (~ 4 bit)!
32-bit is really something for esoterics, shamans and
conspiracist, because here CPU and memory are strained
just to process and store many zeros. So 32-bit adorer
indulge themselves to a kind of "zero religion".
Haha, if you now think about mastering - the final format
from which you draw mp3s or burn CDs - then you need even
fewer bits! Because the loudness-war usually compresses
the dynamics to about 30 dB or less. And then even 8-
bit/44.1 khz are completely sufficient! Nobody could hear a
difference.
The difference between "32-bit floating point" and
"32-bit integer" is huge: Binary numbers in floating point
have a “scientific notation”. That means they have a
23-bit mantissa, an 8-bit exponent and 1-bit plus-minus
indicator.
This leads to a dynamic from - 758 dB to + 770 dB.
Since the greatest difference in sound pressure on Earth
can be about 210 dB, from anechoic chamber to massive
shockwave, 1528 dB is far beyond what will ever be
required to represent acoustical sound amplitude in a
computer file.
32-bit integer, on the other hand, doesn't make sense.
As stated above, it has 192 dB dynamic, but requires the
same amount of harddisc space as 32-bit floating point.
So really - 32-bit integer is "esoteric".
If Reaper should offer this, then it is purely for the sake of
completeness.
The OP claims that other software is not able to decently drive 32 bit DACs from other file formats than 32 bit integer wav. While this supposedly is not true, it's not a Reaper issue whatsoever.
Well, there's stuff like the ESS9019. It claims "32 bit" and it's patented. Internally, it's a duo of 24 bitters, with interpolation. It's good. -120 dB S/R, 135 dB dynamic range.
But it's still a pair of 24 bitters, effectively yielding 21 bit each. And there's overlap between the two ranges. So it doesn't reach 32 real bits. How much overlap? ESS isn't clear, nor is the patent.
While the dual 24 bit at least gives an advantage while recording, I can't see what it would do on the output side.
If the OP hears a difference between float and integer, he needs to take a long hard look at his setup. Software can do funny things when converting. I wouldn't expect it to happen, but I've been amazed before.
__________________
In a time of deceit telling the truth is a revolutionary act.
George Orwell
Technically I suppose doing a more that 24 bit Audio DAC should not be as demanding as doing an appropriate ADC. But reaching 32 bit does not seem possible, at least not without cooling down the device to close to absolute zero to reduce the noise of the resistors.
Technically I suppose doing a more than 24 bit Audio DAC
should not be as demanding as doing an appropriate ADC.
But reaching 32 bit does not seem possible, at least not
without cooling down the device to close to absolute zero
to reduce the noise of the resistors.
Hayeah, but this is already true for 24-bit: Just to
realize the possible 144 dB, the analog preamps would
need such a cooling.
Without cooling, today's good preamps reach about 115 dB,
e.g. the Manley Slam. With cooling down to approx. 10 Kelvin
about 130 dB should be possible.
Hayeah, but this is already true for 24-bit: Just to
realize the possible 144 dB, the analog preamps would
need such a cooling.
Without cooling, today's good preamps reach about 115 dB,
e.g. the Manley Slam. With cooling down to approx. 10 Kelvin
about 130 dB should be possible.
Hence 32 bit in in hardware seems perfect nonsense .
A 32 bit ADDA in the engineering world is 32 bit. Simple enough.
What Sound Devices, Zoom and now Steinberg are doing, is adding 24 bits to 24 bits and calling it 32 bit. An old engineering technique to get better resolution. Only, in engineering nobody would dare to call it 32 bit, cause it isn't.
And, as others have explained, there are no advantages in playback. In recording, there is 1 major advantage to 32 bit FLOAT: you can amplify or attenuate after recording, without worrying too much about clipping or noise. A dream come true for people who record on the fly, like in news gathering, or nature recording.
32 bit integer doesn't have that advantage. But even plain old 24 bit has enough dynamic range, because the theoretical noise limit for a mic preamp is -132 dB. Current best practical figure is -118, IIRC. 24 bit already offers -144 dB. Way more.
There are optical mics, however. Only, outside of a few broadcasters in Korea and Japan, nobody is using those for music or voice...
In a way, it's about creating demand. Like a car that can do 224 mph. What's the use?
__________________
In a time of deceit telling the truth is a revolutionary act.
George Orwell
There is now hardware available that outputs 32-bit integer samples, however my understanding is that they generally use multiple 24-bit ADCs with different gain staging, switching between the ADCs depending on the input level.
So while the signal produced may be 32-bit, the number of significant bits used by those signals is probably still less than 24 (each sample or block of samples will have some zero bits at the top and/or noise bits at the bottom).
Because of this: 32-bit FP should also be able to correctly represent those signals, unless the ADCs used are capable of outputting more than 24 bits of actual signal.
Disclaimer: I haven't actually tested any of these devices.
(also: we added 32-bit integer just for completeness, I can't imagine a real case where you'd really want to use this and not 64FP).
thanks to Reaper, about audio myths and rendering problems
@Justin: thanks for integrating 32 bit integer in the list of rendering formats
@mschnell: wav usually is the format used for files sent to artists, customers and so on (one can just be sure they can listen to the files, other than with AIFF files)
32 bit integer DACs:
I am not an engineer but a tonmeister so I judge not from measurings or theories but from the sonic results. As mentioned in the first message or this thread I am working with advanced Higher Order Ambisonic mixes (Eigenmike em32 recordings, 3D audio). Content: classical orchestral recordings. My (32 bit integer) DAC has an ESS ES9018 chip.
For listening to any of these mixes at the end I use bit reduction (with Dither):
- bit reduction to 32 bit: standard configuration, very spatial sound
- bit reduction to 24 bit: the impression of signals really coming from front or rear is missing to a really high degree (so the listener is not surrounded/enveloped by the audio any more - this is not so much about direct sound sources, it is about Raumschall as explained in the next section)
- bit reduction to 32 bit (without dither or with other than optimal dither parameters): different (and less than optimal) reproduction of low frequency content.
There is no doubt from my side that my DAC indeed converts 32 bit integer.
It does not bother me if it uses 2 24 bit DAC internally as long as it works.
The Dynamic Range myth (@enroe) A:
Dynamic Range in the sense of what enroe is talking about is not what matters in listening to acoustical music.
In concert halls there are thousands or even millions of lowest level reflections adding up to what in German is called Raumschall (reflections and diffuse sound). From what I experience
- it is obvious that the listener in the concert can perceive them
- in stereo recordings (1-dimensional) lots of these lowest level sound signals do not matter very much as the human perceptual system is not able to differentiate them in an 1-dimensional audio reproduction (these signals are squeezed too much in this kind of reproduction)
- in true 3D audio reproduction the situation is similar to the situation in concerts halls and the lowest level sound signals contribute to the overall audio perception.
These lowest level sound signals can only be reproduced by audio with a resolution significantly higher than 24 bit integer. From my point of view in some years we will end up at 40 bit integer.
By the way: The Eigenmike audio signals currently are recorded with 24 bit integer. But these signals are adding up to a soundfield with much deeper a resolution as there are 32 of such signals (and even 64 in the future).
The Dynamic Range myth B:
One is told by many audio professionals that audio with a level lower than 120 or 140 dB does not make any sense as it is masked by the noise of the electronic components. I did an experiment which proved this is not true on the DAC side:
sine tones or music were attenuated in Reaper by 160 or even 180 dB and given to the output (my 32 bit integer DAC). Appropriately amplified in the analog domain the signals are still easily detectable within the noise when I listen to them.
Plugins for HOA transformations (mixing) - 32 vs. 64 bit FP:
Just for your information - HOA (Higher Order Ambisonics) sounds much (!!) better when mixed with 64 bit FP plugins!! Thanks to Reaper for providing a 64 bit FP platform since many years (also Plugins in Reaper are all 64 bit FP)! Blame on almost all other Plugins which still are 32 bit FP.
Rendering problems in Reaper:
I know this is not the right thread for a discussion like this, what I am preparing is something like a bug report. This comes in advance (maybe some of you readers have similar experiences).
All renderings I am doing in Reaper sonically cannot measure up to what I am listening to via the output. This is even true for stereo recordings (96 kHz and mixed with mostly 64 bit plugins). Nulltests show that there is a difference between mixed sound and rendered sound in the region of around -100 dB.
It seems that Reaper is using different processes for audio output and renderings with the rendering processes being significantly inferior. Are they really working in 64 bit FP as do the mixing processes?
As Reaper still is the only platform genuinely suitable for HOA mixes this is highly regrettable as high end HOA mixes are seriously degraded by these renderings. Rumours are there will be an Eigenmike suitable for HOA 7th order as soon as November 2020. Hopefully Reaper will find ways to update the rendering processes soon.