Jump to content

Zoom F6 (a 32bit recorder!)


IronFilm

Recommended Posts

2 hours ago, Mattias Larsen said:

Can NLEs export 32 bit aafs/omf/xml? Can all of them import 32bit even? 

 

I edit primarily in Premier Pro, and it does support 32-bit floating point. I believe it's the only NLE that does. I do not think / know if Avid Media Composer or Final Cut Pro X support anything higher than 24-bits integer. If they do not support the higher bit depth (or sampling rate for that matter), the NLE would conform the file to the sequence settings (usually 48k / 24-bits), which would be fine. With proper gain staging, a 32-bit float file would look and sound just fine in any sequence. The issue would mostly be if the NLE conforms a 32-bit float file that's been gain staged too low. Post sound would have to replace all the sound from the EDLs with the original production sound files to be able to take advantage of the 32-bit float and add as much gain as needed.

 

4 hours ago, JayKay said:

32bit float on the other hand allows the signal to go above 0dBFS and can still capture it just fine.

You could then turn down the audio in post and save all the audio above 0dBFS with the 32bit float file, which wouldn't be possible with 24bit file.

 

0dBFS represents the maximum signal level in the digital domain, regardless of how many bits the file is recorded with.

Link to comment
Share on other sites

6 hours ago, Olle Sjostrom said:

Still, what does it sound like? And also, file sizes? 

Nobody really knows yet, as nobody outside Zoom has got their hands on a demo unit to test. 
But the pre amps are the same as in the other Zoom F series, or in other words: they've very good. 

 

As for file size? 32bits is double the file size of a 16bit file, which is very very tiny. Unlike camera department recording raw, I don't think we've got anything serious to worry about here for data management of 32bit recordings. The biggest worry is post handling it, would never do this without getting the OK from post first as it is very nonstandard. 

Link to comment
Share on other sites

I just think the 32 bit thing is kind of pointless, seeing as people have been able to get good audio out of the zooms in the past and other brands, even on tape. I don't believe it will have a strong impact on our workflow. As of now it's just a selling point. 

However, it's fun to see that zoom are strong believers of "we'll fix it in post". Because if anything, 32 bit workflow and being able to not gain stage properly will only mean more work afterwards. 

Link to comment
Share on other sites

8 hours ago, Jose Frias said:

0dBFS represents the maximum signal level in the digital domain, regardless of how many bits the file is recorded with. 

 

That's true for when you are capturing audio. In the process of analog to digital conversion, 0dBFS is really the maximum signal level you can record without clipping. But once you have your signal in the digital domain and in 32bit float you can go above 0dBFS. This has to do with how 32bit float is encoding the data.

 

You can try it yourself: Take any sound file in your DAW. Your DAW has to work in 32 bit float. Then apply a lot of digital gain to your audio, so that it goes above 0dBFS. Export this sound file as 24bit int and also as 32bit float. Import these two files again and use digtal gain to lower their amplitude. You will see that the audio in the 24bit file is clipped at 0dBFS and with the 32bit float file you can recover the audio above 0dBFS just fine.

Link to comment
Share on other sites

7 hours ago, IronFilm said:

32bits is double the file size of a 16bit file,

 

Please, do the math properly. It's not hard.

32 bits / 8 equals 4 bytes.

With a 8 channel recording, and sample rate of 48000 samples per second, your data rate is

8 hannels * 48000 samples per second * 4 bytes per sample  = 1536000 bytes per second

1536000 / 1024  (to get from bytes to KiloBytes) = 1500

Divide again by 1024 to get from KiloBytes to MegaBytes, and you have 1.46 MB/sec

 

Then you have to add a tiny bit for header / metadata info, but (unless you have a Cantar) that is just a few KB.

 

Link to comment
Share on other sites

5 hours ago, JayKay said:

 

That's true for when you are capturing audio. In the process of analog to digital conversion, 0dBFS is really the maximum signal level you can record without clipping. But once you have your signal in the digital domain and in 32bit float you can go above 0dBFS. This has to do with how 32bit float is encoding the data.

 

You can try it yourself: Take any sound file in your DAW. Your DAW has to work in 32 bit float. Then apply a lot of digital gain to your audio, so that it goes above 0dBFS. Export this sound file as 24bit int and also as 32bit float. Import these two files again and use digtal gain to lower their amplitude. You will see that the audio in the 24bit file is clipped at 0dBFS and with the 32bit float file you can recover the audio above 0dBFS just fine.

 

This is correct, but we are talking about the 32-bit floating point in a >>recording<< application when it comes to the F6. Your original comment I quoted made it seem that you can still >>record<< above 0dBFS: "32bit float on the other hand allows the signal to go above 0dBFS and can still capture it just fine".

 

Unless Zoom is doing some magic I'm unaware of here, their double ADCs are still bound to a fixed point / integer math, and unless they calibrate 0dBFS to not represent the full scale of their ADCs, then 0dBFS does in fact represent the the maximum signal you can >>record<< before clipping.

 

 

Link to comment
Share on other sites

my understanding has always been that any bits get added to the bottom of the scale. unless something changed i would think that those extra 8bits get placed at the bottom of the scale not in the 0 dbfs or above the 0 dbfs. 

 

its really not like tape where you can bias the machine differently along with different formulations. 

 

i compare it to overxposing film/sensor. at some point, if its overexposed you lose all details and sich. the magic halpens when you add more shadow details. its the same thing here i feel like.

 

perhaps i’m wrong and 32 bit float goves you bits above 0dbfs. i never record at 32bit fixed or float. 24bits has great dynamic range as it is.

Link to comment
Share on other sites

18 hours ago, Jose Frias said:

Your original comment I quoted made it seem that you can still >>record<< above 0dBFS

I see, that wasn't my intention.

 

18 hours ago, Jose Frias said:

Unless Zoom is doing some magic I'm unaware of here, their double ADCs are still bound to a fixed point / integer math, and unless they calibrate 0dBFS to not represent the full scale of their ADCs, then 0dBFS does in fact represent the the maximum signal you can >>record<< before clipping.

 

I'm with you and think it is very likely that they will use dual 24bit ADCs and stitch their data together in a 32bit float file. Initially 0dBFS of this file would line up with the clipping point of the more insensitive ADC. At this point 0dBFS is of course still the maximum signal level you can record. It is only after the initial AD conversion that you could push values above 0dBFS with the digital gain. But this would then be recoverable because of the 32bit float.

Link to comment
Share on other sites

while I agree that the fix it in post mindset is often problematic, I can think of situations where recorder with dual ADC and 32bit float would be useful.

 

like for example I've done some acoustic concert filming/recording as one man crew for friends, so I couldn't adjust the trims all the time, so I had to record at rather low gain settings where the quiet parts were recorded on very low levels while I still hit the limiters for the peaks. 

 

with a properly designed dual ACD 32bit recorder, I really wouldn't have to worry about this at all since the high gain preamp/ADC would capture all the quiet parts and the low gain circuit would take capture the loud parts and all would be saved in a file format that makes it impossible to clip or distort, no matter what I set the recorder level at, so it's basically set and forget (and even the set part is reduced to setting the mics and not the levels).

of course there would be extra work involved in post, but even if I have my full attention on riding the trims during the performance, I'll still have to do additional levelling anyway and might even fight with the variable levels that are burnt into the recording.

 

 

I also agree that we can make perfectly fine recordings with the current gear, but personally I think the main reason that prevents 32bit float from being useful is just the post workflow/tools are not established yet because the idea of having a recorder which can basically capture the full output range of any mic without any clipping or distortion sure sounds intriguing to me.

chris

Link to comment
Share on other sites

I think what is left out of the discussion: It can record both the 24 bit and 32 bit SIMULTANEOUSLY.  So depending how it would be actually implemented in practice, I can image one could write in the metadata (of the 24bit?); something like "huge dynamic range in this take, check the 32 bit file if you feel you need it". So the sound editor would just take the 24 bit workflow/files/whatever for editing, and incase for that ONE take/scene that he needs it, he can use the 32 bit to get the (otherwise unrecoverable?) info.  

Link to comment
Share on other sites

8 hours ago, JayKay said:

It is only after the initial AD conversion that you could push values above 0dBFS with the digital gain. But this would then be recoverable because of the 32bit float.

 

We're obviously getting into semantics here, but at that point I think that calling it 0dBFS / full scale is not appropriate, but I see your point.

Link to comment
Share on other sites

Regarding dynamic range I'm still a little confused by the importance placed on digital levels and the ability to work with "overs" when writing audio files (processing is of course an entirely different matter) 

AFAIK 0 dBFS is not a defined level, it doesn't correspond to a signal voltage, it is just a digital value hopefully corresponding to the  analog limits of the system at conversion under some agreeable standard.

 

For example a properly designed AD converter for US broadcast will correspond to the required analog reference of 1.228 Volt / +4dBu = 0VU = -20dBFS. It then follows that 0 dBFS = +24dBu, so the analog maximum input level better be equal or greater than +24dBu, balanced <1% THD, otherwise there will be analog clipping before we hit 0dBFS.

(BTW, In reality this happens all the time because manufacturers like to fudge the numbers and claim max input levels without properly specifying THD, that's why many analog front ends sound distorted/ pinched, non-linear way before the 0dBFS value is reached)

 

Now compare that to a piece of gear that uses .775 Volt / 0dBu = 0VU = -20dBFS, it only has to have a max input level of +20dBu and will be cheaper to build.

If you use .316 Volts /-10dBv as analog reference and define a digital reference level of -14 dBFS then a max input level of around +6 dBu = 0dBFS will suffice, even cheaper. Slapping a 32 bit converter into the unit adds pennies to the cost vs the amounts it would cost to build a studio level device that meets all the criteria.

 

Note that only talking about the digital level expressed in dBFS and how they are expressed in bits tells us nothing about the analog capabilities / limitations of the system and those are the ones that will determine the signal to noise ratio way more than 24 vs 32 bits because while the analog reference levels are in flux as of late even on pro looking devices (seeing an XLR input no longer tells us anything about the reference used) the analog noise floor is still the same and if you operate at lower levels you're closer to the noise floor. I don't believe that the AD converter built around lower operating levels ( cheaper ) will rival the performance of the AD converter built to handle professional level audio signals when it comes to S/N, and even the best of those only achieve 21 bit of dynamic range.

 

Link to comment
Share on other sites

7 hours ago, Werner Althaus said:

only achieve 21 bit

 

To emphasize this (from an IT background), the difference between 21 and 24 is NOT a small difference, it's 700% !
(Compared, you can also say it's 8 times more.)

Each bit doubles the total number that can be described.

1 bit is 0 or 1, so 2 possible values

2 bits is 0 up to 3, so 4 possible values

3 bits can describe 8

4 bits can hold 16 values

 

Largest number in 21 bits is 2097150

compare that to 24 bits:       16777214

 

Classic! (Spinal tap)

 

 

Link to comment
Share on other sites

7 hours ago, Werner Althaus said:

and even the best of those only achieve 21 bit of dynamic range.

 

I recently stumbled upon this one by Linear/Analog Devices. They claim a "true" 32 bit at 145 db. I thought that "true" 32 bit was about ~190db by the way. 

https://www.analog.com/en/products/ltc2508-32.html

Link to comment
Share on other sites

25 minutes ago, Vincent R. said:

They claim a "true" 32 bit at 145 db. I thought that "true" 32 bit was about ~190db by the way. 

My calculator says 192.7 dB.

Or as a rough rule of thumb, at 6 dB per bit, is "roughly" 186 dB.

LEF

Link to comment
Share on other sites

57 minutes ago, LarryF said:

My calculator says 192.7 dB.

Or as a rough rule of thumb, at 6 dB per bit, is "roughly" 186 dB.

 

that's true for 32bit integer linear encoding since this is a very inefficient way to store linear data which basically needs logarithmic precision (you'd have 2 values for the lowest 6dB and 2147483648 values for the highest 6dB).

with other ways of 32bit encodings, like 32bit integer logarithmic encoding or 32bit floating point linear encoding we can get much larger dynamic range. (In the latter case I think it's over 1000 dB but I'm too lazy to research the math).

 

Obviously that's not much use for recording normal audio since we never have a range over 150db of dynamic range, but it also allows for much more precise encoding of very low levels, plus it would allow for having a much higher headroom over "reference audio level" - ie if we would want to have dialogue around -15dB on the digital scale, we can record screams at say +20dB on the digital scale and allow to compress that back below 0dB in post for distribution.

 

To get any real benefit of 32bit float recordings we absolutely need properly implemented high-end dual preamps and ADC (ie without distortions) and good algorithms for combining them. One could basically match one preamp and ADC circuit to the hottest possible mic signal, then have the second tuned to at a lower level and combine the two. (My understanding this is already done by Zaxcom and Sonosax).

And while you probably could record everything at -40dB on a S4+ to allow for enough headroom (like on a car bag drop), it certainly wouldn't make post very happy.

 

But I agree that as long as NLEs and post workflows still work in linear 24bit integer encodings, having 32bit float recordings probably creates more problems then it solves (I wouldn't be surprised if in 15years time this has changed though).

chris

Link to comment
Share on other sites

Can someone PLEASE tell me what this is all about.
You have UNCOMPRESSED, that means, 'AS IT IS'.

How much better do you want it?
As stated by others better than I am, there are no electronics better / exceed the space 24 bits can provide, so why bother?

 

Of course, in post, in a chain of effects where the first effect amplifies and the second one reduces, more bits might be of aid, but in recording, I totally fail to see the advantage.
 

You all record raw now. This is what comes out of your mics. Period. Higher sample rates I can understand (Read, high frequencies are not sines but become blockwaves, that are made back into sines due to the mass/responsiveness of the speakers.)

Please, someone, record the same souce twice in 24 bits, once normal, once 30 dB lower, and amp the second one by 30 dB in post, and do a phase reverse test. Then tell me I'm stupid.

 

 

Link to comment
Share on other sites

40 minutes ago, Bouke said:

Please, someone, record the same souce twice in 24 bits, once normal, once 30 dB lower, and amp the second one by 30 dB in post, and do a phase reverse test.

 

probably not much (if any) difference... but as mentioned:

a) post will not be happy if I hand them files where everything is recorded at -40db (assuming you mean normal at -10dB) and

b) recording at -40db doesn't seem to be a good choice when recording a jazz concert where one minute the main instrument is a solo flute at very low levels and the next minute you have the drum kit and trumpet at full blast.

 

admittedly not a everyday scenario for a production sound mixer, but personally I'd love a really good dual preamp/ADC recorder that save to float for those occasions because if done properly*, you simply don't have to worry about level anymore at all. 

chris

 

Edit: I'd be very curious if somebody from Sonosax could add their view on this since they basically already seem to have half the system in place. 

 

*(probably not going to happen in a 500EUR device like the F6 discussed here though)

 

 

Link to comment
Share on other sites

Makes no sense at all to me.

The whole discussion is about having more bits to accomodate dynamic range, that has been proven a myth, unless some form of 'dynamic compression' is used.
It's (not from ears, but from technical perspective) an 'emperors new cloths' thing.

 

Now:

I AM post! What do you think, that I want overmodulated files?

For your jazz, do you really think that if you record -40 the low parts will be bad? Theory says that it won't make a difference.
You all bitch about more dynamic range while you never ever use the space you have, and let the limiters kick in.
Makes no sense to me at all, unless you do fast turnaround stuff that has no post, but in that case 16 bits limited would be good enough for rock 'n roll.

 

Link to comment
Share on other sites

2 hours ago, chrismedr said:

One could basically match one preamp and ADC circuit to the hottest possible mic signal, then have the second tuned to at a lower level and combine the two. (My understanding this is already done by Zaxcom and Sonosax).

Lectro has done this since the first UM400 to squeeze 3 bits more dynamic range (18 dB) out of the then available low current dual channel ADC's. 

Best Regards,

Larry Fisher

 

 

Link to comment
Share on other sites

1 hour ago, Bouke said:

You all bitch about more dynamic range while you never ever use the space you have, and let the limiters kick in.

 

with all respect, I don't feel like I'm bitching (or even complaining) about the lack of dynamic range.

But believe me or not, I really do think that a recording system which captures the full range of a microphone output in all it's fidelity without having to worry about levels at all is intriguing (that's not to say that I believe for one second that it will make for better films or concert recordings).

 

1 hour ago, Bouke said:

For your jazz, do you really think that if you record -40 the low parts will be bad? Theory says that it won't make a difference.

 

so why don't we all record at -40dB all the time and forget about limiters completely?

I know I don't because a) post/producers would kill me, b) I'd essentially have a 16bit recording for most of the material and c) I feel that the sound quality is better if the gain is set to the proper level.

 

what strikes me as odd is that you're so dead set against the idea. To me mapping the real world sound values to basically infinite values without hard ceiling seems the much more logical way to deal with digital audio recording and processing then having to constantly turn a knob and adjust to an arbitrary defined point (which will seriously compromise my recording if I accidentally mess up).

 

I won't loose any sleep if floating point recording never becomes practical, but I woundn't be surprised if it's the standard in a few years time.

chris 

Link to comment
Share on other sites

12 hours ago, Werner Althaus said:

Regarding dynamic range I'm still a little confused by the importance placed on digital levels and the ability to work with "overs" when writing audio files (processing is of course an entirely different matter) 

AFAIK 0 dBFS is not a defined level, it doesn't correspond to a signal voltage, it is just a digital value hopefully corresponding to the  analog limits of the system at conversion under some agreeable standard.

 

For example a properly designed AD converter for US broadcast will correspond to the required analog reference of 1.228 Volt / +4dBu = 0VU = -20dBFS. It then follows that 0 dBFS = +24dBu, so the analog maximum input level better be equal or greater than +24dBu, balanced <1% THD, otherwise there will be analog clipping before we hit 0dBFS.

(BTW, In reality this happens all the time because manufacturers like to fudge the numbers and claim max input levels without properly specifying THD, that's why many analog front ends sound distorted/ pinched, non-linear way before the 0dBFS value is reached)

 

Now compare that to a piece of gear that uses .775 Volt / 0dBu = 0VU = -20dBFS, it only has to have a max input level of +20dBu and will be cheaper to build.

If you use .316 Volts /-10dBv as analog reference and define a digital reference level of -14 dBFS then a max input level of around +6 dBu = 0dBFS will suffice, even cheaper. Slapping a 32 bit converter into the unit adds pennies to the cost vs the amounts it would cost to build a studio level device that meets all the criteria.

 

Note that only talking about the digital level expressed in dBFS and how they are expressed in bits tells us nothing about the analog capabilities / limitations of the system and those are the ones that will determine the signal to noise ratio way more than 24 vs 32 bits because while the analog reference levels are in flux as of late even on pro looking devices (seeing an XLR input no longer tells us anything about the reference used) the analog noise floor is still the same and if you operate at lower levels you're closer to the noise floor. I don't believe that the AD converter built around lower operating levels ( cheaper ) will rival the performance of the AD converter built to handle professional level audio signals when it comes to S/N, and even the best of those only achieve 21 bit of dynamic range.

 

While this is all true for traditionally designed recorders, it doesn't really apply to the F6. There are no preamps or analog circuitry before the converters which would saturate into nonlinearity. The mic signal directly feeds the converters.

Link to comment
Share on other sites

50 minutes ago, Patrick Farrell said:

There are no preamps or analog circuitry before the converters which would saturate into nonlinearity. The mic signal directly feeds the converters.

Oh wow. Why did i miss this critical piece of info. That's quite a major detail. Where you got this intel from?

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...