The Zoom F6 and 32 bit recording

[lkingston]

The other really cool thing with 32 bit float audio is how it imports into your audio editor. When it imports, it normalizes the audio levels automatically.

As was mentioned in the videos, there are two components to 32bit float audio: a 24 bit component that is like traditional 24 bit audio, and an eight bit component that represents the overall range. When you import the audio, that second eight bit component is automatically ranged so that the 24 bit component is normalized. This is incredibly cool as it is basically like recording 24 bit audio with the levels properly set each time.

There are scenarios where this is INCREDIBLY useful. Imagine you are recording a small acoustic group. You are one of the players. You mic up the players with six mics going into your Zoom F6 or Sound Devices Mix Pre 6 ii. You don’t set any record levels. You do set approximate monitor levels, but not exactly. You press record, and you all play the music. That’s it.

If you were recording at 24 bits, it would be an absolute pain. You would have to have everyone play and try to set each individual level. You would be going back and forth between playing and setting the levels, your own included.

With 32 bit float, it just works! Yeah, the files are way bigger, but they are still smaller than video files. I wouldn’t use it for multitracking audio with overdubs. For me, it is just for live music with all the tracks recorded at once, and where you can’t pay attention to setting the individual levels because you are also playing.

[keith.rogers]

Originally Posted by lkingston

The other really cool thing with 32 bit float audio is how it imports into your audio editor. When it imports, it normalizes the audio levels automatically.

As was mentioned in the videos, there are two components to 32bit float audio: a 24 bit component that is like traditional 24 bit audio, and an eight bit component that represents the overall range. When you import the audio, that second eight bit component is automatically ranged so that the 24 bit component is normalized. This is incredibly cool as it is basically like recording 24 bit audio with the levels properly set each time.

There are scenarios where this is INCREDIBLY useful. Imagine you are recording a small acoustic group. You are one of the players. You mic up the players with six mics going into your Zoom F6 or Sound Devices Mix Pre 6 ii. You donÂ’t set any record levels. You do set approximate monitor levels, but not exactly. You press record, and you all play the music. ThatÂ’s it.

If you were recording at 24 bits, it would be an absolute pain. You would have to have everyone play and try to set each individual level. You would be going back and forth between playing and setting the levels, your own included.

With 32 bit float, it just works! Yeah, the files are way bigger, but they are still smaller than video files. I wouldnÂ’t use it for multitracking audio with overdubs. For me, it is just for live music with all the tracks recorded at once, and where you canÂ’t pay attention to setting the individual levels because you are also playing.

Well.... I know you are a fan of 32-bit float, but I did a fair amount of live recording of groups, even open mics where there was no time to re-set gain/trim, and I can tell you that setting levels is not that impossible in all but the rarest of cases, and then you learn to knock a couple dB off the gain before you hit the red button!

So, my short answer (shortest I can do, anyway) is I would not buy a 32-bit-float capable recorder if it meant I had to give up the ability to set mic gain. For the F6, I would always choose dual mode (both float and 24-bit fixed). And, if you know the players and equipment, setting the gain once and keeping notes is going to solve a lot of problems. Plus, IMO, you'll have tracks that are quicker to use and mix.

Longer answer.... Where I see the problem (from another thread somewhere) is that to achieve the "headroom" (the vast majority of the 32bf dynamic range is, technically, "bottom-room"), the auto-gain is set somewhat arbitrarily low (on the F6). I mean, the designers have no idea whether it's an acoustic guitar a foot away or a trumpet with the mic stuffed in the bell, so "plan for the worst." This gets applied to all the tracks, and frankly, makes realtime monitoring a noisy mess. You're at the mercy of the channel *fader* and headphone amp to provide a level for you to really tell what is getting tracked, and, believe me, that is at least, if not more, important than knowing you cannot clip the signal. (The [new/updated] F8n Pro model still allows gain to be set individually on channels - probably an indicator that its primary use case is in audio recording for "film" production.) My $.02, anyway.

[lkingston]

Originally Posted by keith.rogers

So, my short answer (shortest I can do, anyway) is I would not buy a 32-bit-float capable recorder if it meant I had to give up the ability to set mic gain. For the F6, I would always choose dual mode (both float and 24-bit fixed). And, if you know the players and equipment, setting the gain once and keeping notes is going to solve a lot of problems. Plus, IMO, you'll have tracks that are quicker to use and mix.

Longer answer.... Where I see the problem (from another thread somewhere) is that to achieve the "headroom" (the vast majority of the 32bf dynamic range is, technically, "bottom-room"), the auto-gain is set somewhat arbitrarily low (on the F6). I mean, the designers have no idea whether it's an acoustic guitar a foot away or a trumpet with the mic stuffed in the bell, so "plan for the worst." This gets applied to all the tracks, and frankly, makes realtime monitoring a noisy mess. You're at the mercy of the channel *fader* and headphone amp to provide a level for you to really tell what is getting tracked, and, believe me, that is at least, if not more, important than knowing you cannot clip the signal. (The [new/updated] F8n Pro model still allows gain to be set individually on channels - probably an indicator that its primary use case is in audio recording for "film" production.) My $.02, anyway.

As you noted, the F6 and Sound Devices recorders let you work in either 24 bit or 32 bit float. You aren’t giving anything up.

In addition, the preamps in both the Zoom F series and the Sound Devices are far cleaner (lower background hiss) than the Zoom H series, so they are wonderful 24 bit recorders.

On top of that, the limiters (32 bit float digital look ahead with Zoom, high quality analog with Sound Devices) make recording at 24 a lot safer if you are using a minimal amount of safety headroom.

You really aren’t giving anything up with a Zoom F6 or Sound Devices Mix Pre. It’s everything you had before for pristine 24 bit recording, plus the new 32 bit float options.

[DupleMeter]

Originally Posted by lkingston

Here’s the thing: 24 bit sounds wonderful, but if you record it with too much headroom, you aren’t really getting 24 bits. You are getting 20 bits or some lower number. With 32 bits, you are getting the full 24 bits of resolution regardless of where you set your levels.

Not exactly. I think you failed to fully comprehend the tech behind it.

At 32-bit float you get virtually unlimited headroom (you can still very easily record less that full but depth), but at the expense of resolution (because dB are logarithmic). So each time your 32 bit recording needs to add headroom, your resolution (the ability to represent smaller changes in the signal) degrades. And the more headroom you add, the bigger the resolution gets with each extra bit that needs to be added.

This won’t be noticeable on loud music styles as quickly as it will in acoustic genres.

[rick-slo]

Originally Posted by DupleMeter

Not exactly. I think you failed to fully comprehend the tech behind it.

At 32-bit float you get virtually unlimited headroom (you can still very easily record less that full but depth), but at the expense of resolution (because dB are logarithmic). So each time your 32 bit recording needs to add headroom, your resolution (the ability to represent smaller changes in the signal) degrades. And the more headroom you add, the bigger the resolution gets with each extra bit that needs to be added.

This won’t be noticeable on loud music styles as quickly as it will in acoustic genres.

Interesting last couple of posts especially about the volume interval size for each decibel as input volume goes up or down beyond certain levels. Also in the recording stage the mikes and preamps of course have their usable dynamic range so I don't see to what extent this 32 float no need to set gains idea really works out.

[lkingston]

Originally Posted by DupleMeter

Not exactly. I think you failed to fully comprehend the tech behind it.

At 32-bit float you get virtually unlimited headroom (you can still very easily record less that full but depth), but at the expense of resolution (because dB are logarithmic). So each time your 32 bit recording needs to add headroom, your resolution (the ability to represent smaller changes in the signal) degrades. And the more headroom you add, the bigger the resolution gets with each extra bit that needs to be added.

This won’t be noticeable on loud music styles as quickly as it will in acoustic genres.

You have to understand that the mantissa is a full 24 bits. The “large jumps” in resolution you are referring to are actually measuring the same amounts of air movement you would be capturing at 24 bits after scaling the level with an analog preamp. Nothing is lost in this process that isn’t way below the audible volume range at that floating scale.

I am using this only on acoustic music with a wide dynamic range. That is actually its strength, not it’s weakness.

Loud music styles don’t have the wide dynamic range that would make 32 bit float important. I wouldn’t bother with 32 bit audio on any loud style of music.

[lkingston]

You could purposely make a recording where floating point would fall apart mathematically, but you would have to try:

Imagine a drumbeat that put light taps next to gunshots. 32 bit float wouldn’t be able to make those large jumps accurately. But as long as the loudest and softest part of the music you are recording are within a 24 bit range 144 db theoretically) you will be fine.

24 bit recording can’t do this either. You are either going to get clipping on the gunshots, or lose the taps in the noise floor.

Put another way: 32 bit float doesn’t have a particular advantage (or disadvantage) over 24 bit audio in precision. The precision of 32 bit float vs 24 bit audio with the levels set correctly is exactly the same: 24 bits.

Rather the advantage of 32 bit float is not having to know ahead of time where that full 24 bit dynamic range is going to be.

[jim1960]

Originally Posted by lkingston

Rather the advantage of 32 bit float is not having to know ahead of time where that full 24 bit dynamic range is going to be.

What piece of music requires that kind of dynamic range. I'm pretty sure I've never heard anything, other than perhaps a full movie soundtrack (and that might not even be true), that requires that kind of dynamic range.

[DupleMeter]

Originally Posted by lkingston

You have to understand that the mantissa is a full 24 bits. The “large jumps” in resolution you are referring to are actually measuring the same amounts of air movement you would be capturing at 24 bits after scaling the level with an analog preamp. Nothing is lost in this process that isn’t way below the audible volume range at that floating scale.

I am using this only on acoustic music with a wide dynamic range. That is actually its strength, not it’s weakness.

Loud music styles don’t have the wide dynamic range that would make 32 bit float important. I wouldn’t bother with 32 bit audio on any loud style of music.

No, that’s not right.

The scaling is logarithmic, and by nature gets more coarse as you add headroom. That’s the point. When you clip your converter you have the ability to scale the resolution so that 0db effectively becomes whatever the loudest sample you captured is. This comes at expense of resolution. The rescaling is logarithmic, so the farther past 0dB you went, the more coarse the change becomes. Much like a fader on a console. A movement of 1mm at unity is far less change than that same 1mm change at -10.

And loud music is exactly what 32-bit float is perfect for. You can rest easy knowing that you can recover from overs in the converter.

That said, the benefit only comes into play if you run past 0dB. Otherwise it is the exact resolution (and precision) as 24-bit integer (that’s where the difference between Integer & floating point shows up). Of course, this is all assuming you clip the converter before you clip the analogue path leading to the converter. In which case the bit depth means nothing, you simply have distortion. Albeit, analogue distortion might be ok, depending on the circuit involved.

Here’s an article I wrote for Production Expert that explains the logarithmic scale & how it relates to precision in audio. The article is about decibels, but it covers what I’m referring to: https://www.pro-tools-expert.com/pro...nding-decibels

[lkingston]

Originally Posted by DupleMeter

No, that’s not right.

The scaling is logarithmic, and by nature gets more coarse as you add headroom. That’s the point. When you clip your converter you have the ability to scale the resolution so that 0db effectively becomes whatever the loudest sample you captured is. This comes at expense of resolution. The rescaling is logarithmic, so the farther past 0dB you went, the more coarse the change becomes. Much like a fader on a console. A movement of 1mm at unity is far less change than that same 1mm change at -10.

And loud music is exactly what 32-bit float is perfect for. You can rest easy knowing that you can recover from overs in the converter.

That said, the benefit only comes into play if you run past 0dB. Otherwise it is the exact resolution (and precision) as 24-bit integer (that’s where the difference between Integer & floating point shows up). Of course, this is all assuming you clip the converter before you clip the analogue path leading to the converter. In which case the bit depth means nothing, you simply have distortion. Albeit, analogue distortion might be ok, depending on the circuit involved.

Here’s an article I wrote for Production Expert that explains the logarithmic scale & how it relates to precision in audio. The article is about decibels, but it covers what I’m referring to: https://www.pro-tools-expert.com/pro...nding-decibels

Let’s say you record the same audio on two channels of an F6 or Mix Pre, but one is at mic level and the other is at line level. Then you normalize both so that they are played back at the same level.

The differences between the very large numerical jumps in the line level recording vs the much smaller numerical jumps in the mic level recording will be the same size after normalization of both tracks. There will be neither more more less precision between the two. This is a factor of the mantissa being 24 bits in both cases. Either track could have been recorded at 24 bits after the input was scaled to the proper level.

The advantage of 32 bit float isn’t an increase or decrease in precision. That is the same as it would be had it been recorded at 24 bits with the level properly scaled in the analog domain first.

Rather, the advantage is in being able to record at this 24 bit precision WITHOUT knowing where this range would fall exactly ahead of time.

[TBman]

Two questions, what about the F3? It seems to be a more modest version $$ wise and what about what Derek (Rick-slo) pointed out about latency issues?

[Doug Young]

Originally Posted by TBman

Two questions, what about the F3? It seems to be a more modest version $$ wise and what about what Derek (Rick-slo) pointed out about latency issues?

I don't see how latency even plays into recording solo guitar on a portable recorder. Latency shows up as an issue when overdubbing in a DAW. Maybe you'll need to increase your buffer size if your DAW can't keep up when playing back the larger file. But for a stereo solo guitar track and an effect plugin or two, most computers made in the last 20 years should be fine! And if you do need to increase the buffer size, it's not something you'll even notice.

Also, depending on your DAW, you may not even be working with 32 bits, it may get converted to 24 when you import the file.

[TBman]

Originally Posted by Doug Young

I don't see how latency even plays into recording solo guitar on a portable recorder. Latency shows up as an issue when overdubbing in a DAW. Maybe you'll need to increase your buffer size if your DAW can't keep up when playing back the larger file. But for a stereo solo guitar track and an effect plugin or two, most computers made in the last 20 years should be fine! And if you do need to increase the buffer size, it's not something you'll even notice.

Also, depending on your DAW, you may not even be working with 32 bits, it may get converted to 24 when you import the file.

Ok thanks Doug. I use Reaper 64 bit version which I think does convert to 24 bits. I'll have to look.

[rick-slo]

Originally Posted by TBman

Two questions, what about the F3? It seems to be a more modest version $$ wise and what about what Derek (Rick-slo) pointed out about latency issues?

https://ledgernote.com/columns/studi...ter%20software.

[Chipotle]

Originally Posted by TBman

Ok thanks Doug. I use Reaper 64 bit version which I think does convert to 24 bits. I'll have to look.

Don't confuse the number of bits used in the *Operating system* with the number of bits used to record.

Reaper 64-bit refers to the version of the operating system, Windows or MacOS. Older operating systems used 4 bytes, or 32 bits, for addresses. This meant they could only make use of 4GB of RAM. Eventually, hardware and operating system software had to increase that capacity, so they doubled it to 8 bytes, or 64 bits, for addresses. This potentially allows for 2TB of RAM to be used.

This is completely independent of the number of bits used for each sample that Reaper or other DAW is capturing. Reaper itself just captures whatever the interface is sending it; internally it processes everything with the full 64 bits. You can then save the resulting audio in whatever format you like, including 32-bit float.

(Fun fact: if you ever need to make music to be played on a telephone system, like on-hold music, you need to save it in an 8-bit format! No wonder hold music always sounds like it's being played through a cheese grater.)