Topic: 32Bit Float recording - The Technical view

[Ozpeter]

I am possibly repeating myself, but 32 bit float A to D conversion cannot clip.  The prior analog stages can.  But these devices are designed with gain staging set so that under normal circumstances you would have to work hard to clip the analog front end.  +4dBu is industry standard.

[Ozpeter]

Meanwhile, I remembered while reading this thread that my very first digital recording of any kind, back in 1983, was... 14 bits.  How far we have come!  When I was asked to make it, I had never even heard a digital recording.  I had a LOT to learn.  I hired a Sony F1 / SLF1 system that used betamax tapes.  I borrowed a pair of mics.  Here are a couple of relevant links.  The second link reproduces the Sony blurb which explained what digital recording was all about.  Actually, I have just read there that the F1 could run at 16 bit for those wanting the ultimate quality.  So maybe I did the right thing and used that option in fact.

https://www.palsite.com/slf1ovi.html

https://www.palsite.com/pcmf1ovi.html

Anyway, the end result was no disaster (imho) and it can still be heard on YouTube Music.  At the time it got a gold award from a French magazine review. It's the first album in the playlist of some of my albums which I had a bit of fun putting together today, using the Discogs database to remind myself what I have recorded over the years.  Some albums I had completely forgotten about! 

https://music.youtube.com/playlist?list=PLW7X_zDpTwy3bogeppBSktVXgfkvgzNLf&si=6qedaTGP50lrDuMg

Anyone interested in what a very old digital recording sounds like (possibly mangled by YouTube) might care to listen to the first 30 seconds...

I guess this post is the very definition of off-topic. Sorry!

[Gutbucket]

^ Would be cool if they included an internal function that would both normalize the 32bit float file AND produce a 24bit file from that on the recorder itself, streamlining post work where desirable.  Sound Devices should be able to do that easily since the MixPre IIs already have the capability of supporting internal mixdown to a new file, I believe.

Gut, you're gonna have to demonstrate that there is substantial benefit from doing the processing on the unit versus a DAW. That touchscreen is minuscule, even compared with a laptop monitor. And control bandwidth on the mp-x is again minuscule compared to a kbd and mouse.

Why would I want to do those things on the recorder?

Just as an option which would allow the user to leverage the 32bit float advantages of not having to set input levels ahead of time, but produce a 24bit fixed output file directly on the recorder instead of having to transfer the file to a computer to do that.  Wouldn't be forced to operate in it that way, it would simply be a nice to have option.  Could always still transfer the original 32bit float file to the computer later.. along with its 24bit fixed derivative.

[Chilly Brioschi]

I plan to chain a (borrowed) Beachtek DXA-PRE ahead of the F3 the to see what sort of mayhem ensues.
The owner does corporate gigs, so I'm trying to find a time that he doesn't need it and I have a local gig to tape.

[TheJez]

^ Would be cool if they included an internal function that would both normalize the 32bit float file AND produce a 24bit file from that on the recorder itself, streamlining post work where desirable.  Sound Devices should be able to do that easily since the MixPre IIs already have the capability of supporting internal mixdown to a new file, I believe.

Gut, you're gonna have to demonstrate that there is substantial benefit from doing the processing on the unit versus a DAW. That touchscreen is minuscule, even compared with a laptop monitor. And control bandwidth on the mp-x is again minuscule compared to a kbd and mouse.

Why would I want to do those things on the recorder?

Just as an option which would allow the user to leverage the 32bit float advantages of not having to set input levels ahead of time, but produce a 24bit fixed output file directly on the recorder instead of having to transfer the file to a computer to do that.  Wouldn't be forced to operate in it that way, it would simply be a nice to have option.  Could always still transfer the original 32bit float file to the computer later.. along with its 24bit fixed derivative.

The upcoming Zoom H1 XLR has exactly this option, see screenshot of the manual...

[Ozpeter]

^ which is where we started that discussion on 22 August. 

[Gutbucket]

 
I suspect other recorders will incorporate the same over time.

..and wonder what value the H1 XLR normalizes the highest peak to, since it doesn't say.

[Ozpeter]

I will try to check that function on my M2 Mictrak device tomorrow to see what level that gives when normalised.  I imagine other devices in their stable will work the same way.

[Ozpeter]

Well, using the Zoom M2 I generated a test tone @ 440Hz using Adobe Audition, recorded 30 seconds of it from loudspeaker replay, and then used the built in export function to normalise the recording and export to 16 bit and 24 bit (two rapid operations).  Then I opened the exported files one at a time in  Audition, replayed them, and noted the maximum level shown (and retained) on the meters.  This showed that the file was normalised within the device to 0dB.  Originally the peak value of the 32 bit float file was -11dB approx.

However, this exercise revealed a drawback in using the built in normalisation function.  Right at the start of the recording there was some kind of transient click, either from handling or from pressing the record button.  This was actually louder than the recorded tone.  So the tone wasn't normalised as such - the click dictated how the whole file was normalised.  If I had been doing this with real material I would probably have edited off the click in the DAW, and then normalised the actual wanted part of the recording.  This might not happen all the time, depending on circumstances - I wasn't very careful about how I was holding the device when I recorded the tone, and as it's almost midnight here I replayed the tone at a modest level not wishing to disturb the neighbours.   But it does illustrate a possible hazard in undertaking whole-file internal normalisation.  But, it would give you something reasonably playable in terms of level, prior to further level control operations. 

[Edited to add, while talking about this much maligned device - I just recorded some room silence using the M2 by the time honoured method of shoving it under a folded duvet in the early hours of the morning in a silent bedroom.  I then examined the file in Adobe Audition.  In the spectral frequency display there was absolutely no horizontal lines at all (which would indicate a noise band of some kind) and that was with the range cranked up to 210dB.  Looking at the frequency display there was a maximum level of about -91dB at around 100Hz.  This could be city rumble.  At 1000 Hz the level was about -117dB and at 5Khz the level was -126dB. The level continued to drop slightly so that at 10kHz it was -127dB.  Above that it slightly declined further.  I doubt whether this amount of noise (system noise plus mic noise) would be of any real consequence or concern even in a classical recording - and it seems to me for the price of the device, it's nothing to complain about.  I paid $117 US dollar equivalent back in December.]

[Dan33185]

If line-input via TRS is capable of handling an input level of up to +24dBu, it should handle a patch from the soundboard just fine.  I Just checked the on-line manual and found the note below-

"NOTE
To make handling the levels of input signals easy, the H4essential sets input levels according to the types of plugs connected to the input jacks.
• When connecting mic-level devices, use XLR plugs.
• When connecting line-level devices, use TRS plugs."

Ok, just to make sure I am understanding it...when I am using my mics or connecting to a soundboard/mixer that has an XLR output, use the XLR inputs on the Zoom. If I am connecting to soundboard/mixer with a 1/4" connection, I should be using the "Line In" input? So using a 3.5 MM input with a dual splitter and 1/4" adapters to the soundboard should work? No attenuator needed because it's 32 bit float?

Microphones get plugged into the XLR input on the recorder, regardless of what output connector the microphone actually features, because they will all produce a lower "mic level" output.  Sources that produce a higher level of output, such as a patch from the mixer, soundboard, or the output from an external preamp go into the TRS input, regardless of their output connector. That means you will need to use adapters to convert from XLR to TRS and vice versa whenever necessary, or use dedicated cables with the different appropriate connectors installed on either end. 

Had a thought today...right now I carry a set of cables for each possibility (XLR male to male, XLR female to male, and 1/4" male to 1/4" male). If I were to just buy some adapters such as this:

https://www.sweetwater.com/store/detail/PWP047Z--daddario-planet-waves-pw-p047z-xlr-male-to-1-4-inch-female-balanced-adapter

Would that give me the same results as a dedicated cable? I love the idea of bringing just a 1/4" male to male cable and just adding adapters when needed.

[Gutbucket]

In terms of signal path, yes. You could plug your 1/4" TRS terminated cable in directly for line-in, and add the XLR adapter for mic-in.

In terms of practicality, consider things like how long the adapter connection becomes which is going to stick out from the side of the recorder - how you will manage that, how much leverage it applies, and how good and tight the additional signal path connections are.. stuff like that.  A lot of times I prefer a short "pig-tail" type adapter that features a short wire between the two connectors just long enough to provide strain relief and the ability to fold under the recorder.  The adapter-to-recorder connection then protrudes far less and is much better strain-relieved, and the cable-to-adapter connection can be neatly tucked away under or behind the recorder in the bag.

[Dan33185]

In terms of signal path, yes. You could plug your 1/4" TRS terminated cable in directly for line-in, and add the XLR adapter for mic-in.

In terms of practicality, consider things like how long the adapter connection becomes which is going to stick out from the side of the recorder - how you will manage that, how much leverage it applies, and how good and tight the additional signal path connections are.. stuff like that.  A lot of times I prefer a short "pig-tail" type adapter that features a short wire between the two connectors just long enough to provide strain relief and the ability to fold under the recorder.  The adapter-to-recorder connection then protrudes far less and is much better strain-relieved, and the cable-to-adapter connection can be neatly tucked away under or behind the recorder in the bag.

For the most part, my recorder is at ground level, or on a platform that doesn't cause any stress to it due to cable pull. I record a lot of local, smaller musicians who have a very basic setup, and usually I just plug in and set it on the floor behind their setup. I've bought a couple cases and adapted them to my recorders, so the recorder sits inside the case and I drill holes in the bottom of it to plug cables in, it keeps it safe from harm and together as one unit basically.

[Ozpeter]

Perhaps I am at risk of being accused of not having read the whole thread, but...

In the thread related to the Zoom H4e, a user has mentioned that in the manual for the device (and not in the advertising!) the 3.5mm input is flagged as not being fed through a dual a/d converter.  Subsequent research indicates that some He series devices do not claim to have dual converters at all.  So what is happening?  Is the audio simply going through a 24 bit converter with the output saved in 32 bit float format, and is there any actual benefit in doing that, apart from the sales benefit of having a 32 bit float sticker on the device?  I am confused... again...

[TheJez]

Perhaps I am at risk of being accused of not having read the whole thread, but...

In the thread related to the Zoom H4e, a user has mentioned that in the manual for the device (and not in the advertising!) the 3.5mm input is flagged as not being fed through a dual a/d converter.  Subsequent research indicates that some He series devices do not claim to have dual converters at all.  So what is happening?  Is the audio simply going through a 24 bit converter with the output saved in 32 bit float format, and is there any actual benefit in doing that, apart from the sales benefit of having a 32 bit float sticker on the device?  I am confused... again...

Good question...
Indeed it is very easy to convert the output of a single 24bit ADC to 32bfp format. I do see benefit for storing in 32bfp if math is done on the samples in the digital realm in the recorder (amplification, filtering, equalizing, mixing, limitting etc). Some of these operations could potentially result in samples > 0dB, which are not possible to store as 24bit linear samples without clipping. And even if all samples remain < 0 dB, storing the output of such operations as 24bit linear samples would require rounding, while storing in 32bfp would allow to store the resulting samples more accurately. If this difference is audible is another question...

[Ozpeter]

Thanks for your response - I have to say that I trust it rather more than the response I got from "DeepSeek" (which I am amusing myself with at the moment), which I found more confusing than helpful!  I think I will continue to ask about location recording matters here rather than there in future...  I won't waste space with the full response, but here is the summary bit -

"Summary
There is no single ADC that directly outputs 32-bit float audio.

Dual ADCs are used to capture a wider dynamic range and avoid clipping, which is then processed into 32-bit float format.

32-bit float is a digital format created in post-processing, not a native output of an ADC.

This approach allows portable audio recorders to deliver the benefits of 32-bit float recording, such as immense headroom and dynamic range, even though the ADCs themselves do not natively output 32-bit float data."