Hard disk recorder modes, int32 vs. float32?

[Dr3Tri]

What's the difference with HD recorder modes like int32 and float32?

---edit---

Well, both work and i don't get any difference with them -at least not anything that i could hear.. so i guess it doesn't matter. That's why i ended asking it now, after pretty long use.. it just would be NICE to know.

[strobotone]

float stands for "floating-point" values (like 1.12345..) which basically could handle a higher resolution compared to
int, which stands for "integer" (whole numbers like 1, 2, 3 ...)

but i am also not sure how it is used in the HD-recorder.

[mantratronic]

They are the different formats of .wav files, 32b float is not clipped and can have very very large noise bursts in it, but 32b int is clipped.

[bahador]

Mantratronic, I was not quit following you, what does it mean 32 bit is clipped and the other one is not? Doesn’t clipping occur when you set your limiter to get some amount of reduction? And also I have exported in both mode 32bit and 32 bit float and there was absolutely no change in RMS of any of the channels. so basically I'm gonna have to ask you to be kind to explain more about the difference between these two cuz I too have some issue understanding them and basically the bit rate of songs, and when a signal is not clipped does it necessarily mean that it is less louder than the other one or what are the benefits in general to have it exported to 32bit float or 32 bit only?

[mantratronic]

Ah, sorry to confuse. What i meant was that the 32 bit float can have samples that are above 1.0 or below -1.0 (where the clipping occurs). When this is played, it clips if you have badly mixed the track (ie, forgot to put a limiter in) or creates a noise burst by trying to push the speaker too far. If the float sounds like the int on your machine, that means everything is fine, it's only if a mistake has been made that you hear a problem in the float.

The reason 32 bit floats are dangerous is that if you have accidentally forgotten a limiter or have a noise burst, it could damage someone else's equipment when they are mastering.

[bahador]

Assuming if you have a limiter at the end of the chain and you have already set the ceiling of your limiter to -0.1, is it still able to produce any clipping or noise burst? and what do you recommend as the final export, generally? 32 bit or 32 bit IEEE float?

And also if you had the item I would be very appreciative if you could explain how does a noise burst happens during the production?

[mantratronic]

Assuming if you have a limiter at the end of the chain and you have already set the ceiling of your limiter to -0.1, is it still able to produce any clipping or noise burst? and what do you recommend as the final export, generally? 32 bit or 32 bit IEEE float?

No, the limiter will deal with the problems. I would recommend 32 bit Int given the only sound engineer I know is the one who told me they hate 32 bit floats because of the problems I outlined.

And also if you had the item I would be very appreciative if you could explain how does a noise burst happens during the production?

I wish I knew! I think there are a bunch of reasons, buggy machines for example, but I don't really know.

[bahador]

Thanks a whole lot for your time Mantratronic, appreciate it.

[Joachip]

Buzz operates in 32-bit float internally, but this only gives you 24-bits of actual precision when exporting to integer numbers. So 32-bit int and 24-bit will give you the same sound quality.

When exchanging audio between external companies like mastering studios etc. people normally use 24-bit int for this reason. So when delivering sound to other people, I would suggest sticking to 24-bit int.

[bahador]

Hi Joachim, assuming if I have recorded my vocal lines in 32 bits float in another host, won’t it change the output quality if I then export it as a 24 bit wave file? I mean the hard disk recorder will do any dithering or what’s the process?

And regarding what you have mentioned, buzz is only capable of exporting the files in 24 bits, sorry I was not quit following you and I didn’t get why 24 bit and 32 bit sound the same qulity-wise? and do you say exporting the file to 32 bit float IEEE will also give the same quality precision like 24 bits?

[Joachip]

Yeah. The only real benefit is that float can be louder than 0 dB.
If this is just for your own files, you can use whatever format.
But if you are sending files to someone else, I recommend 24-bit int.

[mcbpete]

That actually sounds real useful, I mean I know you shouldn't really have stuff peaking over 0 but having the ability to go above that limit and then normalising back 'down' to 0 on my mastering DAW seems like it could have its benefits.

Is there a max threshold for this 32-float file format (Reaper goes up to about +10 if I remember correctly) and how does Buzz interpret stuff above 0 internally (I always thought that 0 was the max on the VU meter and above that you just start clipping)

[oskari]

That actually sounds real useful, I mean I know you shouldn't really have stuff peaking over 0 but having the ability to go above that limit and then normalising back 'down' to 0 on my mastering DAW seems like it could have its benefits.

Is there a max threshold for this 32-float file format (Reaper goes up to about +10 if I remember correctly) and how does Buzz interpret stuff above 0 internally (I always thought that 0 was the max on the VU meter and above that you just start clipping)

Float format can go up to +771dB but Buzz is limited to only +680dB internally.

[strobotone]

hehe.. that sure leaves enough headroom.

[mcbpete]

Flipping 'eck Tucker

[bahador]

mcpete how exactly you benefit from something going way beyond 0 ? I mean isn't the max threshold 0 on digital domain?

and assuming if you were to send your stuffs to a mastering studio, you should leave some headroom for the engineers to be able to work on your track so then again you have never used the floating threshold, am I right?

since working with the new hd recorder I have exported all of my files in 32 bit float IEEE, but as i learned from the discussions here it has no usage in the digital domain and as Joachim has suggested we gotta stick to the 24 bit version.

and still it brings a bunch of other questions to my mind that if the max threshold is 0 why there should be a bit rate that is capable of producing frequencies over that threshold, for LPs? (are they mastered in a digital chain?) or what is the main usage of that bit rate in general?

[mcbpete]

Well as I see it, as Oskari says (possibly with tongue in cheek, possibly not) no 0db isn't the peak headroom in buzz, it goes considerably higher than that - so for me if there is something that goes above the 0db limit if I record in 32 bit float but later downsample and normalise with dithering I'll end up with something with a slightly (though no doubt imperceivable) better dynamic range. Even if that's not the case, it's a good safety net in case some resonant effect starts to stray close to peaking but in 32bit float land it'll ensure that it doesn't clip.

Though note this float stuff has nothing to do frequency, only amplitude. Plus I'm only guessing having only heard about this format yeasterday, I could be talking rubbish !

[bahador]

I have recorded my vocal lines in a 32 bit float session but again clipping happens or so I felt, cuz again I'm recording in a digital domain which shows 0 as the max threshold and going way beyond it will be clipped, I tested that by shouting out loud in the mic. but still something not even 60% clear to me.

[UNZ]

I have recorded my vocal lines in a 32 bit float session but again clipping happens or so I felt, cuz again I'm recording in a digital domain which shows 0 as the max threshold and going way beyond it will be clipped, I tested that by shouting out loud in the mic. but still something not even 60% clear to me.

distortion in general has nothing to do with 0db. you can have distortion well below 0db or over, or anywhere...even if there is no clipping at 0db (which there most certainly IS when you record).

also, "headroom" is a stupid concept in the digital domain, it just doesn't apply. with floating point, all that changes in the numbers is, well, the floating point
(1.0 or 10.0 or 100.0, who cares, it represents the same thing if the scale of reference is the same, 1 or 10 or 100...the thing is that you can mix and match BIG AND SMALL numbers without much problems with floating point, all that decreases is the precision with which it represents those numbers if they get ridiculously small or big, but in audio we generally stay around -1.0 to 1.0 which is optimal.) you're better off thinking about dynamic range and signal-to-noise-ratio than headroom. A .wav (lets say 16bit) will not sound better because you leave headroom, to the contrary, you're just wasting bits. The reason mastering engineers tell you to "leave some headroom" is that they dont want to deal with idiots sending them 16bit int or 24bit int .wav that are clipped, but it has absolutely nothing to do with "headroom". (why mastering engineers dont just embrace floating point standard which would remove this problem is beyond me... i use it all the time because i can't be arsed to care about occasional peaks going over 0, no problem with floats there, it doesn't clip, its actually the closest representation of the old analog "headroom" world that that nobody seems to be able to get out of their small heads... first thing they'll do is slap on a limiter anyway, or you could just normalize it and "magically" all the peaks will be below 1.0, no clipping happened at any point...)

[Joachip]

The maximum within the digital domain is different between int and float. I'll try to simplify this for normal Buzz use cases:

int

• Maximum level = 0 dB
• Used when exchanging .wav files between foreign people, studios, sample web sites etc.
• All soundcards use int when sending data from the driver to and from the converters. That is why you can never record or play anything above 0 dB.

float

• Maximum level = +680 dB or more
• Used internally within your software, for temp files and for exchanging files between software within your own home. Not recommended for exchange with foreign people who use unknown software.