in reaper because of nebula and i guess because i can thanks to technology i save renders in 96khz and 32 bit floating point. i've been trying to help a friend mix/master his songs (blind leading the blind, that'll be a separate post to see if i'm even improving anything or not) and i didn't realize ableton was distorting the mixes i was sending over. i thought ableton supported that file but i guess not, my friend thought i was just sending shitty masters that were clipping by 2-3db haha which is kinda funny. sorta.

anyway, i used voxengo r8pro to convert the files to 44.1 24 bit, and i selected "prevent clipping" and those converted files are still clipping at .1 db sometimes. the original file is limited to -.3db. if i change the render options in reaper to 44.1 24 bit it results in clipping. how do i stop this from happening, besides just turning down the converted file until it doesn't clip and rerendering, or is that just how its done? do i have to change my sessions to 44.1 24 bit just to render it without any clipping occurring? it seems like it should be simple so i'm guessing i'm just dumb and missing something.

 

I'm sorry I don't have the answer, but I'm curious... why 32-bit and the extra steps it takes? I read that human hearing can't distinguish anything over 22 bit and I don't know the details.

Thanks!
-Mark

 

well i'm certainly regretting the extra steps now i can tell you lol. AT THE TIME it just seemed like... why not. the 96khz i do because nebula is built around it being at that sample rate, i guess i just picked 32 floating over 24 bit just cause its theoretically got more headroom and my computer could handle it without sweating. also reaper works in 32 bit floating point (or thats where mine is set anyway, i think it goes up to 64 bit floating) so i figured id just save them like what i was working in. now many projects later they're all saved like that.

i dont know why ableton is distorting them it says it supports 32 bit floating point on the site. but it seems like it happens to other people too and the answer seemed to be just convert them lower.

 

32bit FP has 1528dB of dynamic range, which basically means you can clip it (way above 0dBFS) and still maintain the entire waveform. In 24bit audio (144dB of dynamic range) you can't do that - once you clip at 0dB you are fucked and the peaks are beheaded.

So, just use proper and healthy gainstaging (don't clip the audio) and leave some headroom for overs/ISP. Otherwise your converters in your audio interface, which are highly likely 24bit at most, will clip/distort - which seems to have happened here.

Or just send 24bit audio to each other. It's perfectly fine dynamic range, way beyond the human hearing.

When downsampling to 16bit however (if you ever decide to do so for some reason) you might need to add noise (dither) to quantize the bit depth data at the very lowest of volumes, like reverb tails. If you don't you might get sort of like a gating effect on the very low level material (such as said reverb tails). Most notable on classic music (which has very big dynamic range) on CD, where they might have missed doing dithering.
16bit has 96dB of dynamic range.

 

Last edited: May 27, 2024

Easy to solve with these options when rendering

 

Last edited: May 27, 2024

The mentioned down-sampling is also a way of introducing distortion due to quantization errors, and people will often turn to dither.

https://www.izotope.com/en/learn/what-is-dithering-in-audio.html

 

Airwindows' free dithers are fantastic for this (32 to 24). I love Ditherbox's "Vinyl" setting.

 

I work pretty much exclusively @96k/32BFP ever since I obtained the computing power to do so. The 32bit Floating format is more of an insurance policy when in the production phase. You can pretty much do anything you want to that data (within reason) and you wont mess it up including but not limited to sending it somewhere else to continue production, mixing, mastering, etc.

The audio hardware (wherever you are sending it) doesn't necessarily see it that way however, since there are as many varying quality levels of hardware as there are OEMs producing it. If there is sufficient level present (with 32bit FP there can easily be) to run the hardware into the red (audible distortion) it can and certainly will sound crappy even though the digital audio data remains intact. The beauty of working with 32 bit FP (or 64 even) format is if your digital levels are clipping your hardware you can simply adjust your digital data down in "volume" and never gain any anomalous artifacts by doing so. In this way your project can stay intact on pretty much any hardware platform. Of course at the final mastering phase the project will eventually need to be dithered down to an industry standard distribution format. Unless you're mastering for a high end video production or something...

 

Last edited: May 27, 2024

@saccamano Thank you! I've always wondered.

 

Why don't you load the renders as an audio track in Reaper and check they really don't clip?
Besides that I agree with what the others said.

 

Convert thru SRC before limiting. Then use true peak limiting and oversampling for limiting but with care. It is safer, but kills more dynamics and punch. So it is like PCM limiting vs "analog" matched, approximized.

Dither lossless, anyway.

 

YES! I had an Akai S1100 (S1000 plus effects) in the '90s. I heard that gating sound on the end of the reverb tails and always wondered why. It was a 16-bit sampler, but I have no idea what the bit depth of the effects processor was.

 

You know, something i actually realized very recently:
if you're limiting at a high enough SR via oversampling, then it actually doesn't need to be true peak. Because what is true peak limiting other than catching ISP's right? And there's no significant inter-sample peaking if the samples are extremely close together (like oh i don't know, 3.072 million per second? )

Some of the highest quality limiter / clipper plugins (n my opinion), all of which are capable of functioning at or above 384 kHz:

-A.O.M. Invisible Limiter G2
-FabFilter Pro-L 2
-iZotope Ozone 11 Maximizer

-Acustica ASH / ASHULTRA
-SIR Audio Tools StandardCLIP
-Voxengo OVC-128

ALL of these are available on the sister site.