rec wav bit-depth: 24 bit pcm (or 32fp, 64fp?)

[jamie rustic]

hello:

i am using reaper with hardware universally run at 96KHz/24bit and have a question as to how to set bit-depth across Reaper.

* in project settings > media i have an option to choose 24bit pcm, 32bit fp or 64bit fp for the recording bit-depth
* in project settings > advanced i have the option to set track mixing bit-depth as high as 64bit fp

i assume i want to use the highest value (64bit fp) for the track mixing bit depth, but am not sure about the recording bit-depth. is the value gained by using a higher bit-depth and fp worth changing from the matching bit-depth of my hardware? if so, what is my value in using 64bit fp rather than 32bit fp?

thanks, jr

[mschnell]

I suppose using a recording file format greater than the recording hardware will just add some dummy bits and hence do nothing useful.

-Michael

[RJHollins]

In practical terms ... RECORDING at 24-bit depth is considered the minimum, recommended depth for audio.

Key point is knowing the final audio destination format.

In Mastering, as I do, all processing, temp files, all go to 32-bit floating point.

Again ... looking at practical practice.

[jamie rustic]

rj- thanks very much.

the final destination of this audio is likely cd/mp3/internet streaming.

is your suggestion that i should record (best) at 64bit fp?

from that recording bit-depth what bit-depth for mixed/rendered audio would you want to see delivered for you to master?

thanks, jr

[heda]

Use 24 bit for recording

Use 32 bit for rendering/editing recorded material. For example if you render effects to a new file.
Export premaster file as 32 bit. Use it to convert premaster to 16 bit mp3 wavs final formats etc

[Megagoth1702]

Originally Posted by heda

Use 24 bit for recording

Use 32 bit for rendering/editing recorded material. For example if you render effects to a new file.
Export premaster file as 32 bit. Use it to convert premaster to 16 bit mp3 wavs final formats etc

+1
No need to record higher than 24bit. Thisguy.
32bitFP for transfers of sound files between programs. To be precise you only need to export at 32bit as soon as you do ANY GAIN CHANGES. So if you change volume (of a track/item or in an EQ) by 0.1dB -> 32bitFP export makes sense. If you just record at 24bit straight into REAPER, didn't do anything to the audio and just want to export stems, you're fine exporting at 24bit, make sure to not use dither at this point!


If you want to go MP3 in the end, I suggest making the MP3 from a dithered (REAPER's dither is fine, make sure to enable noise shaping) 24bit WAV. 32bitFP WAV -> MP3 does not sound as good because there is no dither. And if you use a dither PLUGIN and export to 32bitFP WAV, you're going to end up with a resolution of 24bit+dither, so the additional bits wont make sense.

Due to the lower noise floor of 24bit WAV files the final MP3 does sound better in the super quiet parts. This is super nerdy-detailed stuff though, noone will notice but I just did the tests and figured that out.


The tests went like this:
-insert tonegenerator, 1k sine at 0dbfS
-insert volume changing JSFX
-automate volume JSFX slowly to 16bit/24bit noise floors (-96dBFS/-144dBFS)
-export as 16bit/24bit/32bitFP WAV with variants (dithered/non-dithered)
-convert to MP3
-import back to REAPER
-normalize the quitest parts of the MP3, just around where the signal is about to go into nothing
-realize that 24bit dithered sounds best, by far
-realize that noone will notice or care
-still feel on Bob Katz Nerd Level

What's interesting is that ACTUALLY you're able to hear the program material BELOW the noise floors! So I still could hear the sine FAR BELOW -96dbFS at 16bit and BELOW -144dbFS at 24bit. Crazy nerd stuff man, I love it.

Also on a side note I am not sure whether REAPER dithers automatically when you export directly to MP3 or so. Will test.

[RJHollins]

Originally Posted by jamie rustic

rj- thanks very much.

the final destination of this audio is likely cd/mp3/internet streaming.

is your suggestion that i should record (best) at 64bit fp?

from that recording bit-depth what bit-depth for mixed/rendered audio would you want to see delivered for you to master?

thanks, jr

As HEDA suggests.

[jamie rustic]

all: thanks for your thoughts.

mega: rec @24bit has always been my understanding.

rj: sorry, heda sounds like a standard i dont know about. can you help clarify, please?

what i have been doing is rendering/exporting my completed (96/24) mixes to 96/24bit wavs without dither. those are then mastered with dither, either by me or others, down to a 44.1/16bit wav. this is then followed by conversion to vbr mp3 for distribution.

is this the best process or should i render/export my 96/24 mixes at 32bit fp to be used in mastering?

thanks. jr

[karbomusic]

sorry, heda sounds like a standard i dont know about. can you help clarify, please?

The long and short of it is that at 32 bit float it is practically impossible to clip audio you process as long as it doesn't hit the DAC. You could literally clip a track to roughly +1000 dBFS, glue it, open it up again, reduce the volume by 1000 dB and it will not be clipped.

[RJHollins]

Although there is the possibility of a 'HEDA Standard', I was referring to Heda, one of the posters in this thread.

As to 'bit-depth', it may be useful to consider, again, the practicalities.

1. Audio interfaces, typically' max in at 24-bit [yes, some do allow larger], but at the cost of storage size.

In its raw format, 24-bit allows for massive dynamic range that easily handle lower peak/RMS tracking levels without a problem.

2. Once tracks have been recorded, the option to expand the bit-depth to 32 fp is something to consider, particularly when in a Mastering scenario. In that place, larger file sizes are manageable, and allow for more possible format exportations [a type of 'future proof'].

3. Within some DAWs, the internal, temp, files can also have bit-depth choices. There, it would still be prudent to maintain [at least], the 24-bit format. The use of temp, 32-bit fp formats could be another choice ... but not mandatory. Remember ... more data .... more processing/storage required.

4. Final Mixes [before mastering] ... minimum 24-bit.

Some Mastering Engineers [like myself], will immediately convert a 24-bit file to 32-bit fp for 'their' base start point.

And if wondering .... there is absolutely NO sonic quality or difference when converting a 24-bit file to 32-bit fp.

hope this helps.

[jamie rustic]

rj: heda... duh

thanks for your thorough explanation. it makes good sense.

[ashcat_lt]

One possible reason to try to record to an fp format might be if you are applying Input FX, or recording the output of a track with FX on it. Whether it actually helps is another question, but it does allow you to be a bit more cavalier with in-the-box gain staging without ruining the recording.

[RJHollins]

24-bit depth has a 144 dB dynamic range.

To quote:

"The most important practical effect of bit depth is that it determines the dynamic range of the signal. In theory, 24-bit digital audio has a maximum dynamic range of 144 dB, compared to 96 dB for 16-bit but today’s digital audio converter technology cannot come close to that upper limit. As of this writing, the 24-bit Burr-Brown converters in StudioLive RM-series rackmount mixers, and Studio 192-series audio interfaces offer a dynamic range of 118 dB. The high-speed converters in Quantum-series audio interfaces provide a dynamic range of 120 dB."

The 'push' for bit-depth beyond 24-bit had little to do with 'audio', but to address micro/macro measurements in Science.

The 32-bit FP format is basically a 24-bit container with a scaling 8-bit mantissa.

Obviously ... work as you choose.

[karbomusic]

Having 32 bit float for ITB for conversions has you covered making it nearly impossible for some operation (glue/fx renders) to clip, it isn't about the converters at all but about the resulting file post conversion. Granted I have no problem generally avoiding that but it's nice to have for those one off instances or oversights when you might need to answer to someone else.

[jamie rustic]

this all makes sense. based on this conversation i plan to use the following settings

PROJECT SETTINGS CONFIG:

* record at 24bit, which is my interface max, and 96KHz
* use 32bit wav for the default format for project/region render, unless there is i am made aware that there is a reason to go higher
* ill use 64bit float(default) as the track mixing bit depth

RENDER (FOR EXPORT) SETTINGS:

* export completed mix at 96KHz/32bit fp to be used by mastering engineer

yea? jr

[RJHollins]

That'll cover it as a high resolution file.

Now the hard part ... creating/recording music that befits the hi-rez quality

[jamie rustic]

thanks all

[Megagoth1702]

Originally Posted by jamie rustic

this all makes sense. based on this conversation i plan to use the following settings

PROJECT SETTINGS CONFIG:

* record at 24bit, which is my interface max, and 96KHz
* use 32bit wav for the default format for project/region render, unless there is i am made aware that there is a reason to go higher
* ill use 64bit float(default) as the track mixing bit depth

RENDER (FOR EXPORT) SETTINGS:

* export completed mix at 96KHz/32bit fp to be used by mastering engineer

yea? jr

@96k: Recording at higher than 48k sampling rate is a questionable topic with lots of discussions on the interwebs. I personally feel like 48k is more than enough for music & voice. I only go higher for sound design source sounds, but then again there is Intermodulation Distortion at higher sample rates, which then again muddies the end result. Everyone will get to their own workflow in the end, I guess.

Other than that - yep. Your final product going to mastering engeneer should be 32bit. Tbh there will probably be no audible difference between 24bit-undithered/32bit but oh well, I feel like it's the "right" way to deliver a file for mastering. I don't feel like lowering the resolution before the very final step.

[bezusheist]

Originally Posted by Megagoth1702

...but then again there is Intermodulation Distortion at higher sample rates, which then again muddies the end result.

IMD from high sample rates is a non issue. That myth is based purely on ignorance. (You didn't start it so I'm not pointing the finger at you)
Any processing that generates harmonics will generate IMD as well. (Assuming more than one sine wave as input)
If you have any audio in the audible range, the IMD from that will greatly outweigh any "extra" from extending the frequency response.
They might as well say avoid any/all non linear processing...at any sample rate.

[mschnell]

Regarding sample frequency (supposedly very true for 32 Bit FP):

There is a nice video pointed to in this forum some time ago showing that the sampling A/D->D/A process exactly reproduces the audio signal, given it does not feature harmonics above half the sampling frequency.

If it contains higher harmonics, any decent A/D converter will cut them appropriately.

-Michael

[Beastmode Beats]

64FP uses too much HD space, it’s just double precision, either way you’ll make the same mistakes twice. You want your final to be a 32 FP file before you master, because all DAW’s cant recognize a 64FP file, so a mastering engineer won’t be able to use those files if their DAW uses an older software version. 32FP is singular precision (it’s used to detect common errors in a 16 bit file to detect clipping). A cd is 16 bit no matter what. So you can set them to 24 bit on a render, but try 32FP for each track so you get the exact frequency spectrum of the original room recordings and use less drive space. Throughout the mixing process its all about getting rid of unused higher frequencies before you render to prevent hiss from being heard on a 16 bit file, cut the highest frequencies to reduce the overall size of a compressed file, by doing this, algorithms in mp3 codecs don’t identify frequencies that you’ve already cut at higher bands while they are compressing the file. So they skip to the next sample bit at the highest frequency band identified in the file, whichever sample is found in the higher frequencies starts to add data in the compression process and sends temp files and store cache files about the file (each frequency band increases the amplitude of a signal, that’s why higher frequencies use more hd space) and 64FP will just kill HD/SSD space

[Beastmode Beats]

Record live instruments at 96kHz 32FP, render all at 32FP (for midi instruments 48kHz is perfectly fine), if you find errors on 16 bit cd later, then do a 64FP render to identify where errors are occurring by checking the RMS levels on the first render.

[Beastmode Beats]

Lastly, if you render 32FP and send it to a mastering engineer they can do more with the song! Because its 32FP levels can be altered.