I don't get your logic here - why you make it sound as though there's a difference between the two. There are no dynamics below the format's noise floor. There are no dynamics above the format's ceiling. The format's noise floor to 0dbfs* IS your dynamic range.

(*float formats can go above 0dbfs)

 

The difference is file size. 144dB is not better quality than 96dB dynamic range. It's just a matter of necessity. What does your project require? Choose your project settings according to your end-result-requirements. Will your end result be 14dB Dynamic Range? If so, why bother with 24-bit, or even 16-bit?! .................. But you should probably stick to 16-bit, simply because it is an "end-result-standard." The CD standard. But also, who the fuck buys CDs anymore? The reason DVD's and BluRay's use 24-bit audio is because it offers a higher dynamic range - It's 144 compared to 96 CD standards. It doesn't "sound better," it's just different specs.

 

Last edited: Mar 3, 2025 at 2:29 PM

As is often the case in life, there are no clear answers to the question of the optimal digitization parameters (sampling rate & word width). Nevertheless, I would like to give a few personal experiences to those who do not want to spend years doing listening comparisons:

1. It depends largely on the quality of the original recording (and thus on the performance of the sound engineer and the mastering engineer). There are excellent recordings in Red Book standard quality and mediocre high-resolution recordings and vice versa. In general, of course, the high-resolution recordings are usually better, at least because they have been mastered more carefully than the same album for release on CD.
2. The quality of today's digital recordings or remasterings far exceeds the capabilities of today's usual - even high-end - playback chains (especially loudspeakers and room acoustics), so the number fetishism that is sometimes practiced when it comes to sampling rates sometimes has an academic character.
3. While I personally cannot perceive any reliable and comprehensible differences in modern recordings between the same recording at different sample rates starting at 44.1kHz, so that in my opinion 44.1kHz or 48kHz are quite sufficient with today's filters, I do think I can detect differences in the word width of the same recording, although in my experience 20-bit recordings also sound significantly better than 16-bit recordings.
4. Things are a little different with older recordings (for me, these are recordings from before the turn of the millennium). Since the anti-aliasing filters used in the past were more invasive for the hearing spectrum than modern filters, I prefer the highest possible resolution for older recordings.
5. For me personally, this means that I always try to get a lossless 24-bit version of every recording. The sample rate is largely irrelevant to me up to 96kHz. If several versions are available, e.g. 48kHz and 96kHz, I prefer a little “headroom” and go for the 96kHz version. In addition, due to the file sizes, I have a preference for “not too high” sample rates, i.e. 88.2kHz/24Bit or 96kHz/24Bit instead of 176.4kHz, 192kHz or 384kHz – not to mention 768kHz. So if the same file is available in 24/96 or 24/192 when downloaded, I prefer the 24/96 version.
6. I consider DSD files (2.82MHz/1Bit) to be comparable in quality to 20Bit/96kHz files, although they sometimes have slightly different sound characteristics, as I explained in detail in my article “PCM compared to DSD”. I would be indifferent between the two versions, although I do notice subtle differences in the sound characteristics, which sometimes work in favor of the DSD version and sometimes in favor of the PCM version.

 

No need for “listening comparisons.” It's maths.

 

Not important to me personally - I'm fine with excessive quality for archiving purposes.

A format for my renders / my samples that doesn't get audibly noisy if I slam OTT on them. 16 bit int is not that.

Yeah, that's the important one. I want my archiving (/ project) format to be something that can take a near infinite number of future processing changes without audibly degrading quality - so that I might derive as many new end results as I need. I.e. render from HQ archiving format to whatever is appropriate for the target platform. 44.1/16 CD, 48/24 video, streaming sites that use Opus for audio.

 

There's an article below in the link if people want to understand most of the pluses and minuses, including Nyquist and the reasons when to use different types of sample rates.

https://producerhive.com/ask-the-hive/48khz-vs-96khz-differences-which-is-best/


P.S. Whoever insinuated that 96 kHz doesn’t make any difference next to a 48 or 44.1 kHz sample selection at accurately reproducing a 20 kHz tone, and it has a bigger file size is correct.

I also know why sound designers use sample rates much higher and their reasons are completely logical. Read the article, it explains this and much more if people are interested.

For anyone doing their own music who is not intending to do a lot of time-stretching, then 44.1 and 48K will fulfil their needs. I prefer 48K, 24bit for most things but that's my own personal choice that has been reliable so far.

 

Higher specs is not better quality. You're just moving further away from the noise floor and increasing your file size and dynamic range.

 

I'm the arsehole/asshole doing that in this thread lol

 

I told you what I want to do and why 16 bit int can not do it. Are you certain you understand how the bit depth noise floor works? (I mean no offense - I'll make drawings if you don't)

 

That's fine, I'm not offended. My answer to you is this: I understand the bit-depth/noise floor relationship to be exactly as it is in the xiph.org video. That understanding is essentially this: The lower the bit depth, the higher the noise floor. The higher the bit depth, the lower the noise floor.

Does that sound correct to you?

 

The xiph videos are great and correct - but in what way is it not clear to you then that 16 bit int sample + slammed OTT = distorted mess?

 

"Slammed?" You can make anything a "distorted mess."

Does "slammed 32-bit floating point" sound better to your ear than "slammed 16-bit?"

The things that you can do inside a 32 or 64-bit mix-engine do not translate to 24 or 16-bit mixdowns/bounces.

 

There are some great audio discussions on this out there online, most of them sound and sensible. A lot suggest that if nobody can hear it and it does not affect the audio that can be heard does it matter.? For some reason that kept reminding me of 'if a bear farts in a forest can you hear it?'.
I think you've suggested somewhere that 48k will provide a sufficiently low noise floor and I'd agree it's more than sufficient.

 

My suggestion was 16-bit/48 kHz.

I THINK (and please correct me if I'm wrong) Dan Worrall "suggests" 24-bit/48 kHz.

 

Of course it sounds better. Because the noise floor for 16/24/32 was burned into my sample the first time it was exported - I can't undo it anymore, it'll always be there. And now I'm making it louder by 60dB with OTT.

So what's going to sound better?

16 bit sample -> -96dB + 60dB = -36dB you can easily hear the noise/distortion now every time the sample oscillates between what used to be silence and -96dB.

The original noise floor is effectively as loud as a 6 bit render now (easy to empirically confirm this yourself).

24 bit sample -> -144dB + 60dB = -84dB still very quiet under most listening conditions.

And when I export my sample after putting OTT on it, I too am going to burn yet another noise floor into the sample. There is no choice, this happens every export/render/bounce. We just get to decide where the next NF is placed on the dbfs scale.

 

You are correct. Each of us has to use what works for us. If you get a great sound at a specific set of bit depths and frequencies, then honestly, it does not matter what anyone says, you get a great sound and that speaks for itself.

 

24/48 makes sense because of DVD/BluRay, however, I'd just like to add that 24-bit is excessive. It equates to 144 dB dynamic range, and no one in this world uses that dynamic range. But then again, the noise floor is lower, so.........

 

Do you use Ableton?

 

I don't (Reaper here) - but why not just recreate the simple test setup I suggested a couple posts up?

 

I say this with respect:

I'm not really interested in recreating test examples. I've been talking about technical specifications and trying to avoid conversations about "audio quality." You have repeatedly brought the conversation back to the subject of "audio quality."

You and I are kinda, slightly, a little bit - having separate conversations alongside each other.