Sample rate conversions - is 24 bit different than 16 bit?

[lfm]

Recently figured out that Cubase is horrible on sample rate conversions compared to all other main daws.

So thinking how much I can do in Cubase?
a) render 24-bit 48k and do both dither and bitreduction and SRC externally?
b) render 16-bit 48k with dither and bitreduction and just do SRC externally?
c) render 24-bit 48k, do SRC externally and then import 44k/16 into Cubase and do dither and bitreduction as last operation?

Are there anything that makes better quality if keeping bitdepth as SRC is done?

Thanks.

http://src.infinitewave.ca/

Below obviously while keeping 24 or 32-bit float. 16-bit has only 96 dB theoretical dynamic range, but artifacts may inflict on final result, is my assumption.

Levels of artifacts are something in range 30-40 db worse in Cubase.
All other main daws have almost completely black background, and very sparse with artifacts above - 140 dB. Looked at PT, Reaper, Sonar, Digital Performer, Samplitude.

[EvilDragon]

You can always use SoX externally.

[lfm]

You can always use SoX externally.

Many thanks, will test that for sure.
Seems to produce as best daws at the infinitewave testing.

But is it significant if 24-bit while doing that - or if SRC algos have nothing to do with bitdepth?
If artifacts in 16-bit even might be better, than 24-bit - in that artifacts are outside range of 96 dB to start with - less influence in audible range - or similar.

Thinking process is a bit simpler if doing normal 48k/16 render and dither in Cubase and just finalize with SRC externally.

Also have to check RightMark Audio Analyzer if it has any graphs that tell about this.

Just assume that 48k->44k is much worse than 96k->44k. Infinitewave had their testing files and don't know if RMAA are very different.

[BertKoor]

Bit depth should not have anything to do with the sampling rate. It's another dimension...

[lfm]

Bit depth should not have anything to do with the sampling rate. It's another dimension...

Thank you, was hoping for something like that.

Fiddling some with RMAA to see if one can see any number being different doing it different ways.

[BertKoor]

Fiddling some with RMAA to see if one can see any number being different doing it different ways.

What you should observe in RMAA is that 16bit material poses technical limitations on the best possible outcomes on things like dynamic range and harmonic distortion. Because a least significant bit weighting at around -90dBfs is quite different than when it's at -140dBfs (and well below the noise floor.)

RMAA was designed in the age when 16bit cards were dominant in the consumer market. All RMAA's metrics can score "excellent" even on a mediocre modern card that's 24bit just on paper.

[lfm]

Fiddling some with RMAA to see if one can see any number being different doing it different ways.

What you should observe in RMAA is that 16bit material poses technical limitations on the best possible outcomes on things like dynamic range and harmonic distortion. Because a least significant bit weighting at around -90dBfs is quite different than when it's at -140dBfs (and well below the noise floor.)

RMAA was designed in the age when 16bit cards were dominant in the consumer market. All RMAA's metrics can score "excellent" even on a mediocre modern card that's 24bit just on paper.

Still think it's modernized a bit - showing -200 dB range etc.
The test patterns show as -180 dB noise floor.

It's been very useful to me. I figured why a dvd player I bought sounded so awfaul. And using spdif outs it did strange cheapo something 44k conversions for cd when playing. It also had a volume control working on spdif out, which in itself suggest heavy digital volume of some sort. It showed about 3% thd etc.

Anyway, I returned that dvd player - not good for anything.

[lfm]

A few things I found interesting testing r8brain, rmaa test, cubase 9, and sonar.
You can actually see in two distortion numbers how Cubase is the worst.
Even if in fourth decimal, it shows which was one thing I wondered about testing further.

Something did not work in RMAA generating 96k pattern, so I did double conversion - first 44-96, then 96-44 again. RMAA column is the reference test pattern and how it looks in analyzer.

I could never import from SoX into RMAA, don't know why.



I did this 96-44 as comparison to infinitewave site, so will have a look at how it looks 48-44 later.