64bit float files + HISE convolution / audio loop player

[Dan Korneff]

@Christoph-Hart Ever since cubase 10 came out, that's all I've been using..... 96kHz too.

[Lindon]

@dustbro wow - thats a LOT of wasted disk space...

Theres a great (simple) video on youTube that deals with the mis-conception that > sample rate = "smoother" results (spoiler:it just doesnt) but I cant find it. But heres a wothwhile read:

24/192 Music Downloads are Very Silly Indeed

(people.xiph.org)

[Dan Korneff]

@Lindon I dunno. At high track counts 96K always sounds better to me. 44.1k gets cloudy and less defined. I can't speak for a comparison ITB though... my studio is mostly analog

[Lindon]

@dustbro - read the article - you are actually making it worse. not better., and costing 6x the disk space. - but your call obviously.

As it turns out the url I point to includes the video I mentioned:

Xiph.Org Video Presentations: Digital Show & Tell

(xiph.org)

[ustk]

@dustbro so the problem might be your converters... But I respect that, on my side, I am working mostly at 96khz

But for the bit depth, honestly 24 is MORE than enough, and 32 float if you want to be free of any internal clipping regarding your DAW. since the plugins are generally coded to work better in a zone under 0dbfs, there is no reason to go higher (especially if your studio is analog oriented), thus makes 32 float unnecessary
64 has no sense since 32 float can handle everything...

[Christoph Hart]

OK, sample rate is another thing, and some effects just will sound better at 96 kHz because it's basically oversampling. Also the artifacts of bad resampling is less noticeable because it usually creates mud in the upper frequency part, which is again, above the 20kHz reception limit.

But I still would like to see (and not hear, I mean see in the sense of a theoretical explanation) why anyone would want to store a signal's amplitude with 64bits of data. Don't get me wrong, double precision values have their right of existence in DSP - usually the frequency counter of an oscillator needs to be a double, and filter coefficients have to be double because some of them get ridiculously small and would get truncated to zero in normal float precision. But for storing the amplitude, it's like Lindon says, just a waste of disk space (and processing power, the bigger the data type, the higher the cache pressure on the CPU).

[Dan Korneff]

@Christoph-Hart said in 64bit float files + HISE convolution / audio loop player:

But for storing the amplitude, it's like Lindon says, just a waste of disk space (and processing power,

From what I understand, it's the opposite.
This is what a developer @ Steinberg says:

Any critical process is executed within double precision (64bit). That was so within the old engine, and is still so in the new.
Plugins handling critical processes also upsample to double precision.
Old VST2 standard was 32bit in- and out.
VST 3 is 64bit in- and out.
Making the engine full double precision eliminates the need for upsampling and truncating before and after each insert slot and from each channel to bus/bus/output. So since the plugins are (or can be) 64bit nowadays, there is no reason at all for "converting" before and after the inster slots.

[Christoph Hart]

Sorry, but I call BS. First of all, this is not upsampling (it's changing the bit depth), and there is no "critical" process that couldn't be done in 32bit.

My guess is that the marketing department wanted to create a random spec improvement over the competition.

Also the entire signal chain in HISE is 32bit float, so even if you drop in your precious 64bit material, it will get sliced to 32bit by the SWORD OF REASON :)

[Lindon]

yep what @Christoph says - thats B**sh*t

[Dan Korneff]

‍️‍️
It'll be interesting to see this thread in 10 years.
"262,144 bit files won't load in my AI chamber"
"261,032 bit files are sufficient for any real processing"