Sample Rate Converters: Do DAWs actually "sound different"?

Please stop worrying. We talked about sample rate conversion, not “sound” of the DAW. Once again: there is no “sound” of a DAW, either it’s transparent (like all DAWs in the market) or not.

I bet he would have said the same only few years ago when PT had 24-bit fixed point audio engine against Cubase’s (scientifically superior) 32-bit floating point engine.

In case of Cubase, we’re talking about -120dB (and not of full-scale, but current signal level) or less.

There’s not “buildup” in parallel processing. It’s just the old “stacking myth”. If you have 1% of 100 (1) plus 1% of 100 (1) it’s still 1% of 200 (2). “Buildup” or “stacking” is just as stupid as thinking that when recording to 1" tape you get twice as much distortion as recording to 1/2" tape … or four times as much as recording to 1/4" tape … don’t even ask how much better C-casette would be compared to 1" tape if you believe in “buildup”.

I bet he would have said the same only few years ago when PT had 24-bit fixed point audio engine against Cubase’s (scientifically superior) 32-bit floating point engine

actually he mentioned that now protools has better bit processing or whatever, but we now have Pro on cubase 8 name so we are good… but seriously oits all another topic and i heard perfect sounding tracks made on cubase too.( i guess more professionals work on protools for mixing for many “real” reasons and historical reasons so probably those pro engineers sound better anyway and its the best commercial for protools

all this numbers don’t say that much to me’ as im not into it technically, but i presume it matters eventually to have as good as possible SRC and audio processing in general. maybe its only me but i do feel lot of times that its hard to get full pleasant sound out of cubase when mixing

To wrap it up from my point of view, pending any opinions that may be posted to the contrary of course :slight_smile::

Cubase’s sweep plot may be visually “uglier” in terms of alias banding, but they are so low in amplitude that they are probably sonically irrelevant*.

The Cubase Sweep plot (the 3rd plot down in my post above) does look pretty ugly, chock full of alias bands. But I hadn’t noticed the scale on the right (which you have to go to http://src.infinitewave.ca/ to see) - in the audible range (20 kHz or lower) they are all attenuated by 90 dBFS or more. This is consistent with the transition filter plot (the 1st of the four plots in that same post) - at 2 kHz above Nyquist (i.e., at 24 kHz), the attenuation is at least -90 dBFS, so we know that all alias band reflections in the audible zone will be of that amplitude or smaller.

  • The only potential caveat is perhaps summing all the alias bands below 20 kHz might bring them up to an audible amplitude. I don’t know how to do that math, but my gut tells me not … but can anyone look at that plot and come up with a more quantitative answer?

Close-mic’d, and recorded at high levels: tambourine, cymbals, high strings …

My voice? :laughing:

Now that I have read about all this, it worries me too. UAD-2 I’m confident will be doing it right, besides, they are adding so much distortion to “match” the classic hardware consoles, any aliasing artifacts are already built into the final product’s “desirable sound”.

But others, I do worry about like you say. I just checked, the Melda plug-ins I just bought do up-sample, but only when that option is chosen, so that is OK since I don’t. I suppose if I ever do want to upsample, I’ll need to listen very closely to make sure there are no artifacts (sine wave I guess would be the best test?). Voxengo I think is the only other plug-in I have that might, and their R8 SRC is supposed to be pretty good. Oh, and Soundtoys, but it’s an effect (Microshift), so like UAD-2, that’s OK too.

Plugins oversampling are always multiples of the DAW sample rate so I don’t think you have to worry so much

It’s something like:
20*log10(10^(dB1/20) + 10^(dB2/20) + … +10^(dBn/20))






Nop, you have to pass thru so many devices that for sure won´t go any up of 18Khz~20Khz (check mic specs, DA/AD specs) also, anything that high needs SO many spl to be heard.

Also probably your monitors can’t reproduce anything above 20Khz, at least you have something like a sonic solid speaker (like the ones that have medical uses)

I think what Mr. Jarno might be saying is that though the subtraction band might be present well down in the audible frequencies (for example an alias band at 19 kHz combining with true signal at 15 kHz resulting in a subtraction band well within the audible frequency range at 4 kHz), it would be swamped: by true signal, as well as non-alias associated subtraction bands generated from the true signal … unless the amount of true source signal down there is so sparse that the alias-generated subtraction band is able to stand out.

So in that “strange” situation (to use his term), aliasing might cause audible artifact - not from the alias band itself, but from the resulting combination subtraction bands being in the audible frequency range (the subraction bands that result from the alias and the true signal’s frequencies).

Maybe?

And as far as the analogue hardware … I’d think decent mics/speakers/ADDAs would do fine up to 20 kHz, but are you saying they generally wouldn’t?

(And I still wonder if close-mic’d sources such as cymbals, percussion, etc. (i.e., instruments without much natural low-to-mid frequency content) when recorded at high levels might result in that “strange” situation where the alias-generated subtraction bands are audible.)

Thanks -

I’m quite rusty to be honest about this (long days since my classes) but I think that is basically intermodulation distortion.

Take a look at this: http://productionadvice.co.uk/high-sample-rates-make-your-music-sound-worse/ good reading I think!

Check the specs of the “almighty” Neuman u87 for instance, and see how is the response around 20K (and combine this with the energy that the source have to make)
https://www.neumann.com/zoom.php?zoomimg=./assets/diagrams/u87ai_diagrams.htm&zoomlabel=Diagram&w=878&h=295




This reminds me a discussion or question about recording cymbals that said: “Since the source moves while sounding, it will create phase problems. How to avoid it?”

Sometimes there are things that could happen and shouldn’t now worried us too much if our ears tell us that everything sounds OK.

Cheers!

I don’t worry at all. My comment about this was just to put things in perspective: nobody worries about plugin oversampling, but there is entire website showing SRC charactristics while probably oversampling plugins do worse job than worst of the DAW SRCs.

And once again: it doesn’t matter if up/downsampling uses multiples of sampling rates. It’s still exactly the same job as with non-multiple sample rates (except in upsampling, where using multiples (2x) saves you 50% of the work; with downsampling it doesn’t).

Stop worrying! It’s not a big deal. If it were even a minimal problem, there would be hundreads of posts about this “issue” in gearsnobs.com

At least percussion instruments are not an issue. They don’t even produce harmonical audio content. They produce noisy sound. Little bit of IM distortion doesn’t make any difference there, it will be masked out by real audio content. High strings is about the “worst” possible natural instrument possible to create this “problem” … but I wouldn’t worry about them either. What I was referring to in my post was something like high-pitched (base frequency over something like 8kHz and a waveform producing lots of harmonics) synth oscillator stuff played in solo (no other instruments present). My point was only to tell there’s theoretical possible condition where you may hear this ultrasonic->aliasing->IM distrotion artifact, but in real life … nah … it doesn’t happen with any natural sound source and not even with any synthetic source which any human would consider “music”.

Exactly.

DAW projects at sample rates other than the native rates of samples – typically 44.1k – will have to real-time upsample all sample streams, putting much more aggregated reliance upon the DAW’s SRC quality, as well as increasing CPU usage.

However, as you write, it all depends upon the levels of the artefacts.

Perhaps I might consider using our RX4 for future SRC.

this is a bug or feature programming cubase?,it’s time to fix it, more than 10 years cubase audio engine… stop marketing it’s time to switch from “creative First” to a “stability and quality first”
interesting to hear the official support?)

Fix what? I can’t see anything is broken:
Distortion figures: not so untypical to any signal processing done with 32-bit floating point calculations. You probably do 32-bit float processing all day long with some of your favourite plugins and don’t complain.
Anti-alias filter: Yes, a bit shallow one, but still more than 60dB attenuation in frequency range which matters. Nothing to worry about. Steeper filter would just introduce more pre/post ringing.

How can you fix something which isn’t broken?

Not broken of course, but filters not as steep/good as Wavelab 8.5 Crystal Resampler, for example (by my understanding of the plots at the src site linked to in the OP).

Are you saying that Wavelab has more pre/post ringing than Cubase? I’ve never used it, so I don’t know myself, an honest question!

Go and see yourself. Go to comparison site, choose Cubase & Wavelab and choose impulse plot.

You see, there’s no free lunch when it comes to filters. Steep filter introduces more ringing. Does this produce audible artifacts with filters like the one found in WaveLab? Probably not. But then does Cubase’s shallow filter produce audible artifacts? Same answer.

It’s a compromise you decide when you design a SRC. If there were a perfect filter everyone would use it. These are not trade secrets. Everybody in the industry knows how filters behave.

If you are interested on the subject go on and explore plots in the comparison site. You’ll find for example:

  • Steeper the filter, more pre/post ringing
  • If no pre-ringing, phase response is compromised and there’s lots of post-ringing
    One nice example is r8brain (Free 1.9 compared to Pro Minimum Phase Ultra Steep) where you find different settings on the same product produces very different characteristics on the anti-alias filter.

to sum up this thread – “i don’t really understand those graphs but something tells me there is something scarily wrong with cubase’s sound”. :slight_smile:

read the faq/help on the site.

i’d also recommend reading the faq/help on the site, did i mention?

(i use r8brain for conversion btw.)

I haven’t read the thread from top to bottom in one sitting, but though some may feel that way, I don’t have the impression most do.

There are very accomplished and well-respected engineers who voice the opinion that the differences in SRC noted in this thread can make a difference to the final sound. That some accomplished and well-respected engineers on this thread/forum feel differently doesn’t make the opposite view one to belittle, IMHO.

For me, I think I’ve understood the graphs for some time now, at least as far as the faq/help on that site can help me do so, along with some independent reading. As to whether the differences are audible or not … I personally can’t say, and in the end, I’m happy enough with what I’ve got … I don’t think the SRC is the weakest link in my chain, to put it mildly!

I’m grateful to everyone who contributed in a helpful way to this topic! :slight_smile:

I’ll try to give a simple answer - Yes, DAW’s sample rate converters actually sound different. Now the difference in most pro DAWs is not huge but it’s audible. You can hear it easier in the high frequency of complicated full mixes or lush synth sounds. For resampling of high quality solo tracks before mixing, most DAWs are good choice and you can compensate resampling artefacts through the mixing. Resampling is critical at final stage of mixing or mastering. I think the Cubase and even Wavelab SRC must be upgraded in future updates. I made tests and personally prefer Voxengo r8brain PRO for it’s more natural sound and clearer attack in high frequency.
And for DAW’s sound engines, I hear very subtle difference in different versions of Cubase for example. Maybe the difference is in the way of mixing tracks. I think Cubase Pro 8 sounds a little bit more soft, warm and analog like than Cubase 7 and I like it!

A sample rate converter that sounds more analog?
heh.