Another thread about summing in scope

[Psilion]

I tried different ASIO drivers including float and record in 32 bit float, but having it running straight trough the scope mixer just sounds clearer. If I send them back to scope on separate channels it's a bit better but still not the same.

Could you please clarify this for me?

Basically what I usually do is run bass and kick as live devices in Scope.
Sometimes I have wanted to record them down as audio and play them back from Ableton instead.
If I record them down to 32 bit float and play them back (unstretched and unprocessed of course) I can do that either by mixing them in Live or by sending them on separate ASIO channels and mixing them on the Scope mixer.
Regardless of which of the methods I use I cannot get the same sound as when running them live, which doesn't make sense to me considering that as far as I understand in both cases the signal stays 32 bit float all the time and should be identical.
I have not written much audio software myself, and by no means claim to know everything about digital audio and DSP, but the result completely contradicts any knowledge I have.

Also I guess that the difference I perceive when sub-mixing the kick and bass on the scope mixer might have more to do with the fact that I'm using a sub mixer rather than actual differences in the summing, but the effect is quite noticeable in my ears and it's fine with regards to work flow so I stick with that approach.

[FrancisHarmany]

I tried different ASIO drivers including float and record in 32 bit float, but having it running straight trough the scope mixer just sounds clearer. If I send them back to scope on separate channels it's a bit better but still not the same.

Could you please clarify this for me?

Did you try VDAT ? It would be interesting to know if it makes a difference.

[astroman]

what Psilion describes is exactly the same phenomenon I mentioned a couple of pages back.
Like me he doesn't have an explanation and we even agree that it contradicts everything one learned - it doesn't make sense, but it's happening.
It's not appreciated either, as it disturbs the workflow, noone would volunteer for such a placebo ...
The 'difference' is mentioned by almost everyone using VDAT (and doing a comparison).
Afaik VDAT just records the datastream travelling to and from the DSPs without conversion - so it's as 'live' as any other source within Scope.

cheers, Tom

ps: of course it's not night-and-day, but just a slight difference - and most likely not related to 'summing'

[ChampionSound]

yes, and I read something similar in some thread here. It was mentioned that the problem could be that scope's asio drivers, including the 24bit ones, aren't the best at converting audio to another platform through ASIO.
I never have tested if the same happens when routing the ASIO channels from a sequencer app. back into scope.
Is the signal deterioration only from scope to seqencer (ASIO dest)? If it is from sequencer to scope too (ASIO source), you'll lose quality/definition even before the scope mixer. That's not likely to be the case because some people prefer to mix in scope as they experience it sounding slightly better, including me, so the problem could be the ASIO destination drivers only. Hopefully to be fixed in V5 (if the problem ever was recognised by SC )

[Sounddesigner]

what Psilion describes is exactly the same phenomenon I mentioned a couple of pages back.
Like me he doesn't have an explanation and we even agree that it contradicts everything one learned - it doesn't make sense, but it's happening.
It's not appreciated either, as it disturbs the workflow, noone would volunteer for such a placebo ...
The 'difference' is mentioned by almost everyone using VDAT (and doing a comparison).
Afaik VDAT just records the datastream travelling to and from the DSPs without conversion - so it's as 'live' as any other source within Scope.

cheers, Tom

ps: of course it's not night-and-day, but just a slight difference - and most likely not related to 'summing'

I think Psilion hit the nail on the head especially saying there maybe other variables. When i was on gearslutz forum i read several posts from the developer of Metric Halo and some of the things he said if i recall correctly is that a simple null test may not reveal the problems a mixer has. Also he stated that tho 32bit should be sufficient that when you build a mix-engine there are many places you can go wrong as a developer that can cause problems, wich suggest its not as simple as many think. Also i wonder if different hosts may communicate with the soundcard differently even tho they may sum accurately. I think there may be many variables that can cause problems and if so justifies in one trusting there ears. But even if i am wrong on quality differences i hear in different hosts i still am justified in my subjective view if it works for me. I don't state it as fact or teach it in some college as that, just as my belief. Ultimately my ears is going to have the final authority in most situations not because i think i'm some golden-ear engineer but because comfort, inspiration and loving to have tools i trust, etc. I'll post the link to the discussion the metric-halo developer had later, and some can go question him.

[ReD_MuZe]

at the end of the day, if you take 32 channels sum them up, in several daws and they all cancel each other out on inverted phase, then you have a placebo effect (logic8, protools7, samplitude8 and cubase sx3 cancel each other perfectly).
if not, the software is not up to par, or out-dated.

for me its the workflow, and processing quality that matters. i mix in scope but not for some voodoo summing. its just another way to get people to buy silly summing mixers and other useless gear.

and if it is a placebo effect, so what? use your ears! they are the end receiver eventually. people are not FFTs, psychoacoustics is a big part of music. you should embrace it and use it to your benefit.

[Tau]

Basically what I usually do is run bass and kick as live devices in Scope.
Sometimes I have wanted to record them down as audio and play them back from Ableton instead.
If I record them down to 32 bit float and play them back (unstretched and unprocessed of course) I can do that either by mixing them in Live or by sending them on separate ASIO channels and mixing them on the Scope mixer.
Regardless of which of the methods I use I cannot get the same sound as when running them live, which doesn't make sense to me considering that as far as I understand in both cases the signal stays 32 bit float all the time and should be identical.

In this case, what might be happening is a slight delay introduced when recording audio into Ableton, shifting the wave a little bit out of phase, thus making it sound different even before it's mixed and processed.

[Psilion]

at the end of the day, if you take 32 channels sum them up, in several daws and they all cancel each other out on inverted phase, then you have a placebo effect (logic8, protools7, samplitude8 and cubase sx3 cancel each other perfectly).
if not, the software is not up to par, or out-dated.

The problem is that one have to devise a test where you are sure that the host is only summing, and then the result would only have theoretical value.
Presumably it should be fair to compare hosts using 32 bit float source files and a 32 bit float engine and assume that it will result in no bit rate conversions with dithering or truncation. Then one would record to 32 bit float and do the null test with those.
That would hopefully result in a pure summing test, and I really doubt that that could result in anything else than signals that cancel out each other when phase inverted. And even the worst of programmers could not get something like "sum=ch1+ch2+ch3" wrong.

But it has happened that they get something wrong in the process of importing 24 bit files and playing back then summing them. And I really cannot put my experiences with Orion, Reason and FL down to placebo since it follows such a distinct pattern. If I thought that cheap software must sound worse, it wouldn't matter if I played back three or 32 tracks at the same time...it would still have that cheap sound. Also I never set out with that preconception...I was keen on using Orion since I liked many things about it and was fed up with using sequencers that usually felt far to bloated for my purposes.
So IMHO the talk about differences in sound between audio engines is not totally unfounded, although I bet the will be very little difference, if any, in the DAW's you mention.

In this case, what might be happening is a slight delay introduced when recording audio into Ableton, shifting the wave a little bit out of phase, thus making it sound different even before it's mixed and processed.

If it was stereo audio and one channel got offset that would cause phase issues, but the position of kick and bass compared to the the rest of the mix should not.

[Psilion]

yes, and I read something similar in some thread here. It was mentioned that the problem could be that scope's asio drivers, including the 24bit ones, aren't the best at converting audio to another platform through ASIO.
I never have tested if the same happens when routing the ASIO channels from a sequencer app. back into scope.
Is the signal deterioration only from scope to seqencer (ASIO dest)? If it is from sequencer to scope too (ASIO source), you'll lose quality/definition even before the scope mixer. That's not likely to be the case because some people prefer to mix in scope as they experience it sounding slightly better, including me, so the problem could be the ASIO destination drivers only. Hopefully to be fixed in V5 (if the problem ever was recognised by SC )

I would have assumed that using the 32 bit float drivers would not result in any conversion with most DAW's though?

[ChampionSound]

I would have assumed that using the 32 bit float drivers would not result in any conversion with most DAW's though?

You're right, at 32 bit float there should be no audible differences, as there are no bits being truncated. But somehow scope's ASIO drivers don't "translate" scope audio to the sequencer in a totally transparent manner, not even at 32 bit float. I really don't know why.

[siriusbliss]

I would have assumed that using the 32 bit float drivers would not result in any conversion with most DAW's though?

You're right, at 32 bit float there should be no audible differences, as there are no bits being truncated. But somehow scope's ASIO drivers don't "translate" scope audio to the sequencer in a totally transparent manner, not even at 32 bit float. I really don't know why.

And you know this how?

I see no loss in transparency at 32float here - for the past 9 years.
(using Samplitude Pro).

Greg

[ReD_MuZe]

ofcouse there is a difference in resolution between 32bit float(sequencers) and 32bit integer(scope)
but in the audio path it is neglectable, and there is no single person in the world who can tell apart 32bit float recorded material and 32bit integer.

besides there are no 32bit converters so you are always listening in 24bit eventually anyhow, and thats only in your studio.

the summing occurs in the internal processing engine which is in most daws 64bit float, and in scope it is 40 bit float. by the time it gets quantized back to 24bit you will not hear any difference.

besides imho talking about summing is waste of time. i don't know a single person who only sums his mixes and thats it. a mix includes much more than summing and that is responsible to your end sound.

my daw experiment is valid but the only way you can know that is try for yourself:
take 32 channels of 24 bits (no you cant record in 32 bits there are no converters for that )
sum them up and bounce them back to 24 bits (no you cant play 32bits, there are no converters for 32bits ) in as many daws as you like.
after that invert one of the tracks and mix it with one of the other tracks - then normalize it.
results:
1) you have scilence? - its exactly the same data then
2) you have some white noise - you gome some dithering differences
3) if you can identify your music - you ither did something wrong, or accept that the software u are using is not so well programmed.

[alfonso]

When making comparisons something to remember is to check you have the same pan law. Anyway, I experienced a difference by ear, for example, between the older Cubase VST32 and the Cubase SX3, not only in summing (the difference there is massive as VST32 is very muddy in comparison, Steinberg itself claimed a much better audio engine when launching SX and phase issues have been recognized in VST32) but even in the playback of a single stereo file. I've also experienced a difference between playback through VDAT and Cubase, but there is the ASIO module making a substantial difference, probably. I'm not an expert of audio engines in any way but I would think that as there isn't perfection anywhere in the world and even less in software, those imperfections can differ between products. How much these differences affect the sound is pretty easy to say in the VST32 vs. SX example above, even taking care of making pan law equal and leaving the faders at 0, because all the other elements (ASIO used and converters and in our case Scope environment) can be exactly the same, and the difference is not subtle. The difference is not subtle between VDAT and Cubase too, which means more correctly in the first place with or without ASIO in between. Probably the best way to make a null test could be to record all these situations in real time trough the wave drivers on a different program like Sound Forge, changing daws and keeping the same asio modules to connect them, in order to keep everything exactly the same, except the element to be tested, and keeping consistency in the way the result is recorded.
This setup should give the most correct testing procedure.

[ReD_MuZe]

cubase VST did have major flaws with its audio engine playback.
it has nothing to do with summing, 1 channel on SX sounds different. thats just bad programing.

[alfonso]

cubase VST did have major flaws with its audio engine playback.
it has nothing to do with summing, 1 channel on SX sounds different. thats just bad programing.

Yes that is well known.....but it might be that programming differences still exist between other daws and there are also minor flaws instead of major ones....so it's not an abstract difference between different algos (fixed vs. float or 32 vs. 64) but the quality of the programming, which might differ and probably does....as shown in the above example and as expected from all human activities.

[ReD_MuZe]

indeed!
thats why summing isn't the case at all.

the "serious" daws all have a clean audio engine in the current versions. cubase was the last to join, in SX2.

[Mike Goodwin]

I would have assumed that using the 32 bit float drivers would not result in any conversion with most DAW's though?

You're right, at 32 bit float there should be no audible differences, as there are no bits being truncated. But somehow scope's ASIO drivers don't "translate" scope audio to the sequencer in a totally transparent manner, not even at 32 bit float. I really don't know why.

I sent an email off to SC about this issue and got a very detailed reply within a day as usual. In short if you are using the 32 bit ASIO drivers no bits are truncated and there is no difference. Bit for bit digital transfer. I would post the email but I know many companies do not appreciate people posting 1 to 1 emails in public forums. So it stays in my in box.

[ReD_MuZe]

I sent an email off to SC about this issue and got a very detailed reply within a day as usual. In short if you are using the 32 bit ASIO drivers no bits are truncated and there is no difference. Bit for bit digital transfer. I would post the email but I know many companies do not appreciate people posting 1 to 1 emails in public forums. So it stays in my in box.

on signals being synthesized or processed inside scope bits are lost. from the conversion between integer32bit to float 32bit.

however, if you record audio, you will never be able to get more than 24bits out of the signal path without extra processing, and its not like you have ever heard those "internal bits" because you are using 24bit converters at home after all...

its really not so important for mixing as it is for processing.

[Mike Goodwin]

I sent an email off to SC about this issue and got a very detailed reply within a day as usual. In short if you are using the 32 bit ASIO drivers no bits are truncated and there is no difference. Bit for bit digital transfer. I would post the email but I know many companies do not appreciate people posting 1 to 1 emails in public forums. So it stays in my in box.

on signals being synthesized or processed inside scope bits are lost. from the conversion between integer32bit to float 32bit.

That makes sence.

however, if you record audio, you will never be able to get more than 24bits out of the signal path without extra processing, and its not like you have ever heard those "internal bits" because you are using 24bit converters at home after all...

its really not so important for mixing as it is for processing.

This I am not sure about as in Ableton Live for example I can record at 32bit, or are you talking about something different?
EDIT; I should clairify that for me 95% of my recording is simply resampling or rendering down my internal synths at 32 bits.

[ReD_MuZe]

do you have 32bit a/d d/a ?

besides what is 32bit float? it has the resolution of 24bit integer :>