Audio engine bug - Consolidate not as per manual.

[Khazul]

From the live manual:

Consolidate actually creates a new sample for every track in the selection. The new samples are essentially recordings of the time-warping engine's audio output, prior to processing in the track's effects chain and mixer. Hence, the new sample incorporates the effects of in-clip attenuation, time-warping and pitch shifting, and of the respective clip envelopes; however, it does not incorporate the effects.

So my take is it should be like a bounce - ie perfect copy of what you hear within the limitations of the selected recording format - which if set to 32 bit - means perfect to intents and purposes.

So, if one were to take a short loop, warp it, set a groove from the groove pool, pitch it up, isolate a bit you want by simple cutting the rest off, then consolidating it back into a loopable length (ie a bar), then it should sound exactly like the original pre-consolidated version?

The following is a short bounce of the project linked below looped 4 times.
consolidatetest.mp3

What you hear at follows:

3 tracks - a conga sample to indicate the beats - 128 bpm.
Two bars of original warped and sliced sample.
Two bars of consolidated sample phase inverted - allready listening carefull you might hear a tiny timing displacement and certain some distortion.
Four bars of alternating
one bar summing the warped clip with another copy in another track phase inverted - it cancels as it should leaving just the conga
one bar summing the warped clip with the consolidated clip phase inverted - it it doesnt cancel, in fact the differences are huge.

This whole 8 bars is repeated 4 times.

The following is a test project.

Relevant Preferences set as follows:
Record format - 32 bit wav
Warp short - auto
Warp mode - beat

consolidatetest.zip

In short - consolidate seems ******* useless for this kind of thing (should it work or not? Manual seems to suggest it should be) - back to bouncing.
I havn't bounced the entire track that contained this sample yet, so dont know if it affects the export - one might expect consolidate and export to use the same code, but whos knows.

Ive known for ages that consolidate wasnt quite right, just not had an example that was as blatantly bad as this.

Report sent though probably in a black hole, posting here so other are at least aware and so know to use z manual bounce when appropriate.

[Tone Deft]

I'd have to try the .zip file to really see what you did.

one thing I'd note is that it doesn't read like it's a bounce, the new copy is out of the engine and not out of the track. a bounce is the output of the track, consolidate seems like a bounce out of the interior of the track, pre-FX etc.

unfortunately manuals aren't written by the engineers that really understand this stuff. I'd give the Ableton manual a C, so many topics are left out and the white paper is meh.

thanks for the interesting thread.

[kb420]

I listened to your audio example.

I downloaded and opened your Live set demonstrating the phenomena.

You are absolutely correct. There is definitely going on there and I can't figure it out. I even tried to consolidate the original file and I watched as the waveform changed. There is no way that the two would ever "null". You may be on to something. Theoretically, based on what the manual says, it should null perfectly.

[kb420]

one thing I'd note is that it doesn't read like it's a bounce, the new copy is out of the engine and not out of the track. a bounce is the output of the track, consolidate seems like a bounce out of the interior of the track, pre-FX etc.

If it's not a "bounce" then what exactly is it???

In this example, if you aren't using anything in the effects chain or mixer, then why wouldn't it be considered a bounce, and why doesn't it "null"?

[Khazul]

a bounce is the output of the track, consolidate seems like a bounce out of the interior of the track, pre-FX etc.

Unless its a 'pre-fx' bounce

[kb420]

Unless its a 'pre-fx' bounce

So in this case, it is a pre-fx bounce? Right??????

[Tone Deft]

a bounce is the output of the track, consolidate seems like a bounce out of the interior of the track, pre-FX etc.

Unless its a 'pre-fx' bounce

of all the users on here I trust you more than most to do this thoroughly and correctly. I just saw some weirdness in the wording.

this will be cool, I love hearing and playing with the weird after effects from wonky audio processes.

[Khazul]

I even tried to consolidate the original file and I watched as the waveform changed.

The waveform will nearly always change during consolidation due to the normalising that occurs. In this case i would expect a huge visual change anyway - its in beat mode and the sample is pitch shifted which means it will get shortened. The groove in use has a quantise factor, which with the warp will shift the timing etc, so thats all fine, but still I would think the end result should be audibly identical to the real time audio output and not audibly time-shifted and distorted.

It seems a sampling of other similarly chopped up sample loops have suffered a similar fate in consolidating in this project - so looks like loads of manual real-time pre-fx bouncing instead - grrrr

[kb420]

I did realize that the waveforms wouldn't look the same after I made my last post. I was going to edit my post, but I was waiting for some other users to try out the set.

I experimented a little on my own. I took the unconsolidated data and consolidated it. Then I turned off the Groove for that clip, as it doesn't default to no Groove after consolidation (which I think it should theoretically), and it didn't matter. I also turned off the warp (which I think should automatically be off also once you consolidate), but no dice. It still didn't null, and I can't understand why it doesn't.

[Khazul]

Interesting - OK, some more:

Loop over second half of bar 7.

Now remove the groove from both the audio 2 and audio 3 samples. They should still cancel.
Select the entire looped half bar in audio 3 and consolidate.
Rewind play position to say beginning of bar 7 and play and listen very carefully to the first pass - A tiny HF hiss. One second and subsequent passes it cancel as it should.
However we have lost the groove, so now reapply the groove to the audio 2 sample in the looped section.
Now apply it to the audio 3 sample - end result same as the others.

Another oddity - Undo all of the above, or move the looped section to second half of bar 5 and work there.
In the sample in audio 3 select the clip and Commit the groove. End result cancels to a tinny HH type sound - its shifted it by a tiny bit I think, causing different slices to be passed to the beat mode warp engine.
If you now commit the audio 2 sample as well - they will cancel again.
If you now warp audio 3 sample, then as before - hiss on the first pass and finally cancelling.

If you take a copy of one of the other warped samples from audio 2 and paste in, then it also nearly cancels excepting for an artifact similar to that of groove commit on audio 3 only.

So seems maybe a non transparent (but close enough for my purposes) workaround is to do various steps separately as evidently warping does not deal with grooves properly and/or ignores them - seems it cant make its mind up what to do TBH.

The fact that commit isnt transparent when combined with beat mode warping isnt entirely surprising (at least not without significant over-sampling which AFAIK live doesn't do), but its probably close enough - so long as the manual includes a suitable warning (as for many producers its common practice to duplicate samples into another track for time selective fx etc) and I could easily see this being another audio quality type turn off if they dont understand whats going on.

However what does need fixing is warping when groove is applied, ie - decide whether to account for the groove or not and exactly how. To be consistent with midi consolidate, then should work as follows:

Remove groove
Consolidate Warp
Copy in warp marker and transient markers from original clip(s) and move them to the equivalent positions in the new clip.
Re-Enable to groove. (or another way to look at groove - consolidate just ignores it as per midi).

However to maximise resolution of processing when consolidating grooved clips, then an option in consolidate should allow for committing of grooves at the same time - ie a feature add to consolidate (useful for midi too as I often consolidate midi and commit it before exporting clips to something else for whatever reason.)

[kb420]

Interesting. Very interesting. I really haven't used the "Commit" yet. I actually wanted to start my own thread because I have questions about it in reference to the "Random" setting. If you "Commit" a groove, then that kills the Random parameter, right???? I mean, once a groove is "Committed" to midi or audio, that clip's setting should be "None" and it should sound just like it did before being committed, but at that point, nothing will be Random anymore, theoretically.

Right?

[Tone Deft]

the clips in the "3 audio" track in measures 6 and 8 aren't even remotely the same as their supposed dopplegangers.

after being online all day, I mean after being at work all day I'm pretty beat. I'll mess around with consolidate, the groove pool and warping a bit tonight.

hopefully the people at Ableton will make use of your write ups, at this point I'm just too tired to follow them instruction by instruction.

from an end user perspective, drop some track markers with labels to show what each means and loose the click track, although it is cool to hear what other people use.

whatever this is I smell an addendum to the Live audio engine white paper.

so where are all the haters at? it's a ghost town in this thread.

[macmurphy]

so where are all the haters at? it's a ghost town in this thread.

it's been presented in an intelligent and none confrontational manner. i'm certain that has much to do with it.

@Khazul - very interesting stuff. keep it up

[kb420]

Did I miss something?

We all took a look at this, and something is definitely wrong.

Did Ableton chime in to give us some insight, or did I mis something????

[Khazul]

Not a squeak (bug report sent along with link to this thread).

Anyway - doesn't really matter - not stuck on it, now know not to waste time again trying to use this feature, and TBH I always knew it was off anyway, though the differences I've come across before (obviously not using a groove I guess) were barely audible even in isolation.

That said, I did do all the stuff that needed doing via the way that works(-ish) - ie commit grooves, then consolidate, buts been done across lots of chopped samples, hacked up loops etc and the end resulting mix seems to have lost something in the process (dull transients), so looks like I'll have to go back to an old save and manually pre-fx bounce a load of tracks instead - grrr!

A good lesson to stick with the way of doing stuff that you normally use and you know works - ie manual pre-fx bounces. Probably Live's revenge on me for trying to prep it all for export to Logic