FREEZE tracks sound different to Live playback

[MIAR]

I'm freezing a VSTi Organ. So there is no WARP involved although the newly Frozen version will not allow any file manipulation to check this.

This is very concerning. I also supersized at the lack or replies on SK's post from last year. If FREEZE is not indentical (and from what I hear, can be significantly worse), this should be a pretty big issue for this DAW.

[Stromkraft]

I'm freezing a VSTi Organ. So there is no WARP involved although the newly Frozen version will not allow any file manipulation to check this.

This is very concerning. I also supersized at the lack or replies on SK's post from last year. If FREEZE is not indentical (and from what I hear, can be significantly worse), this should be a pretty big issue for this DAW.

If you mean my post it's from about 4 months ago and I haven't contacted Ableton support about it, as I planned to do some improved re-testing. Also my workaround recording directly to a new audio track — sometimes moving the MIDI track to song folder in my User Library as backup — and moving over processing and automation works very well for me.

Admittedly freezing/flattening would be more convenient, but it helps my productivity to view the MIDI track as the performance and the audio track as the recording.

I still use freezing for when I need to free up CPU, but I never flatten since I did those tests. The difference is very very small though and I'm just being anal about it.

I have also experienced clicks in frozen tracks during a brief period when using an earlier beta version, which I did have contact the beta team about. However, those issues went away and I couldn't repeat them.

[MIAR]

Oh sorry I must of been reading a date somewhere else. But still so few see or have noticed this as an issue? Its surly not system specific and must be the DAW itself.

Did Ableton get back to you on it?

[Stromkraft]

Its surly not system specific and must be the DAW itself.

The system in my case have not been ruled out. I intend to repeat tests on a newer and faster machine. I'd say it's clear I don't understand completely how audio in Live works, but I have actually never experienced any production stopping issues on the other machines in use by my musical partners nor do I currently feel hindered producing on own old work horse.

Did Ableton get back to you on it?

We had a conversation about the clicks. I'll renew the contact when I've been able to present them with a specific case that can be repeated, but I don't have this irritating problem since a longer time.

We should note there are 2 issues, possibly unrelated, that have been mentioned:

• That frozen/flattened/exported tracks don't seem to null the track they originate from, even if the differences are commonly not audible.
• There can be audible clicks in frozen tracks under some circumstances, hitherto these circumstances are unknown.

[MIAR]

Buts its a mathematical rendering operation that needs to be carried out within the DAW. Why could the system itself be a factor?

[tedlogan]

• That frozen/flattened/exported tracks don't seem to null the track they originate from, even if the differences are commonly not audible.

Not surprising though, if one is using random phase-triggered synths for example, or any FX with very subtle random variations in sound. The lot would never sound 100% the same each time, though these differences shouldn't really be noticeable except in a null test.

[Stromkraft]

• That frozen/flattened/exported tracks don't seem to null the track they originate from, even if the differences are commonly not audible.

Not surprising though, if one is using random phase-triggered synths for example, or any FX with very subtle random variations in sound. The lot would never sound 100% the same each time, though these differences shouldn't really be noticeable except in a null test.

If you read the linked test suite that I made this spring, you'll see that test:"8. Live MIDI track with effects and recorded audio" are 2 test runs, both with the exactly same Live track, that when repeated did cancel against a recording of itself.

If your concerns, which are very valid, was a factor when I did my tests then we would have the same total cancellation result with "4: Frozen track with effects" which did not cancel out.

Of course, the effects used and their dynamic nature will affect cancellation tests, so that's a very valid concern of yours. The same is true for the signal path, that in fact in those cases where I had to use 3 tracks could affect my previous inability to make frozen tracks cancel.

I'm not claiming my tests runs were very scientific, can't be improved — I was very dissatisfied with not being able to have simple signal path repeatable in every test — or that they can't contain some fallacies that changes the picture, but I'm not a total idiot either. I did the test primarily because of my assumptions about frozen tracks being bit-identical to a level adjusted recording, directly or via resampling, or to an exported track and the fact that a handful of folks kept saying "frozen tracks sound bad". I'm not at all convinced this is indeed the case, but I do think it's healthy to challenge your assumptions.

I do think that if you read all the 8 tests that I did and assume they are about right, then you'll agree with me that flattened tracks aren't that bad, even if recording pre-fader seems like an interesting alternative. It's probably wise to export songs with tracks unfrozen too, because frozen tracks seem to not be as bit-perfect.

I'm well aware of the fact that the test runs need to be improved and repeated. One possibly more reliable method would be to use the actual frozen audio file re-added to a normal audio track to be able to compare without strange signal paths. If any of you want to contribute to set up a test suite that anyone could run, I welcome your input and suggestions in the "Finding the best way to get tracks down" discussion.

[tedlogan]

• That frozen/flattened/exported tracks don't seem to null the track they originate from, even if the differences are commonly not audible.

Not surprising though, if one is using random phase-triggered synths for example, or any FX with very subtle random variations in sound. The lot would never sound 100% the same each time, though these differences shouldn't really be noticeable except in a null test.

If you read the linked test suite that I made this spring, you'll see that test:"8. Live MIDI track with effects and recorded audio" are 2 test runs, both with the exactly same Live track, that when repeated did cancel against a recording of itself.

If your concerns, which are very valid, was a factor when I did my tests then we would have the same total cancellation result with "4: Frozen track with effects" which did not cancel out.

Of course, the effects used and their dynamic nature will affect cancellation tests, so that's a very valid concern of yours. The same is true for the signal path, that in fact in those cases where I had to use 3 tracks could affect my previous inability to make frozen tracks cancel.

I'm not claiming my tests runs were very scientific, can't be improved — I was very dissatisfied with not being able to have simple recording path repeatable in every test — or that they can't contain some fallacies that changes the picture, but I'm not a total idiot either. I did the test primarily because of my assumptions about frozen tracks being bit-identical to a level adjusted recording, directly or via resampling, or to an exported track and the fact that a handful of folks kept saying "frozen tracks sound bad". I'm not at all convinced this is indeed the case, but I do think it's healthy to challenge your assumptions.

I do think that if you read all the 8 tests that I did and assume they are about right, then you'll agree with me that flattened tracks aren't that bad, even if recording pre-fader seems like an interesting alternative. It's probably wise to export songs with tracks unfrozen too, because frozen tracks seem to not be as bit-perfect.

I'm well aware of the fact that the test runs need to be improved and repeated. One possibly more reliable method would be to use the actual frozen audio file re-added to a normal audio track to be able to compare without strange signal paths. If any of you want to contribute to set up a test suite that anyone could run, I welcome your input and suggestions in the "Finding the best way to get tracks down" discussion.

Ahh - but I am not in disagreement with you - I used to have to freeze tracks all the time on my previous old PC. Sounded fine to me. I was merely throwing that thought of random variation out there, nothing more, nothing less.

I like your dedication to detail, but I'm not interested myself in these ultra-fine variations as they have zero impact on me enjoying music or attempting to make it, so I have not read your findings. I'm sure they have merit and all that, I just don't care about these kind of things - like whether 96khz is "better" than 44khz, or mp3 vs vinyl vs CD vs YouTube etc - I can happily listen to music on my new TDK "bass boost" in-ear phones which I bought for £3.99 on eBuyer's flash sale, mainly to play pink noise tweaked to sound like the Enterprise to block out babies and others' headphones on the journey to work. Didn't take long to start listening to more than pink noise. I kinda like hearing music from less-than-ideal sources I've found.

Anyway, I'm digressing as usual - and I see I have forgotten that the OP did say no randomness in the audio source etc, so my thought was redundant. Apologies.

[Stromkraft]

Ahh - but I am not in disagreement with you - I used to have to freeze tracks all the time on my previous old PC. Sounded fine to me. I was merely throwing that thought of random variation out there, nothing more, nothing less.

I like your dedication to detail, but I'm not interested myself in these ultra-fine variations as they have zero impact on me enjoying music or attempting to make it, so I have not read your findings. I'm sure they have merit and all that, I just don't care about these kind of things - like whether 96khz is "better" than 44khz, or mp3 vs vinyl vs CD vs YouTube etc - I can happily listen to music on my new TDK "bass boost" in-ear phones which I bought for £3.99 on eBuyer's flash sale, mainly to play pink noise tweaked to sound like the Enterprise to block out babies and others' headphones on the journey to work. Didn't take long to start listening to more than pink noise. I kinda like hearing music from less-than-ideal sources I've found.

Anyway, I'm digressing as usual - and I see I have forgotten that the OP did say no randomness in the audio source etc, so my thought was redundant. Apologies.

Ah, OK. I most certainly found your thoughts enlightening and it got me thinking about this issue in another way, which I appreciate.

As I've alluded to I do think at least flattening is close enough to the original in most cases to not having to record every track that I want to make audio of. But I do share the concerns the OP expresses about clicks and other audible problems with frozen tracks.

If I learned anything from those tests it was that I don't understand the attraction of exporting to audio and putting it back into the project, nor do I understand why I should re-sample from the master when I can record pre-fader directly to audio (from several tracks as well). While direct recording runs some risk in a heavy project of recording some glitches, that's why I run some tracks frozen (before I record them unfrozen) giving the CPU a work load it can handle.

[fishmonkey]

did you guys rule out all the non-neutral operations that affect freezing that are mentioned in the Live user manual?

[dhpetrescu]

Why not just re sample instead of freezing?

I've experienced something similar, for me it was due to either:
1. Compression or a few other complex effect on the track. Freezing was only possible after inactivating these effects.
2. Like someone else mentioned, if there I a randomized modulation, the track will never be exactly the same.

did you guys rule out all the non-neutral operations that affect freezing that are mentioned in the Live user manual?

[Stromkraft]

Why not just re sample instead of freezing?

Why, when recording is usually more carbon copy? I default to 32bit projects usually. Turn off everything in the signal chain? Why would I do that when recording pre-fader exists? I suppose there's must be a reason and maybe the difference is very small. I still think it's a bit convoluted.

How is "resampling" better than recording? I can't see that it is, but I'm eager to learn.

[Stromkraft]

did you guys rule out all the non-neutral operations that affect freezing that are mentioned in the Live user manual?

I think I did for the test-runs I made earlier in the other thread. I didn't use automation, recorded pre-fader, didn't use audio files from outside of Live, defaulted to 32bit so I could avoid dithering. I might have missed something, so I'm open for suggestions.

[MIAR]

Never really apreciated what the use of "Resampling is.

SONAR has a "Bounce to track" option and most other DAWs do to. Why do we still need realtime playback or freeze/flatten or export to achieve the same.

[Stromkraft]

Never really apreciated what the use of "Resampling is.

SONAR has a "Bounce to track" option and most other DAWs do to. Why do we still need realtime playback or freeze/flatten or export to achieve the same.

Exactly. Except that besides real time playback it doesn't seem to be the same. My biggest disappointment with my test runs was that exported files were totally off compared to frozen/flattened. Maybe it's possible to get exact exports with some set up, but I failed.
Hopefully I did something wrong, or any issues have been fixed since. Right now I can't retest as I don't have the time.

In my current project I use a combination of recording MIDI tracks and freezing/flattening. Automation loss is the biggest issue I've had so far, but I've found workarounds in most cases and instrument automation is acceptable to lose of course. I continue to be sceptic to exporting and re-sampling, but will look into those methods again at some point.