Jump to content
sreams

Elastique algorithms have serious timing issues

Recommended Posts

Posted (edited)

So here's what I have that shows this issue:

Started with a 7 minute long drum performance that was not played to any click
Tapped out the tempo onto a MIDI track while listening to match the drum performance
Used "Fit to Improvisation" to perfectly match the tempo map to the human performance
In AudioSnap, set the clip Follow Options to "Auto stretch"
Checked the box for "Follow Proj Tempo"

After that last box is checked (I have made no timing changes to the drum tracks yet), everything looks fine on the timeline. Playback is perfect with the clip set to use "Groove" for online render. As soon as I switch to "Elastique Efficient" or "Elastique Pro", playback no longer matches the visible waveform. It sounds kind of okay for the first few measures, but playback is a little fast and gets more and more ahead of the click as playback progresses. Again... the visible waveform looks fine and matches beats/measures perfectly. Switching back to "Groove" or "Percussion" completely fixes the issue. It makes the Elastique algorithms pretty much unusable.

I can upload trimmed-down project to show this behavior if it is helpful. It is 100% reproducible.

Edited by sreams

Share this post


Link to post
Share on other sites

Radius Mix works and sounds great, but is not available for realtime processing. The issue I'm describing affects realtime playback. It means "Groove" and "Percussion" algorithms are usable, but both Elastique algorithms are not.

Share this post


Link to post
Share on other sites
21 hours ago, sreams said:

I can upload trimmed-down project to show this behavior if it is helpful. It is 100% reproducible.

If you can upload examples where the algorithm isn't working we can pass this along to zplane.

Share this post


Link to post
Share on other sites
1 hour ago, Jonathan Sasor said:

If you can upload examples where the algorithm isn't working we can pass this along to zplane.

Too big to attach here... so here you go:

http://www.mistercatstudios.com/mtest.zip

1) Open the project. It contains one audio track of an overhead mic from the drum kit. The tempo map has already been adjusted to match the human performance. No timing changes have been made to the audio at all.

2) Check playback from various points along the timeline. The metronome should match the drums perfectly.

3) Select the clip and open the AudioSnap Palette. Change the Online Stretch Method to "Elastique Efficient" or "Elastique Pro".

4) Check playback again. The further along the timeline you go, the further away the drums get from the metronome.

Share this post


Link to post
Share on other sites
13 minutes ago, Craig Reeves said:

What Cakewalk needs to do is just simply allow users to use elastique in the same way we would normally use Groove-clip looping (ALT+7) instead of having to go through AudioSnap which has essentially been broken for years. Audio Snap seems to try to do too much and "overthinks".

I don't agree to that! I have tested several audio stretching implementations of other DAWs and also read about some more and IMHO they don't suit for professional work or only restricted to certain material (sometimes even confirmed by their DAW gurus)!

The problem is that all of them that detect the transients automatically make a lot of incorrect detections and there you must have the possibility to adjust this like with AudioSnap! You must be aware that there is no algorithm to detect the transients absolutely correct, because there are instruments (like vocals) that allow to choose the transients in various positions to sound "correct".

So all those flashy programs like Ableton that do it the "easy" way, they just cheat you! Because if you have wrong detections and no possibility to adjust it, the stretching is not usable at all, unless you allow the performance being destroyed! And some other programs (like Reaper) even lack a "real" stretching functionality. They provide only a very basic stretching support that is really labor-intensive!

Back to AudioSnap: Aside from its brilliant design I agree with you that there are some issues (and I still hope it is being fixed one time), but they may be caused more by the stretching algorithms and by their application.

Share this post


Link to post
Share on other sites
Posted (edited)
2 minutes ago, marled said:

I don't agree to that! I have tested several audio stretching implementations of other DAWs and also read about some more and IMHO they don't suit for professional work or only restricted to certain material (sometimes even confirmed by their DAW gurus)!

The problem is that all of them that detect the transients automatically make a lot of incorrect detections and there you must have the possibility to adjust this like with AudioSnap! You must be aware that there is no algorithm to detect the transients absolutely correct, because there are instruments (like vocals) that allow to choose the transients in various positions to sound "correct".

So all those flashy programs like Ableton that do it the "easy" way, they just cheat you! Because if you have wrong detections and no possibility to adjust it, the stretching is not usable at all, unless you allow the performance being destroyed! And some other programs (like Reaper) even lack a "real" stretching functionality. They provide only a very basic stretching support that is really labor-intensive!

Back to AudioSnap: Aside from its brilliant design I agree with you that there are some issues (and I still hope it is being fixed one time), but they may be caused more by the stretching algorithms and by their application.

As I pointed out, Groove-clip stretching works perfectly fine. It doesn't need to find the transients. What needs to happen is that if you increase the tempo by 6%, the audio should be stretched accordingly without having to manually slip-stretch each one of them yourself.

Edited by Craig Reeves

Share this post


Link to post
Share on other sites
2 minutes ago, Craig Reeves said:

As I pointed out, Groove-clip stretching works perfectly fine. It doesn't need to find the transients. What needs to happen is that if you increase the tempo by 6%, the audio should be stretched accordingly without having to manually slip-stretch each one of them yourself.

Okay, I understand now! 😉 You only want to stretch the clip as a whole, no intention to quantize audio! Sorry, in this case I have misunderstood!

Share this post


Link to post
Share on other sites
9 minutes ago, marled said:

Okay, I understand now! 😉 You only want to stretch the clip as a whole, no intention to quantize audio! Sorry, in this case I have misunderstood!

Exactly.

Share this post


Link to post
Share on other sites
31 minutes ago, Craig Reeves said:

Exactly.

These are all valid points, but kind of off-topic. I'm referring to a very specific bug that occurs with Elastique algorithms only. The usefulness of AudioSnap is a separate conversation that deserves its own thread.

Share this post


Link to post
Share on other sites
16 minutes ago, sreams said:

These are all valid points, but kind of off-topic. I'm referring to a very specific bug that occurs with Elastique algorithms only. The usefulness of AudioSnap is a separate conversation that deserves its own thread.

You're correct. I'll delete the post.

Share this post


Link to post
Share on other sites

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now

×
×
  • Create New...