Timing problem with stretch markers

[mawi]

I have this strange timing problem with stretch markers:

The bass track with stretch markers is delayed. Even when I glue or freeze this track. Is this a bug?

[mawi]

I also have this stretch marker rhythm glitch. If I make a split before the rhythm glitch, the glitch disappears.
Video demo

Here the demo project to understand.
Download demo project

Also ReaPitch has timing problems. Seems like all pitch and timestretch algorithm in Reaper would cause the timing errors.

Here the demo project with ReaPitch.
Download demo project

I just found out that it is an old problem (2010).
http://forum.cockos.com/showpost.php...55&postcount=7

[mawi]

I just have the problem again that an item with stretch markers does not syncronize to the metronome. the longer the playback runs, the worse it gets. does anyone have a solution?

[mawi]

Steps to follow:
1. insert an audio file into an empty project
2. open the media item properties and set the stretch marker fade size to 0.0 ms.
3. adjust the audio tempo to the project tempo using stretch markers.
4. start playback with metronome switched on, now you can hear how the audio file comes out of the beat after a while, even though you have moved it to the beat with stretch marks.

[mawi]

the bug occurs in step 2 when you set the stretch marker fade size to 0.0 ms.

[mawi]

Why do I use a fade size of 0.0 ms? because I hear the fade otherwise.

simple example:
1. add the sine-wave audio file to a new project.
2. create some stretch markers and move them.
3. with the default fade size, you can now clearly hear annoying fades. i don't know why you need these fades at all? my suggestion would be to use less noticeable equal gain crossfades.

but it would be nice if at least the timing bug with 0.0 ms fade size, as described in post 4, would be fixed.

thank you

[mawi]

Push.

[mawi]

See post #4.

[mawi]

Bump.

[Dafarkias]

Confirm and bump.

This is actually one of the first bugs I ever found in REAPER. Never reported it though (I think).

[Justin]

Confirmed, looking into this now. This is a tricky and complicated issue depending on the timestretch engine used.

In 5.16 we changed it so that setting a 0ms fade size did not reset the timestretch engine at each warp marker, so it is ultimately dependent on the timestretch engine's accounting of time. Unfortunately it seems that some modes are better than others.

Before I go further, there is a side note: if you're using a 0ms fade, you should set the transient marker mode to "No pre-echo reduction". This is a straight bug, I'll change it to be correct for the future.

Anyway, if you want correctly-accounted timestretch with 0ms fades: it seems like your options are Elastique 2.28 Efficient or Soloist, or Elastique 3 Soloist. RBL might also work (it was doing something that wasn't technically wrong, maybe just undesirable, here?). The other modes produce poor timing, and in these instances you should use a fade (0.1ms will do, even).

[Dafarkias]

Originally Posted by Justin

Before I go further, there is a side note: if you're using a 0ms fade, you should set the transient marker mode to "No pre-echo reduction". This is a straight bug, I'll change it to be correct for the future.

Anyway, if you want correctly-accounted timestretch with 0ms fades: it seems like your options are Elastique 2.28 Efficient or Soloist, or Elastique 3 Soloist. RBL might also work (it was doing something that wasn't technically wrong, maybe just undesirable, here?). The other modes produce poor timing, and in these instances you should use a fade (0.1ms will do, even).

Cool!

I'll give those different settings a try! I never really messed around with them because I assumed that the 'pro' algorithms produce a higher quality stretch.

I did figure out about using the .1 as a workaround, which is why I never thought the 0ms issue was a huge deal...

A personal notion I've developed throughout my usage of REAPER is that I can typically get a cleaner sounding result from elastique's 2.28 (pro), with less obvious artifacts, vs its version 3 counterpart.

It would be awesome if there were to exist maybe a somewhat technical breakdown of how these different settings 'stretch' audio,or perhaps a guideline pertaining to their distinctions amongst each other, in some level of descriptive detail.

(Not trying to hyjack this thread,I just like SM's, as you can tell from my avatar.)

Signed,

Dafarkalarkle

[mawi]

Hello Justin, thank you for taking a closer look and perhaps finding a solution. Finding the optimal stretch marker setting for highest quality of the different signals is complicated and like a puzzle for someone who doesn't know his way around. after a few short tests i found the following settings where i could hear the least artifacts (like clicks or tremolo effects on stretch marker fades on long sustaining notes, or smearing of transients on percussion and drums).

• for long sounding notes on vocals, e-guitar, bass, etc.
  elastique 3.3.0 Soloist, stretch marker fade size 2.5 ms, Mode: Tonal optimized
  no clicks or tremolo effects on fades.
  
  
• for percussion and drums
  elastique 3.3.0 Pro, stretch marker fade size 0.1 ms, Mode: Transient-optimized
  every now and then clicks, strong transients
  
  
• for tracks that contain both long notes and drums/percussion
  elastique 3.3.0 Pro, stretch marker fade size 0.0 ms, Mode: Balanced
  no clicks or tremolo effects on fades, relatively strong transients, unfortunately not in sync.

not true
that's what i found out today. everything without guarantee.

[mawi]

I just made a small table where I tested élastique 3.3.0 Timestretch Engine (I don't have time for all other Timestretch Engines). The sound source was an instrumental with loud drums and bass and instruments with long tones. It turns out that the élastique 3.3.0 Pro with 0.0 ms fadesize and tonal optimized mode was the best choice for this special sound source and requirements. Although not perfect. Then élastique 3.3.0 Soloist Monophonic with 0.0 ms fadesize and tonal optimized mode comes into consideration.
https://www.dropbox.com/s/5bdqj32c12...ngine.pdf?dl=0
Maybe Justin could use that information.

edit: it actually seems that the élastique 3.3.0 Pro/Soloist in tonal-optimized mode and 0.0 ms fadesize is even in sync. not true

Sorry, it was late when I did that test and there are false testimonies.

[mawi]

Sorry, Justin, I can't help you much right now. As you mentioned before, it's very complex. The "No pre-echo reduction" setting definitely helps with some settings. But Elastique 2.28 Efficient/Soloist, Elastique 3 Soloist, or RBL with 0.0 ms or 0.1 ms didn't quite fit for me. Either the sound quality of the Timestretch Engine didn't fit, the fade artifacts or it wasn't in sync. I'm trying to adapt a foreign long sample with drums and instruments to my instrumental. At the moment I use Elastique 3 Pro with 2.5 ms fade size and Tonal optimized mode. The fade artifacts are relatively small, and it's in sync, but unfortunately the transients are not so strong. Maybe someone has another idea.

[Dafarkias]

I've only ever used two other DAW's, so I can't say that REAPER is lacking anything by comparison (due to limited experience with other DAW's), but I wish the Elastique algo's had more controls incorporated into REAPER, and included documentation for those that like to 'look under the hood.'

Of course I feel like an @55 for saying that...

[mawi]

v5.984 has been released with Stretch Marker improvements.
https://forum.cockos.com/showthread.php?t=225994
https://forum.cockos.com/showthread.php?t=225998
Unfortunately I can't see any improvement.

Dafarkias also sees very little improvement.
https://forum.cockos.com/showpost.ph...40&postcount=4

[Dafarkias]

^^^^^ Confirmed. I'm not really convinced that the 0ms issue was resolved in any of the settings, or maybe there is just an existing issue that lies beneath.

For instance, when I was trying the various settings with the 0ms fade, the timing issue rarely manifested itself until over 5 seconds of continuous playback.

...Strange

[Justin]

Fix coming soon for this! (the latest +dev builds in the prerelease forum...)

[MZe]

Originally Posted by Justin

Fix coming soon for this! (the latest +dev builds in the prerelease forum...)

Were you referring to the version v6.03?
- Stretch markers: fix timing issues with tempo ramps and/or 0ms fade-times

Because in this version (and the latest 6.04), the issue persists. With any Time-stretching algorithm selected, fade size, and mode, the stretched audio doesn't align with the waveform displayed. When I render the item with the stretch markers to a new take, I can actually see the offset between the two.

EDIT: I found the issue with my project, I had "correct compressor latency/length (recommended)" disabled in my item's source property. The stretch markers were not the cause of the lag and enabling this setting removed successfully the lag with and without the stretch markers.

[skylar_battles]

Glueing items with stretch markers at 0ms still doesn't provide an identical result to the item before gluing. The files don't null.