Minimizing elastic audio artifacts

[camshash]

Hi guys, currently working on a couple sessions where the artist wants to change the tempo of songs a couple bpm faster in mixing/post production state.

One of the songs I was able to go from 98 to 100 bpm with no perceivable artifacts and it sounds very natural. However, on another song going from 100 to 102 is resulting in a lot of artifacts/skipping sounds in the vocals regardless of whichever algorithm I'm using, and when I commit them or use rendered playback the artifacts are still present. I don't have any issues with the drums in rhythmic, bass in monophonic. It's only with the Vocals and polyphonic instruments like guitars and keys.

X-form sounds the worst by far which is weird since it is supposed to be the most natural but has a lot of artifacts, which leads me to believe I'm doing something wrong.

All tracks/buses are in tick mode and the timing/phase of everything is as it should be.

What are some other things I might be neglecting when changing the tempo of a song?

Is increasing the tempo by 2-3 bpm asking too much? Is it always a case-by-case basis?

I appreciate any and all feedback. I'm trying to give as much info as possible, I can include sound files if needed. I've used elastic audio mainly for individual tracks and loops, but never for whole songs. I'd much prefer to re-track at the new tempo than compress the audio but deadlines are tight with this one.

Running 2022.12/Ventura 13.6.3 M1 Max in rosetta

Thanks!

[1m1]

In my experience it is case-by-case. Different algorithms work for different things, so try them all.

Another option is to consolidate the clips into a single file, then use an Audiosuite plugin to change the length. Surprisingly the original old school built in time stretch plugin was really good on voice sometimes.

[midnightrambler]

Yeah I'm afraid what you're discovering there is that Elastic is definitely NOT a "set and forget". I've whinged about this countless times on here before. You basically have to go through each track and determine which algorithm sounds best, and also inspect/edit/delete the analysis markers to minimise audio "lumps". I've often found that if you're doing a global tempo shift, just completely deleting all analysis markers can yield the best sound quality results, though at the expense of super-accurate timing, so even that is a bit of a fudge. YMMV.

[Sardi]

Originally Posted by midnightrambler

Yeah I'm afraid what you're discovering there is that Elastic is definitely NOT a "set and forget". I've whinged about this countless times on here before. You basically have to go through each track and determine which algorithm sounds best, and also inspect/edit/delete the analysis markers to minimise audio "lumps". I've often found that if you're doing a global tempo shift, just completely deleting all analysis markers can yield the best sound quality results, though at the expense of super-accurate timing, so even that is a bit of a fudge. YMMV.

This.

Removing a LOT of analysis markers helps immensely. It can be quite transparent in polyphonic mode if you cull most of the analysis markers and only use them on the transient sections or starting phrases of words etc. All the ones on the body of the sound should be removed IMHO.

X-Form, contrary to what most say, is not the best. It sounds very phasey. It never was this way originally. AVID changed some setting in the algorithm somewhere along the line and it altered the way it sounds. Quite annoying.


Sent from my iPhone using Tapatalk

[skylar_battles]

Really not sure myself that removing analysis markers for this case makes a difference. I tested with 2 unstretched clips. One with all analysis markers deleted, one with sensitivity at 100%. I changed the tempo with elastic audio on and tracks set to ticks with the polyphonic algorithm.

After changing the tempo in order to stretch the clips, going into warp view reveals that the analysis markers have not even been turned into warp markers by stretching. So my understanding is it really just stretches the whole clip to fit. Like if you did with TC/E. What's odd is the 2 clips don't null and I'd think that they should. When I zoom in to max everything looks lined up at the sample level so I'm not sure what is going on here. I looped it and was switching between the two clips and I didn't notice a difference to be honest even though they don't null. Weird.

[Stephen Bond]

Lower the amount of analysis markers can help but it will also make the audio less tight in the timing department.

Ideally you need a warp marker per note and no more or less.

Stephen

[camshash]

This is all helpful information,

One other method I tried which so far seems to be very successful is using varispeed and then pitching the audio back down with Melodyne, Ive calculated that going from 100->103bpm results in a 51 cent difference.

The results sound very natural so far. Works best with vocals in melodyne melodic algorithm. I thought the universal algorithm would work best but the formants sounded a little off.

[albee1952]

Interesting as(for me) X-Form has always had the least artifacts. Serato Pitch 'n Time Pro is probably the best tool for this and they might offer a free demo that could solve your problem this time around(its a bit expensive to buy)

[camshash]

Originally Posted by albee1952

Interesting as(for me) X-Form has always had the least artifacts.

What sources are you using X-form on? Every time I use x-form it sounds very phasey and 'digital' for lack of a better word, however there are no lumps in the audio like I'll get from all the other algorithms. What version pro tools are you running? I experienced the same thing with elastique-pro in 2024.3, very phasey sounding.

Is it possible there are other settings within the session/preferences that are causing elastic audio to not work properly? I had one song pre-mix with very little if any plugins/automation and elastic audio worked flawlessly, but on a session in post-mix stage elastic audio failed

[Rich Breen]

FWIW, EA is great for tweezing an individual track to a grid or to another instrument, but I wouldn't use it for this particular task; to compress/stretch everything in a session by a specified amount I'd use one of the dedicated TCE algos on a per-track basis; Pitch 'n Time or Soundshifter are both capable of good results - I've also sometimes used Zplane Elastique (audiosuite) and Izotope Radius with good results.