DAW: Which Has Best Time-Stretch Algorithm?

[frankpain]

Other than Serato sample, which is a plugin, I think Reaper and Tracktion Waveform have the best Time stretch algorithms, FL Studio is also good, Ableton is good, MPC Software is good. Cubase sucks, Cakewalk is not too good either.

[tommyzai]

Other than Serato sample, which is a plugin, I think Reaper and Tracktion Waveform have the best Time stretch algorithms, FL Studio is also good, Ableton is good, MPC Software is good. Cubase sucks, Cakewalk is not too good either.

Serato Studio comes with the Sample algorithm. MuLab's proprietary stretching gets rave reviews.

[tommyzai]

Any recent thoughts on the best time-stretching DAW?

[soundmodel]

All of them suck to be honest.

It's just not possible to stretch audio without adding artificially sounding fillers.

You might be better off by slicing the audio manually and playing with masking in mixing if there are other tracks going on.

Granular and phase vocoder algorithms do this, but what they use for filling is less sophisticated.

[audiouser720]

Ableton Live

[Danilo Villanova]

REAPER has r8brain from Voxengo, which is supposed to sound better and use less CPU than Elastique.

[lfm]

When I ran Reaper 3.x-4.x I started a tab with a reference track beside my project.
- reference music was from a CD and 44k
- my project was 48k

When I switch to reference tab I got overs +1.6dB and wondering WTF that was
- a purchased cd with overs???

What I did not realize at first was that Reaper did not switch to 44k for reference music tab
- it resampled 44k over to 48k!!!!
- the clip with reference music was marked "resampled"

Then a light bulb moment, and I changed to most cpu hungry algos
- if it was 192 something
- overs went down to +0.6 dB from +1.6 dB
- if running reference track as 44k separately, it showed no overs of course

So this tells a story how resampling do things to audio.

I cannot say I could hear anything in my face that was bad, but there were numbers telling it is not without penalty you resample.

AFAIK Reaper uses same algos for stretching and SRC
- at least then

I reported as a bug but got no confirmation from Cockos
- rather than this must be intersample peaks and things
- this was BS of course
- I even checked with a free plugin from SSL that oversampled x8
- resampling, stretching and whatnot do things to your audio

As I read it now uses r8brain for SRC seems better. You spend as much money you can afford on preamps and everything to have as nice sound as possible. You don't want to ruin any of that with poor math somewhere.

[jrwaltb]

You can find out here: https://src.infinitewave.ca/

Reaper is better that I thought and so is Studio One 6. I think Sox and Rx is the top two with Rx being the best.

[Tj Shredder]

I reported as a bug but got no confirmation from Cockos
- rather than this must be intersample peaks and things
- this was BS of course

This is no BS at all, its a simple explanation. And as you work within any DAW with 32-bit float, it does not affect the audio quality at all. If you want to go the save way, do the resampling in 32-bit right away… But as you recognized already, you can’t hear the difference…

[jancivil]

I use Cubase's time stretch pretty much every project. The elastique pro list is all I've used at least since I can remember, and it's brilliant, flawless.

If I want more than just time 'correction' ie., to get 'creative' its formant varieties suit me, and there's "Tape" and its Formant version.
'it sucks' articulates nothing. "It's just not possible to stretch audio without adding artificially sounding fillers." is just inaccurate as far as Cubase Pro versions (insofar as my usage anyway), as most of what I do is just to make a piece of audio I'd just as soon remain committed to fit changes in my timeline. And this per realistic orchestral kind of samples most of the time.
This is to say not as regards radical or dramatic way-out-of-whack time differences, which there are plugins made specifically for.

[apoclypse]

All of them suck to be honest.

It's just not possible to stretch audio without adding artificially sounding fillers.

You might be better off by slicing the audio manually and playing with masking in mixing if there are other tracks going on.

Granular and phase vocoder algorithms do this, but what they use for filling is less sophisticated.

I wonder if AI could help here. I'd assume with enough training it would work well. I think this is where things are going.

[soundmodel]

All of them suck to be honest.

It's just not possible to stretch audio without adding artificially sounding fillers.

You might be better off by slicing the audio manually and playing with masking in mixing if there are other tracks going on.

Granular and phase vocoder algorithms do this, but what they use for filling is less sophisticated.

I wonder if AI could help here. I'd assume with enough training it would work well. I think this is where things are going.

Likely.

You could for example search for cyclical waveforms, because those are what can be repeated without sounding artificial at all. Well, up to the point where it becomes repetitive. This repetition is what the phase vocoder approach also does, but it repeats pure waveforms (sines essentially).

I've stretched some sounds this way. You just seek for a part that can be repeated without the modification sounding artificial. Or you seek a point that's a zero cross and has something nice before it so that you can add a custom filler sound that sounds like it fits.

There's some sweet spot in terms of time such that if the additions are small, then the ear will not pick it up compared to the major parts going on. Same as with masking in mix (you mainly just hear variations around the main parts, you don't pay attention to edit mistakes in small parts etc.).

[lfm]

I reported as a bug but got no confirmation from Cockos
- rather than this must be intersample peaks and things
- this was BS of course

This is no BS at all, its a simple explanation. And as you work within any DAW with 32-bit float, it does not affect the audio quality at all. If you want to go the save way, do the resampling in 32-bit right away… But as you recognized already, you can’t hear the difference…

I tested at 44k and the ssl oversampled plugin and there are no intersample peaks.
- that's what I meant, explanation was BS
- algos do stuff that we may accept for creativity or not

And I said, not in my face different sounding, which is not the same that it isn't sounding different.

Getting overs was enough for me, didn't listen further.
- one reason I dropped Reaper as main daw
- src for export would use same algo then

But good if that now changed to r8brain doing src when doing export.
- so it was recognized as maybe not top solution to use stretch algos for simple src

[Tj Shredder]

These inter sample peaks are something to beware of in general. That is why its better to never normalize to the peaks, but around half a dB below. Your DAC will produce them anyway, you can’t avoid that. This has nothing to do with a DAW doing something bad. To drop Reaper for that reason is a bad decision, its not Reapers fault. If it would lower the volume without asking to avoid those overs would be much worse!

[lfm]

This has nothing to do with a DAW doing something bad. To drop Reaper for that reason is a bad decision, its not Reapers fault. If it would lower the volume without asking to avoid those overs would be much worse!

You are in heavy denial here, sorry, but that is what it appears to be.
- and note it was about a reference music track, not for final rendering

I changed from default resample algo which gave +1.6dB overs to highest at the time, 192 something, and it fell to +0.6dB, and yet it is nothing to do with Reapers algos?

If running it at default 44k no overs was created, since algos need not do anything.

Reample algos alter audio enough to do this.