Live 6 versus SX 3.1 sound quality test

[Tarekith]

I set out this morning to compare the mixdowns of Live 6, versus Cubase SX 3.1. Here's how I did it:

I compared two files for this test. The first was an 8 bar segment from my new song rendered as a 24/44.1 wav file directly from within Live 6. I placed a 1 sample long pulse at the start of the file to facilitate lining up the audio files exactly later on. The Clips in this song used a combination of warp types.

Next, I rendered each of the Live 6 tracks as separate 24/44.1 wav files (in one command with Live 6's new Render All Tracks function), then imported them into SX and dropped into it's Project View on separate tracks. No SX settings were adjusted at all, the mixer was flat, pan law was -6dB to mimic Live's pan law. This was then exported as a 24/44.1 wav file mixdown as well.

In a fresh SX project, I loaded the Live 6 render, and the SX render, lined up the 1 sample pulses in each file exactly, and then reversed the polarity (phase) of the Live 6 track. Playing back both simultaneously, you could clearly hear the differences between the two tracks with the polarity reversed. Namely in the guitar and pad parts, which are quite audible when silence is what you'd expect to hear. I exported this reverse polarity playback as another 8 bar loop for you to listen to.

Clearly the two apps are different sounding. Listen to the original files and A/B them now. The SX one has clearer cymbal tails IMO, you can hear the slight reverb on them, where as on the Live 6 file this is hardly audible.

Originally, it had been my intention to compare one other file as well. I wanted to open 12 rewire channels in SX (this is how many tracks were in the Live song), and feed the outputs of each individual Live 6 track into SX's mixer this way, then render. However, I forgot that I used a combination of VST plug ins, as well as Live 6's own plug ins for this song. Since you cannot open VST effects in Live when Live is a rewire slave, and SX cannot load Live's effects ever, it was impossible for me to do this. I could have done some work arounds, but I felt the test would not be a one to one comparison then.

If you think of any other ways this test may be flawed, please let me know and I'll be happy to attempt to correct it.

Here's the files if you want to listen for yourself:

http://tarekith.com/assets/Sworl-SX3_vs_Live_6.zip

[sqook]

Interesting test. I downloaded the zipfile and ran both of these waveforms through a Fourier Spectrum analysis. Here is the graph for Live 6:



And here's the one for SX:



Pretty similar, I'd say.

[Tarekith]

I noticed the same thing using Inspector XL while doing the test. Yet still different enough that they don't cancel each other out with teh playback test. It's interesting.

[kb420]

Digital is all just 0's and 1's, but it's amazing how all of these different programs sound a little different, ain't it?

[djsynchro]

pan law was -6dB to mimic Live's pan law.

You should do the test again with all pans set to zero "mimick" is not good enough - if there is a minute microscopic minimal (get my drift?) difference tracks will not cancel out completely.

Or render a bunch of mono tracks.

Or put just one mono track through Cubase, if there is a difference it will show with one track.

If the mixer gain structure to the master is different there will be again a slight difference and files won't cancel out. But I think if you do it right you will get them to cancel out. Did you have dithering switched off in Cubase? Dithering adds noise to the audio so it would make the rendered result different from the Live render.

[Tarekith]

No dither in SX (not that it would matter if you hear the reversal file), and all pans were set to 0. Gain structure was identical in both projects, that's what's so interesting.

[djsynchro]

Interesting.
Wanna try it again with just the 1 mono track? (Just to eliminate anything that could make a difference).

[b0unce]

Digital is all just 0's and 1's, but it's amazing how all of these different programs sound a little different, ain't it?

no its not, not at all.

column A: 100111010101010101111101101010100010

column B: 011000010101000101000010101110011101


column A & column B are not the same even tho , -gasp- , the digital environment "is just 1's & 0's"

whats done with the 1's and 0's is whats important, and obviously each group of coders is going to write different mathmatical approaches to the same tasks, to satisfy their own tastes, needs, requirements etc etc...

[marky]

Pretty similar, I'd say.

Hey Sqook, are you sure those FFTs cover the entire files?

I'm thinking not, since they're only using N=4096 samples, whereas the files obviously contain much more (44,100 * 2 *17 secs). I think the FFTs listed are only covering a small part of file.

Wouldn't you need a waterfall plot to get an accurate account of the details of the files?

Unless I'm missing something (entirely possible since iit's been a long time since I looked at this stuff).

[Tarekith]

Interesting.
Wanna try it again with just the 1 mono track? (Just to eliminate anything that could make a difference).

Sure, here you go:


http://tarekith.com/assets/SworlMono-SX3_vs_Live_6.zip

Same results, just in mono

[Tarekith]

BTW, I just want to say that I'm no trying to point out any deficiencies in Live, just the difference between the two apps. I had a lot of fun using Live to make this song, just takes some getting used to

[sqook]

Pretty similar, I'd say.

Hey Sqook, are you sure those FFTs cover the entire files?

I selected all sample points in each file. N=4096 is simply the chunk size used in the FFT algorithm.

[djsynchro]

Ahem!

I loaded them up in Live 6 beta and had a look at the (new peak reading) meters...
The 2 rendered files are peaking at different volumes, so they can't cancel out.

There needs to be at least one reference point at 0db and you need to verify in a wave editor that they both hit 0db indeed, because different meters sometimes show clipping at different levels.

Wanna do it again?

[Tarekith]

Ahem!

I loaded them up in Live 6 beta and had a look at the (new peak reading) meters...
The 2 rendered files are peaking at different volumes, so they can't cancel out.

I think you misunderstand what I am doing here. I am NOT trying to make the files cancel out, I am showing that with similar gain structures, both programs render the ouput differently. One of these differences (as you point out) is the gain of the resulting files.

I can certainly normalize both and redo the test, but that negates one of it's key purposes. In fact, here you go:

http://tarekith.com/assets/SworlMono0dB ... Live_6.zip

[Michael-SW]

The waveforms doesn't look identical when you look at them in Soundforge, so something is indeed going on. They are of course very similar, but sometimes there might be extra details in one of them. What might be a smoothly rounded peak in Cubase might have a little extra dip in the middle in Live.

But I can't really hear any difference between the two normalized files, not after only a couple of listenings at least.