Blind A/B Shootout: Upsampling Your Mix

[Funkybot's Evil Twin]

Ok, since no one posted a serious comparison in the other thread and I got sent home from work early due to strep throat, here it is, the A/B test we've all been waiting for.

First, here's what I did. Took 7 stems from a song I had (drum overheads/rooms, drum directs, bass, guitars, keys, lead vox, and background vox), and exported them as 24bit 44.1khz wave files. I turned off all bus effects, and all lead vocal effects prior to export. Then I used Audiomove to upsample the mix from 44.1khz to 88.2khz. I then created a new 88.2lhz project with the stems, mimicing the routing I used in the previous project as well as the effect presets. I did some additional mix tweaking, and when I came upon a good result exported the new mix to 88.2khz, 16bit. I then used R8brain Free to convert from 88.2 to 44.1 (I used R8brain instead of Audiomove on the downsampling because Audiomove was reducing the RMS by about 1db). Next, I took that EXACT same project, deleted the 88.2khz files, changed the samplerate to 44.1khz, and reimported the original 44.1khz stems. I then exported the 44.1khz stems as 44.1 (obviously), 16bit. Finally, I opened up a wave editor, shredded off the same piece of the song from each mix, and put them next to each other (I'm obviously not telling you the order of which one is first).

Now, I firmly believe that this is the best way this A/B comparison can be done, as there is no post-tweaking (like Bduffy says he does) on one vs. the other. Other than the samplerate, these are the exact same mixes, same effect settings, etc. I think the difference is actually quite clear (using an RME Fireface800 w/ ADAM A7's, so a fairly good monitoring environment), but lets see what you guys think. Sorry for the super long post, but I wanted to get every detail in so no one accuses me of screwing up the blind test. If anyone wants to take the file from sendspace and host it elsewhere for conveniences sake, please do so. Here is the link to the file (8MBs, uncompressed .wav):

http://www.sendspace.com/file/8s7pq1

UPDATE: Unmastered version (same file order) here:

http://www.sendspace.com/file/6sn71c

[bduffy]

Ooooh...a poll...wicked, downloading now!

[Funkybot's Evil Twin]

How'd I know you'd be first to post in this thread .

[Kingston]

I used R8brain instead of Audiomove on the downsampling because Audiomove was reducing the RMS by about 1db

are you quite sure your measuring equipment is up to the task? it doesn't do that.

anyway, the difference I would say is negligible and totally not worth it in this case. I don't know what plugins you're using but it's quite clear they don't benefit from this oversampling ordeal a whole lot - if at all. Your source material is also a bit questionable and - how should I put it - "china sound". The harsh transient quality pretty much masks all the areas oversampling *should* be improving (imaging especially).

no offence but 96khz isn't really useful for a home demo sound. sorry if I come across as bit harsh (pun very much intended) but for a sound like this you might as well stick to 44.1khz.

[Joe Simard]

whoa a bit condesending there... lets be happy

[Kingston]

just being brutally honest.

[Joe Simard]

exactly. brutality: not nice.

[bduffy]

Hmmm...I'll probably have to reserve judgement until I get home; I'm on the cans here at work. That's weird that you experienced a -1dB drop! I've been doing conversions all weekend, and the converted files are 100% exact; you simply cannot tell the difference.

[Funkybot's Evil Twin]

are you quite sure your measuring equipment is up to the task? it doesn't do that.

I did it twice (couldn't believe that it wasn't just user error the first time around). I could post a screenshot of the files if you want, as you can see the difference. It dropped the volume by about 1.1 dbs. Here's what I had, and what I did, opened the 16bit 88.2khz wave file (which was at the correct volume), selected the output format as 16bit 44.1khz, and converted. The converted file was 1.1db's quieter.

Also, I'm not going to take offense about the sound, I know it's a quite harsh (this was recorded through the converters in my M-Audio Omni, prior to the RME purchase). On the same front, the previous mixes I did weren't anywhere near as harsh, as they were considerably darker/less compressed but I wanted to try one bright and loud (90's rock-ish). It also didn't sound quite so bad until I started clipping the hell out of it with T-Racks, but I wanted to see what happen in an extreme environment. I still think the difference can be clearly heard though.

BTW, did you vote?

[JeffreyCreel]

brutally honest, honestly brutal. Does it really matter?

[bduffy]

FET, I cannot tell. Sounds the same to me. I'll check again at home, but I don't think I can vote atm...but I will later.

What EQ's, compressors, etc were you using? Just wondering if we can establish if they benefited at all from the oversampling.

[Kingston]

I did it twice (couldn't believe that it wasn't just user error the first time around). I could post a screenshot of the files if you want, as you can see the difference. It dropped the volume by about 1.1 dbs. Here's what I had, and what I did, opened the 16bit 88.2khz wave file (which was at the correct volume), selected the output format as 16bit 44.1khz, and converted. The converted file was 1.1db's quieter.

it's still a measuring or user error somewhere there. I just checked it and wavelab global audio file analysis tells me they are *always* equal. None of the numbers changed on any conversion direction or amount, either peak or RMS.

what do you measure with? I suggest you ditch it.

BTW, did you vote?

I didn't. I can't discern a favourable difference between either of them. If I had to guess you've done no drastic EQing in the mix and there are probably no saturation/distortion plugins there at all (which benefit the most), and very little compression - which makes the whole ordeal just too subtle.

(well I think/guess/imagine the first part of the audio file is the oversampled one, but it's very subtle and not quite the difference I'm used to hearing with these things.)

[Funkybot's Evil Twin]

Here's the exact setup (so Kingston can really poke fun ):

Overheads: Nick Crowe's Tubedriver .93, ElectriQ (-6.6 db lowshelf at 60hz, -2.5dbs @ 500hz, 1.6db highshelf @ 11k), Kjaerhus GPP-1

Overheads sent to a bus along with Drum Directs and Voxengo's Marquis

Bass: ElectriQ (-4.5 db low shelf at 40hz, -3.5 dbs at 375hz, +2.5 dbs @ 2.5k, moog lowpass at 7k) GCO-1

Guitars: ElectriQ (-3db low shelf @ 130hz -1.1 db high shelf @ 8750hz)

Vocals: GAG-1, Electri-Q (high pass @120hz, -2.6 @ 800hz, +1.8dbs @ 4.5k, +3.5db highself @ 11k)

On the master was:

Voxengo Marquis, PSP MasterQ, PSP StereoEnhancer, and T-Racks (set to destroy)

[Funkybot's Evil Twin]

it's still a measuring or user error somewhere there. I just checked it and wavelab global audio file analysis tells me they are *always* equal. None of the numbers changed on any conversion direction or amount, either peak or RMS.

what do you measure with? I suggest you ditch it.

I instantly heard the difference when I put the files for A/B comparison, then when I went and looked at the waveform you could clearly see it. I'm using Audiomove 1.14, perhaps it's a bug in an older version? Are you doing 88.2 to 44.1 with it, or just 96 to 44.1? I swear, I tried it twice, and the difference was obvious. BTW I was measuring with level meters from InspectorXL (formerly from EA now by Roger Nichols), I know better than to trust the hosts metering. Again, I assure you, this is really happening time and time again.

[bduffy]

Here's the exact setup (so Kingston can really poke fun ):

Overheads: Nick Crowe's Tubedriver .93, ElectriQ (-6.6 db lowshelf at 60hz, -2.5dbs @ 500hz, 1.6db highshelf @ 11k), Kjaerhus GPP-1

Overheads sent to a bus along with Drum Directs and Voxengo's Marquis

Bass: ElectriQ (-4.5 db low shelf at 40hz, -3.5 dbs at 375hz, +2.5 dbs @ 2.5k, moog lowpass at 7k) GCO-1

Guitars: ElectriQ (-3db low shelf @ 130hz -1.1 db high shelf @ 8750hz)

Vocals: GAG-1, Electri-Q (high pass @120hz, -2.6 @ 800hz, +1.8dbs @ 4.5k, +3.5db highself @ 11k)

On the master was:

Voxengo Marquis, PSP MasterQ, PSP StereoEnhancer, and T-Racks (set to destroy)

Dude, just so you know, I am passing judgement on your mix bus. I just cannot abide that.

Well, your effects are of the higher quality variety! I would think they would benefit from it, but you may have squashed it beyond recognition...perhaps no mastering next time?