Monday, January 28, 2013

Coulton's Audio Tracks Part 2

In my previous post, I discovered the that the bass track used by Glee had been subtly altered from Coulton's track. At the end, I tentatively made the naive assumption that there would be no reason for them to re-record the bass track if they were working off of Coulton's original. But science doesn't assume anything. In fact, Paul Potts reminded me of a very good reason to re-record the bass. Namely center-canceling. He explains it well on his blog, and goes on from there to compellingly demonstrate which specific version and file format of Coulton's audio track would have likely been used -- a feat which I find nothing short of remarkable.

Essentially, center-canceling (aka center-kill or vocal-kill) is a technique where the left channel and right channel are split and one is inverted before adding them back together. This effectively mutes any sounds which are dead-center in the track. It's often used to eliminate lead vocals to create karaoke tracks, but since the bass is usually centered, it also removes the bass. Paul was looking at using the technique to try removing (most of) the duck quack from Coulton's source, but then he struck upon the genius idea of also center-cancelling Glee's track.

I was able to reproduce this result, and it results in a much cleaner track, with no bass or lead vocal. That lets me do what I had originally had intended to do -- compare the rhythms of the banjo and mandolin tracks. Unfortunately, these parts still get easily buried in a wash of strings, backing vocals, and compression, even after I apply heavy EQ. They're still kind of quiet, but I think there's enough there to go off of. In addition to using a center-cancelled and EQ'd version of the backing track, I also slowed the tempo waaaay down, which makes the precision of the timing easier to hear. It also lowers the pitch. A lot. What were originally high-pitched compression artifacts in the mp3 become annoyingly noisy background chatter.

I'll let the results speak for themselves.

Example 1 -- Intro Riff and Beginning of First Verse

This is the longest example, because I wanted to demonstrate how well the parts stay together over a long period of time. What you will hear on this tracks is: (1) only JoCo's solo banjo and mandolin parts mixed together on only the left speaker; followed by (2) only the Glee version, modified as described above, on just the right speaker; finally followed by (3) the two of them together (JoCo on the left, and Glee on the right).

First, we'll hear it at full speed:

Then, we'll slow the whole thing down to a staggering 30% of full speed. This track is over 3 minutes long, but it really demonstrates the precision over a long period of time:

To my mind, it is absolutely inconceivable to consider the possibility that someone could deliberately perform something so precisely at full speed.

Example 2 -- Added Fret Noise

As I was looking for distinctive spots on the banjo track to test, I came across this one distinctive little riff between phrases (after the lyrics "average groupie"). The banjo and mandolin have a sort of quick syncopation going on here, but what made it extra distinct was that it had extra fret noise in the track, which I discovered wasn't in Coulton's track. The riff itself is dead on, but it appears that the audio engineers actually added a fret noise to the track as they were editing. I can only speculate why they might do that (this particular section of the verse seems to have been cut from the final screen version, so it probably wasn't a foley effect). As before, this will start with JoCo on the left, then Glee on the right, then both. Note that while I'm only playing the banjo and mandolin (as before) I did verify that this noise isn't in the other tracks.

We'll listen at 50%:

Example 3 -- The Quack

As long as we're cleaning things up and slowing things down, we may as well listen to the "ghost of a quack" that Paul discovered in the track. Bear in mind that we are not listening for an actual quack, but for the artifact that was left over when the quack was almost center-cancelled. This sound comes just at the end of the following clip. As before, I start with Coulton's banjo & mandolin on the left (for reference, but no quack), then just Glee on the right (there's a sort of "click" near the end), and then the combined version.

But in case you don't hear it, after I play the combined version, I begin to loop over the quack. And while it loops, I do something almost magical... I discovered that the quack corresponds to a sharp spike around 4 kHz, so I slowly fade away all the other frequencies except for a narrow band at 4 kHz. This effectively isolates the sound.

Here's what it sound like at 40% speed:

And here's what it sounds like when you bring it back up to tempo (just the mixed version this time). This actually sounds pretty annoying:

In Closing

At this point, if I were on a jury, I don't see how there could be any reasonable doubt that these are the same tracks. To get something to match so closely at such a speed would be a super-human feat. And we already saw from the bass analysis that they just weren't that concerned with getting an exact match (nor would any sane person). So while the bass track appears to have been removed with the duck (and maybe lead vocals), the banjo and mandolin tracks do indeed seem to be none other than Coulton's. Which is what everyone suspected all along.


BoffoYuxDudes said...

Caleb. You and Paul are doing some amazing stuff here. Well done.

You guys now are gonna inspire me get off my butt and finish the Devil Dogs video. Consider it a treat for all your hard work.


Paul Potts said...

You make a really compelling case, Caleb! I agree that it's pretty much impossible to believe that they re-recorded the stringed instruments. I couldn't replicate a guitar part that precisely, and it certainly wouldn't stand up to scrutiny under a super-slow-speed microscope.

Just to clarify, the last two tracks are (1) the center-cancelled Karaoke, and (2) the center-cancelled Glee track, right? So you're asserting that that quack "squeak" is in the Glee track because they used the center-cancelled Karaoke track as a backing?

I have gone back and forth a little on how they did this, because it seems like I _also_ hear some leftover lower frequencies of the quack in the finished Glee track. They shouldn't be there if they Glee really started with only the center-cancelled Karaoke track, because there would be nowhere else in their final mixed track for those frequences to _come from_. I wonder if maybe they _mostly_ center-cancelled it? Like, just attenuated the center enough to kill most of it while doing less damage to the other instruments? Perhaps instead of doing the full center-kill, they used a plug-in for this.

Rusty said...

I'm so impressed with the methodical, scientific and thoroughly convincing manner with which your and Paul R. Potts have attacked (and decimated) this issue. FOX and GLEE should be embarrassed. And they owe Coulton an apology and compensation. He should get paid as an arranger and musician on the track. And they should pay him damages.

Caleb said...

Paul, the only thing here that comes (directly) from JoCo is the banjo & mandolin tracks (on the left speaker). Everything else (on the right) is from the glee track (as ripped from YT). I don't have the Karaoke track to test with.

I didn't notice a lower frequency sound, but once I found that higher pitched spike, I focused in on it (so there could be more there?). It's also possible that what I'm hearing there got transformed via the conversion to/from YT. It doesn't seem like a typical MP3 artifact to me, though.

Paul Potts said...

It wouldn't be an MP3 artifact per se. It would be more like the residue of cancelling both the correctly reconstructed audio and the artifacts, which would tend to emphasize the artifacts, since the correctly reconstructed audio actually would cancel.

In any case, while I am still really curious about exactly how they did it, we may unfortunately never find out or see any testimony. It is not really necessary to prove that every detail of the track matches or exactly what transformations they may have done. What we've shown already ought to be sufficient to prove that they used a substantial portion of Coulton's audio. I mean, keep in mind that artists can wind up liable for infringement for using only a _sample_ without permission.

JoeCovenant said...

What is groovy about this is that it scientifically confirms what the ears hear! And being the techinical vacuum that I am, it's nice to know that aural confirmation still works! ;' )