Saturday, July 07, 2007

Deconstructing Gwen

Gwen Hughes is an angelic Diva that I am honored to call my friend. While I am merely internationally ignored, she is internationally renown for her performances and songwriting. Yeah, she's that good! If you want to hear for yourself, why not take a moment to check out her Electronic Press Kit?

She's humored my pesterings of musical questions and nurtured my ambition to produce a new, modern-Euro sound for her music in the past, but now she's gone one further... actually collaborating with me on two new songs of hers!

Now Gwen and I are two very different musicians (if I can even call myself that). We often talk at great lengths about how different our worlds are. True, I have modest keyboarding and music-producing abilities, but this woman sings and plays piano for a living... a real professional.

So I eagerly accept the opportunity to "re-produce" or "re-mix" one or two of her songs... but can I do it???

Well, I found it quite amusing that path I had to walk to be able to absorb her music, and so I created this post... let it bear witness to the genius that is Gwen.

Input

Two mp3s arrive in my mailbox. I eagerly listen to them. On first glance, they seem like complete songs... drums/bass/keys/vocals (all done by her). I play along with them, figuring out the chords (more or less) as the songs play and I think, "yeah, I can do this."

Now I should spend a moment discussing the ways one can go about re-mastering a song. The first method, identical to the way a DJ would re-mix a song off the radio, is to use the song itself as the building blocks to create a different song. The entails sampling the song into lots of little bits, which itself might require some audio processing.

For example, let's say I want to use a bit of the song as a base for a new song, but the part I want to use has vocals on it that I don't want. Well, you can consider a song to be like a layer cake: at the bottom layer (the lower frequencies) is the bass, and other instruments are found occupying higher layers. (Drums, especially hi-hits and snares, comprise the highest frequencies.) By using audio processing techniques, I can "cut out" instruments (even voice) from the original recording!

But, due to the fact that sound "leaks" between the layers (as harmonics) the end result isn't as nice as if I had started with clean recordings of each individual instrument.

Add to the fact that I'm serious... I want to create a 24-bit, 44.8KHz recording... and that I'm starting with a .mp3 (which is already lossy) and you'll soon see that any end result I do would have to again be redone with higher-quality source material.

It's not like me to do something 1/2-way.

The second approach, also widely used, is simply to re-create the song first, using local instruments, so that one can start with high-quality recordings. A bonus of this approach is that if done properly, one gets more than just a recording... one gets a capture of the creation process that can be edited (say, changing a note, or a trumpet to a sax)!

I knew before Gwen ever sent me anything that I would want to use the latter approach. Why? Because I'm a musician... I feel like I have a larger expression with MIDI data than I do with MP3 data. Besides, if I'm going to help write a song, I think it's reasonable to assume that I could play the song.

The Devil's in the Details

But with Gwen... that turned out to be a much larger problem! See, Gwen is a professional jazz musician... her stuff is a bit more complex than the usual drivel that I was used to on the radio. (Even the "sophisticated" stuff I listened to is pretty simple by comparison to jazz.)

While I could play along with her songs, I was a LONG way from being able to reproduce what she sent. I could add to it, but not create from scratch. No matter how hard I tried to hear what was going on, it was stuff I had never heard or seen before and I was stymied.

I knew however, that if I just watched her play the song once, I would be able to reproduce it. God has blessed me with that skill. So, frustrated, I called her and asked for the chord charts. This would at least verify that my chord guesses were correct.

Well, after busting my chops (rightly so) for not putting more time into learning the song, she sent me the charts. (Thanks!) I now owned a few pages of symbols that, when presented to a musician might actually "sound" like a song- but in my hands, were just stark reminders of how much I was not a jazz musician. Armed with a map to the song, I began my quest.

Creating Audio Samples

The first step was to take her mp3 and break it down structurally in a way that it would be represented on a chord chart. To do this, I loaded the song into CoolEdit and meticulously set markers corresponding to the start of key measures in the song.

The next step was to get my recording studio setup. For this part of the song, I needed a mixer, drum machine, and samplers for the bass, piano, and strings. That was the easy part, harder is programming the instruments!

I started with drums. The song only has a basic 16-beat drum pattern looped, so I figured that would be a good place to start. Using the samples I just created, I figured out the beats per minute (113) and set up my recorder to loop over the first 16 beats while simultaneously playing the first 16 beats of her song. I listened to that drum pattern over and over on a loop until I could anticipate each beat. Once I could play the drum pattern, I programmed the drum machine and took a deep breath in preparation of doing the same thing with the bass and piano.

Normally, I can pretty much play along with a song on first hearing, but *replicating* a song is completely different. "Anal-retentive" attention to detail comes to mind. I won't say how many times I had to listen to just the first bar of the bass+piano, but a few hours later, I had something close!

However, this was just "replicating" ... this wasn't "playing." With heavy rhythmic devices like drums and bass, this is okay, maybe even preferred, but with other instruments, the result sounds ... mechanical. So, I quantized the bass (forced the notes to align on specific beats) and armed with the chordal structure as well as a close approximation of exactly what notes were being played, I put the the drums and bass on a loop and tried to "feel" the song. In other words, I had to "figure out" the unconscious little motions that Gwen used to make her music. (This is why it's easier to do if you can actually watch someone do it!)

Now, I won't say I replicated exactly her fingers, but it's really close. If anything, I have a tendency to "hear" more than what is actually there, so my recording might be a bit "heavy-handed" compared to her delicate fingerings. That's okay; now that I have the notes in MIDI format, I can delete (or edit) them later if needed.

And so, after a total of around 8 hours of work, I've replicated about 25 seconds of the song!

Recall that the point of doing all of this isn't to replicate the song, but to START from that point. So, I'm looking at more hours of tedious work before I can actually start to be creative and start contributing to the song.

Well... no one said it would be easy! Of course, I'm sure for some people, it is!

If you're curious... here's Gwen's original recording and then my version. The instruments might sound different (easily changed) and the volumes might be different (easily changed) but the notes and rhythms should be there. And, even if they are a little off, I can change them too without having to re-record. I love making music ... well, I will when I actually get to make it! :)