Yay for Digital Recording

I record digitally, of course — passionate debates about the Lost Virtues of Analog aside, it's the only reasonable way to make an affordable home studio. And heck, I think it sounds pretty darned good.

The computer I record to is a lovely Powerbook G4 (15", 1.6 GHz, 1GB RAM, in case anybody cares). It's a magnificent machine which replaced my previous Powerbook G4 after four years of hard use. The hardware and the OS have withstood some fairly significant beating (I don't go easy on my computers), and are a pleasure to work with. Apple really does do fine work.

To connect the mics to the computer, I use a Presonus Firepod. It also has proved an excellent machine. I had a little problem with it on OS X 10.4 where sometimes no sound would come out; Presonus support quickly pointed me to a setting in "Audio MIDI Setup" to fix the problem ("Clock Source" must be "Device," not "Mac"). That exchange was one of the finest support interactions I've had in a long time — and since then, the Firepod has worked reliably and consistently sounds great. You can read my full review (and others') at Harmony Central.

Before the Firepod, I used an M-Audio USB Duo. It sounds good (many of the recordings on this site were made with it), but was an unending headache to use. After that awful experience, I strongly recommend against buying anything from M-Audio.

Recording Software

I make my recordings in either Peak LE or Logic Express, depending on what sort of work I'm doing. (I sometimes find Peak easier for recording a whole session into a large file, but prefer recording snippets of spoken commentary into Logic.)

I'm not wildly enthusiastic about Peak or Logic: they're just a bit awkward and unpolished all around, with numerous glitches of the "minor irritant" variety. Using them just makes me long for Adobe Illustrator. But they're good programs and do produce very fine audio; in particular — of note to home studio builders considering less expensive software options! — both use audibly better algorithms for sample rate conversion, downsampling, and EQ than Audacity and every shareware option I tried. That right there is really the one good reason to pay $100 for Peak LE instead of going with the cheaper options; it has little else to offer above them. Logic Express is notably more robust than Audacity in numerous ways.

I don't know much about Soundtrack Pro, Digital Performer, Cubase, or Pro Tools, but they're all well-respected programs and I hear good things about them.

Making the Recording

I record at 96 kHz, 24 bits. To my surprise, I really can hear the different between 96/24 and regular 44.1/16 CD quality. So I do this with two thoughts: (1) eventually, higher-quality audio may be available to listeners, and (2) in the meantime, mastering work done at a higher precision will have fewer artifacts when I render to CD quality.

I'm extremely meticulous about placing the mics in exactly the same location every time I record. This means that, once I've found a position I like, there's no further fussing with positioning every time I sit down to a new session. If you're not careful about where you place the mics, and their position varies slightly every time, you're going to cause yourself a world of hurt when you get to the mastering!

The one thing I do have to adjust every time I record is the input level — some pieces are louder than others! I play through some of the louder sections of the piece I'm recording, watching the peak level. In the past, the advice was always "record as hot as you possibly can," but I think this advice is increasingly invalid for modern technology: new A/D interfaces have incredibly low noise floors, so 6dB of extra noise is still negligible — and if you're recording at 24 bits, you've still got plenty of resolution to work with if you have to add gain later. Recording hot doesn't have the big advantages it used to. It still has the same problem, however: even the smallest amount of digital clipping sounds awful and is nearly impossible to correct. I'd thus rather err on the side of being a bit too quiet: I give myself a bare minimum of 6dB of headroom.

Once I've set the level, I play a few repeated chords right in the middle of the piano, with the pedal down, and adjust the left and right channels to come out more or less equal. This adjustment is much easier to make before recording than it is while mastering: my recording technique creates a very wide stereo image, and with a full piece going all over the keyboard, it can be hard to tell which channel is really louder.

Then comes the music. Playing for a microphone often creates the same kind of nerves as playing for an audience. In both cases, I find music struggling at first to wrench myself away from listening with the outer ear ("What is the listener hearing? What will they think?"), and toward the inner ear that listens inside the music — but in both cases, the inner ear takes over as I warm up and gather momentum. The keys to a good recording session are all the obvious things: good practicing, a calm mind, a decent meal, good rest, open ears, patience....

Editing

If I don't have a good single complete take, the first task after recording is editing. This isn't a problem for the improvisations, which I publish just as they happened, but can take quite quite a long time with compositions. Just selecting which takes to use is quite time-consuming, and then making the splices smooth can make this whole phase add up to hours of work.

An aside on the philosophy of splicing: I'm skeptical of classical recordings with dozens or hundreds of splices that are pieced together entirely in the studio — it works well when the studio is part of the compositional process (Beatles), but in live performance, where the purpose of such splicing can only be obsessive perfectionism, it leads dangerously to a clinically perfect but spiritless recording without the organic expressiveness that makes music magical. Excessive splicing is tempting but dangerous! However, neither am I in the camp that decries splicing as some kind of hoodwink or moral failing; recordings are recordings, not live performances, and a musician's job is to work their medium's full potental to produce the best possible experience. So when I splice, I try to strike a balance between correcting really conspicuous mistakes that disturb the flow of the music, and preserving that flow in its natural form. End of aside.

There are two essential criteria for a good splice:

• It should be acoustically invisible: somebody listening to the sound of final mix should be unable to recognize that a splice occurred. That means no clicks or pops, stutters, burps, unnatural note starts and stops, sudden loudening or softening, etc.
• It should be musically successful: the music should flow naturally and be emotionally compelling. That means no sudden changes in tempo or dynamics — unless such a sudden change fits, of course.

It is easy to focus on one of these criteria at the expense of the other; keeping both in mind at once is difficult. The crossfade / edge blending features of modern editing programs help tremendously, but it's still a real trick. I find that the best crossfade time can be anywhere from 5ms to 2000ms; it's very much a case-by-case issue, and experimentation is key. Finding the right splice point is thus a matter of long trial and error.

Unless you have exceptionally good studio monitors, splicing is a job for headphones. (I do splicing on a pair of Grado SR-60s, which are excellent and surprisingly cheap.)

If you have reverb in your audio chain, turn it off when you splice. Reverb can mask problems.

I do the splicing in Logic, using multiple regions on the same track. Some approaches I often use:

• Splicing just before a conspicuous attack. This is my most common technique. I first choose a promising-sounding point in the music, a clearly identifiable attack which is fairly consistent between the two takes I want to combine. This method usually works best when there is a pedal change (or no pedal at all) at the point of attack, but can work in other circumstances if sustained notes in the background are at consistent volumes between the two takes. With the two takes in separate regions, I trim close to the attack in both, with a bit of slop:
  
  I then whittle each region down to the precise point of the attack, starting with visual inspection, but finishing with careful listening. I make sure that in the region on the left, I can't hear any of the attack before it cuts to silence — and in the region on the right, I can hear the complete, natural-sounding attack with as little as possible of the previous note.
  
  I drop the two regions right next to each other, no overlap and no gap, and take a quick listen. The splice will sound quite obvious, but I can get an idea of whether it's going to work.
  
  I then start dragging the right region's left endpoint to the left, to produce a crossfade just before the attack. Sometimes a very brief overlap works best, sometimes a longer one. Repeated listening and experimentation is key here.
  
  When using this technique, watch out for stutters, hiccups, a sudden "switched on" attack that doesn't include the hammer strike, and notes sustained or ringing in the background suddenly getting louder or softer at the splice point.
• Crossfading over the body of a note. This approach can work well when attacks in the two takes are indistinct, or there is too much pedal to get a single clean break. This approach starts exactly like the previous one: find the same note in the two takes, trim to that note, stick the regions next to one another. But then, instead of crossfading immediately before an attack, I drag the endpoints of the regions to produce a long fade (sometimes several seconds!) during the body of a sustained note or chord, well clear of attacks on either side. When using this technique, listen for notes growing louder or softer in unnatural ways, and for phasing or interference effects caused by the two takes being slightly out of tune with each other.
• Crossfading over silence. This is nice work if you can get it — but be careful! "Silence" is not actually silent. A piano keeps reverberating even if the pedal is down, and there's always room noise, which may change over time. There's a temptation to just do a quick "snip," but that rarely works. Although it's easier to find a good point, splicing empty space takes the same care as splicing a note. I usually use the previous technique to crossfade a long period of silence, listening with the gain up so that I can clearly hear reverberation and room noise. When using this technique, listen for reverberations suddenly cutting out, and for background hiss and rumble changing suddenly.

A final warning on splicing pianos: the darned things have a lot of notes, and they reverberate like nobody's business. The transition from one note to the next is rarely as clear as our ears believe: the preceding notes ring into the next one a great deal.

We pianists are also far less consistent in tempo and dynamics than we believe — especially when starting in the middle of a passage. What we play depends on what we've been playing, and the sort of momentum we've built up.

For both these reasons, when you have a mistake you want to correct or a passage you want to redo, it works far better to start well in advance of the actually passage in question. Work your way into it; never start playing on the exact note you want to splice — even if it's easier to start there, even if it's the start of a new section, even if there's a rest before it. Back up and work your way in. You'll save yourself a bunch of heartache later!