Recording Music

Live music is best, no doubt in my mind, you can't do better than live. But practicality makes it impossible for us to go some place and play music or listen to others play music and so we record music so that we can listen to it later.

Recorded music sounds best when it sounds like it did live, but that is often a lot more difficult to achieve than one would think. Some of the issues are technical, some are psychological.

A big psychological factor is room acoustics. If we are in a room that is acoustically hard, that is to say it contains a lot of reflective surfaces and so we hear the direct sound produced by the musicians and a lot of reflections off of various objects, delayed by some milliseconds, our brains tend to adjust what we hear. We'll hear the echoes but they will seem to be at a low level relative to the direct sound, even if they are not, and will add to the ambiance of the sound unless the echo is extreme.

However, if we record that performance, and then later listen to it with headphones, so that we don't add echo from the acoustics of the room we are listening in, the reverberations will often seem overwhelmingly loud and annoying. For some reason, in the absence of the original environment, our brains do not perform the same filtering that they did when we were there.

If we take steps to eliminate room acoustics, feed electric instruments directly into a mixer and close mic those which are not electric, mix the whole mess down and record it, then if we listen to that it will sound flat and dull.

To sound natural, we need to hear some reverberations but they need to be at a lower level than usually are naturally present when recorded. Commercial recordings often address this by taking the direct feed and close mic approach and then adding some fake reverb.

When recording live music, there is a sweet-spot, a distance from the performance that, when recorded, will have just the right amount of reverb to sound natural without being overwhelming. In my experience this is usually fairly close, but not right up to the instruments. With amplified music this is not always possible because the amplified sound output may come from multiple diverse sources.

Assuming you can get a decent acoustical sound to record, there are various additional challenges. Live music has a high peak-to-average ratio. Musical peaks may be 20 decibels or more above the average level. These peaks are too brief in nature to read on VU meters, and sometimes to brief for peak led circuitry to response, although the latter definitely will respond faster than a physical meter. If the led peak indicators do not have a capture function, they may flash too briefly for you to see.

Analog tape recordings didn't clip hard, that is, instead of sound reproduction remaining linear up to some hard limit, analog recorders got progressively less linear as their design levels were exceeded. This has to do with the non-linear characteristics of magnetic recording heads and media. Because of this "soft" clipping, analog recordings wouldn't suffer greatly if an occasional brief peak exceeded the intended recording level.

Digital recording isn't like this. Once the bits are all ones or zeros you can't represent a signal that goes beyond this. The signal is hard clipped. This can result in many high order harmonics and intermodulation distortion products. The intermodulation products are the most objectionable because they are not harmonically related. When you make a recording on digital media, you have to allow enough overhead for the peaks to be recorded without clipping.

Consumer equipment, like the camera I recorded Brian Aubert with will generally do this automatically, within limits, but if you are using professional equipment you need to allow plenty of overhead. Keep in mind that a sixteen bit digital recording should have a signal-to-noise ratio of around 96db, so you are not obligated to keep the level right at the maximum to avoid noise.

Sample rate is another thing to be aware of. People often cite the Nyquist limit, which states that the maximum frequency you can digitize is half the sampling rate, as an argument suggesting that 44 KHz is more than adequate. I am going to disagree and I will elaborate on why.

First, because it takes at least two samples to record a cycle of a waveform, aliasing occurs as you approach half the sampling rate. Consider for example, sampling a 22 KHz signal and a 44 KHz sample rate. The phase relationship becomes important. If sampling and the signal are in phase with each other, then you sample once at the positive peak of the input signal, and then the next sample at the negative peak, and the resulting sample has a high amplitude. But what happens if you shift the sampling relative to the input signal by 90°? Then one sample happens just as the signal is crossing the zero line from positive to negative and you get a value of zero, and then the next signal happens as the signal crosses the zero line from the negative to the positive, and again you get a value of zero. The signal has completely vanished!

Instead of sampling a 22 KHz signal, let's sample a 21,990 Hz signal, what happens? The sampling goes in and out of phase with the sampled signal ten times per second. In effect, the 21,990 Hz signal is modulated at 10 Hz. You will get an alias signal at 22,010 Hz because the encoding isn't entirely linear, particularly when the signal is a high enough frequency that it is only sampled twice per cycle. What's worse, if you actually get audible information above 22 KHz, it will alias to frequencies below 22 KHz. For example, a 25 KHz signal will produce an alias at 19 Khz.

To avoid this aliasing, low pass filters are employed to roll-off frequency response before half the sampling rate is reached. So for example, you might roll off audio that is going to be sampled at 44 KHz at 20 KHz. However, there is a problem, filters introduce phase shift, and the more steep the roll-off, the greater the phase shift. There are those who will argue that humans can't hear phase relationships, I would argue they are full of it. I believe distorted phase relationships ruin your ability to acoustically locate a sounds source.

Sampling at 48 KHz verses 44 KHz gives you another 2 KHz usable audio spectrum, or allows the use of a less steep low pass filter. Better still is sampling at 96 KHz or 192 KHz which really allows the full audio spectrum to go unmolested. I buy music in DVD format when I can because I can hear a non-trivial difference in DVD audio verses CD audio (assuming the DVD audio is not compressed, usually with a movie it will be compressed because otherwise it will take too much space). DVD's can be recorded with either 48 KHz or 96 Khz audio sample rate, and can be either uncompressed (LPCM) or compressed (Dolby Digital).

The 96 KHz rate doesn't get used much but for the best possible audio quality, that's the way to go. Understand though it will eat a lot of space, so whether or not you can use that will depend in part on how much program material, the quality of video, etc. If you want to include decent quality video and the program is long then you're either going to have to use compression or a lower sample rate or both.

Most audio compression schemes are "lossy", which means they save space by throwing away some of the data. Generally, audio is sampled and then converted via a fast Fourier transform into what frequencies were present at what amplitudes in a time slice.

People are less sensitive at the higher and lower frequencies than to midrange frequencies, so generally signals that are below a threshold amplitude at a given frequency are eliminated.

A signal of a given frequency will mask out other nearby signals that are substantially below the amplitude of the masking frequency, so those are eliminated.

Humans have more temporal resolution at higher frequencies, that is, we can detect changes in the amplitude of a higher frequency more frequently than a lower frequency, so, some encoding algorithms like Ogg, don't include low frequency sample data as often as high frequency data.

Most higher frequency content in music is harmonically related to lower frequency content so some schemes eliminate data for frequencies above 7.5 KHz, and then try to reconstruct the content based upon frequencies that are half of those in the 7.5 KHz - 15 KHz range. This is known as band replication. This eliminates non-harmonically related frequencies, it also assumes a constant power relationship between fundamental and harmonics, but in real musical instruments, the harmonic relationships vary from instrument to instrument and are responsible for each instruments unique sound.

Fewer bits can be used to encode a wider dynamic range if the encoding is not linear, some compression algorithms take advantage of this by using u-law or other non-linear encoding schemes.

Every encoding scheme makes trade-offs and there is some variability in an individuals sensitivity to various factors. Some encoding schemes seem to work best with certain types of input.

My own experience with compression is limited to Digital Dolby, MP3, and Ogg. Unfortunately, the math is way over my head and the explanations of the differences aren't too useful without understanding the math behind them, so I can speak only from my experience.

In my experience, if you can go with uncompressed, if the space is there, then do that. No compression means no compression artifacts. It also means a big file which on a fixed size media where the audio has to compete with other things, may be problematic.

If you have to use some form of compression, I've found Digital Dolby/AC3 and AAC pretty good in terms of minimizing damage over all, at relatively high bit rates. AC3 seems to work well for a broad range of musical and non-musical sounds.

I have found MP3's generally are ugly, the ambiance seems to be missing and MP3 recordings sound flat. In addition, if the music is complex in nature, it seems to "blur" the sound for lack of a better term, individual instruments become hard to distinguish. Also, the temporal resolution of MP3 is inadequate at high frequencies and does ugly things to percussion. MP3's work best for very simple music, flute, guitar, or piano solos for example, and then only with VBR or 320 Kb/s or greater bit rates.

For low bit rates and complex music, Ogg seems to do well. It doesn't tend to lose the ambiance the way MP3 does. However, if the bit rate on Ogg is too low, it sounds almost like an old cassette recorder with a bad capstan, that is, a flutter effect. Ogg also seems to sometimes fail on very simple music like a piano or guitar solo. It's very strange, but in those situations it sounds as if the fundamental is attenuated and only the harmonics are audible or are exaggerated relative to the fundamental giving a stringed instrument a hollow sound. Oddly, this does not seem to be a problem if the music is more complex, multiple instruments are playing, chords are being played, etc.

Now with all that background, my complaints about modern recorded music follow. I am convinced that MOST recording studio engineers are deaf. I can think of no other explanation for the crap that is churned out.

First, clipping, damn idiots, if you've got a media with 96 decibels of dynamic range (typical CD), you do NOT need to make the recording as loud as possible, allow some headroom. Digital clipping sounds UGLY. Cymbals sound like gated white noise, their metallic resonances gone. Snare drums sound like gated white noise. In fact, everything above 4 Khz sounds like gated white noise when the high frequencies are clipped.

Bass, don't clip the damned bass either! If bass is allowed to clip the entire recording then bad intermodulation distortion results. I like the Weezer song Island in the Sun, but damn the whole recording is clipped on the CD, the bass intermodulates everything else, and the percussion sounds like gated white noise. Having heard the song performed live without the IM distortion and with real instruments it sounds SO much better, I want a recording that sounds like that. Why, deaf engineer recording idiots, can't I have that?

Same is true for the Red Hot Chili Peppers song Snow. Not nearly as bad, but still in the louder portions of the song it is notably distorted and clipped, much better than Island In The Sun, only the peaks are chopped off while Island In The Sun is pretty much clipped to hell all the way through the song.

Another complaint, what you do with EQ, DAMN YOU idiot tone deaf engineers for ruining so much good music. A kick drum has a striker and it makes a friggin' noise when it hits the drum that is PART OF THE SOUND, and there are actually some transients in that sound. When you run it through a mush filter and roll off everything past 40 Hz it SOUNDS LIKE SHIT! A kick drum doesn't go wum wum, it goes THUMP THUMP! On the other end of the spectrum, cymbals, THEY HAVE SOUND BELOW 5 KHZ! A lot of the resonant characteristics that gives each cymbal it's unique sound comes from mid-range fundamentals. Filtering those out turns it into gated white noise hiss hiss hiss sounds that don't resemble the natural sound of the instrument at all. Horns same thing, you eliminate lows and highs and you totally destroy the total characteristics of the instrument. Half the time I can't even tell what type of horn is being played in a recording it's so badly screwed up. WHY? I've got a vinyl album of the Crazy 8's (for those who were don't know, before CDs we used to get music in the form of a spiral groove pressed into a 12-inch diameter plastic platter. Oddly, this primitive technology often produced surprisingly good fidelity at least initially but the durability had a lot to be desired. At any rate; the horns on this pressed piece of vinyl sound just like being there as they should, but most CDs I have suck, but there are notable exceptions.

Compression, reducing the dynamic range of audio. Again, with vinyl, when you had at best 65 decibels of dynamic range to work with, compression sometimes was necessary, and if it was reversible, Dolby, it really made sense, improved signal to noise ratio without destroying program content. On a CD with 96 decibels of dynamic range, virtually any music ought to fit within that without destroying it with compression. Granted, I can understand when you're recording belly-button girls or RAP instead of music, and your goal is to make it as loud as possible on a boom-box or crappy car stereo, then yea, I can understand ironing the audio as flat as a pancake in those applications, but STOP DESTROYING REAL MUSIC.

I have noticed a lot of the destruction seems to be something record companies inflict intentionally. I say this because so often I've heard artists start out on small independent labels, and the recording quality is excellent, a good example is Hearts initial album Dreamboat Annie on Mushroom records, the highs were crisp and clean, the overall recording was excellent. Then their next album came out on ABC Dunhill, YUCK! Gone were the delicate highs in the guitars and vocals that so contributed to the overall sound of the first album, in place of those, the usual clipped distorted mush.

I've repurchased, on CDs, much music over the years that I originally bought on vinyl or Beta Hi-Fi, but which has been destroyed over the years, or stolen (I had a bunch of my records stolen at one time years ago). Again, the CD copies almost always have been inferior to the original release. When there is considerably lower distortion, more dynamic range, and better frequency response on the CD, there is absolutely no excuse for this.

The Ventures and Surfaris on Vinyl were both great recordings, very clean, unclipped, but these were stolen from my parents house and later I re-purchased them on CD, yuck. The clean highs were replaced with clipped mush.

I have a Duran Duran compilation on Beta Hi-Fi tape. Beta Hi-Fi was a really interesting format in which the audio tracks were recorded as FM subcarriers on the video track. The fast head-to-tape speed of beta (relative to VHS) allowed a wide subcarrier with a high modulation index to be used. The result was superb audio, low distortion, good frequency response, high signal-to-noise ratio. Rio, by Duran Duran, had the cleanest wonderful highs in that recording. But my beta Hi-Fi deck wore out, and a modern replacement would run about $1,000 (yes they still make them if you're willing to pay a grand for a VCR), insane, so instead I opted to try to replace the content that was important to me. I first bought a CD with the songs on it, quality was absolute crap, highs on Rio were all clipped and distorted, voices were muffled, overall sound definition was lost. I then was given a DVD compilation as a gift, and it was much cleaner, closer to the original recordings, but still not as clean as the old Beta Hi-Fi tape.

Now, I know this isn't a problem with the DVD technology itself, because I have some, in fact most of my DVD records, are good to excellent. One in particular, a DVD of the Puffy Ami-Yumi Jet Tour, that's as clean of a recording as you can ask for. But most CDs are junk, most DVDs are better but most still have room for improvement.

What really bugs me is that these companies have multi-million dollar studios to record with or very expensive high end equipment to record live performance, yet most of the recordings are worse than what I get with relatively low-end consumer equipment.

Aside from the above, there are acoustical issues, the standard studio approach of putting everyone in their own booth, recording separate tracks, mixing it down, adding fake reverb, sometimes that produces acceptable results, most of the time it sucks and doesn't sound natural. The best recordings I've heard have either been live or where they've put everyone in a large studio, mic'd the instruments close enough to avoid excessive reverb but not so close as to get none and picked up just enough natural reverb.

Why am I writing all this? I dunno, I guess it hurts just a little less than physically banging my head against the wall although I feel that in all probability the end result will be the same. Maybe though, just maybe, it will encourage people to actually listen to the music they buy and be just a little more selective, and in doing so force the industry to start paying some attention to sound quality.