Audio Loudness: Two Disturbing Transgressions

Speaking of live music recording…those of you who’ve also tackled this challenge will likely agree with me that optimizing per-channel input levels represents one of the more difficult aspects of the project. On the one hand, you want to get the peak input level as close to 0 dBfs as possible, in order to maximize the SNR (specifically, the ratio between the tunes you’re attempting to capture and the ambient environmental noise). But on the other hand, if you exceed 0 dBfs, you end up with clipping.

Unlike in the analog past, where clipping distortion was merely unpleasant, digital clipping is highly undesireable. And subsequent restoration of a clipped recording is impossible, although if the distortion is minimal, interpolation-based tools can muster a semblance of repair by replacing clipped samples with synthesized alternatives. All of which showcases the value of fast-response level meters…along with the ability to (if possible) set levels during a band’s sound check, so that you’re not unhappily surprised once the set begins and the musicians crank up a high-volume jam.

All this talk of loudness reminds me of two related topics to which I thought I’d devote the remainder of this particular post. The first is the ‘normalize’ setting often found in audio encoding and playback programs such as iTunes and Windows Media Player. Its purpose is to adjust the volumes of multiple audio tracks to a common level. Admittedly, I conceptually understand its value; in listening environments with high ambient noise levels (cars, headphones while on-the-go, etc), normalization precludes listeners from having to constantly fiddle with the volume setting in order to hear quiet passages while not subsequently being blasted by loud passages.

And admittedly, I also realize that most music "listeners" (I make charitable use of the term in this case) don’t enjoy their tunes in low-ambient-noise critical-auditioning settings. But the idea of normalizing music tracks during playback still makes me cringe; it’s counter to the original intention of the musicians and the recording engineer team (then again, I rarely listen to an album’s tracks out-of-order, or track-at-a-time…call me a purist, it’s ok). And I would definitely never advocate normalization as part of lossy- or lossless-encoding tracks for archive. Once you’ve done so, you’ve permanently lost the original recording’s dynamic range and other desireable attributes.

Speaking of ‘the original recording team’, they’re the instigators of another highly disturbing trend related to loudness. Check out this YouTube clip:

An increasing number of musicians and recording engineers, in no small part motivated by their labels’ demands, are over-compressing recordings in order to make them ‘louder’ on average through their entire playback duration. The result? As IEEE Spectrum said in a recent article:

From the mid 1980s to now, the average loudness of CDs increased by a factor of 10, and the peaks of songs are now one-tenth of what they used to be.

For more on this disturbing trend, check out the following chronologically ordered links I’ve been archiving over the past year-plus:

• Why Music Really Is Getting Louder
• The Future Of Music: Part One (with Ars Technica and Slashdot commentary) and Part Two

p.s…speaking of lossy compression, its prevalence is influencing recording engineers’ techniques, too. Sigh…