Louder, Until It Stopped Working

Play two versions of the same track, identical except that one is a decibel louder, and listeners prefer the louder one. Reliably, without being able to say why, and this includes trained engineers.

That single perceptual fact produced thirty years of degradation across the recorded catalogue.

The reasoning at each step was sound. A record that sounds bigger on the radio, in a shop, next to its competitors, sells better. Compression reduces the peaks, which allows the average level to be raised. So masters got louder, and then louder, because whatever level everyone else reached, being above it was an advantage.

It is a textbook arms race with no stable resting point. Every participant gains individually by being louder than the rest, and collectively everyone ends up at the ceiling with nothing gained.

The cost is dynamic range: the distance between the quiet parts and the loud ones. Squash that and the chorus cannot be bigger than the verse, because both are already at maximum. The music loses its light and shade, and beyond a point the waveform is clipped and audible distortion appears.

It became a public issue with records where listeners could hear the damage. The 2008 Metallica album Death Magnetic was mastered so aggressively that fans noticed the distortion, and the versions released for a video game (mastered with more headroom) were widely preferred. Thousands signed a petition asking for a remaster. It is one of the rare cases where an audience identified a technical mastering decision and objected to it by name.

Appeals to restraint achieved nothing for two decades, and they could not, because any engineer who complied was punished immediately: their record sounded smaller than everything around it, and the client noticed.

What ended it was a change to the rules.

Streaming services introduced loudness normalisation: playback adjusted so every track hits the same perceived level. A hyper-compressed master is now simply turned down, at which point it does not sound bigger. It sounds flat, because it has no dynamics and it no longer has the volume advantage that was concealing that.

The incentive reversed, and mastering practice followed within a few years. Nobody was persuaded. The referee turned everything down to the same level, and the behaviour changed because the payoff changed.

There is a lesson here that reaches past audio. A coordination problem in which everyone's individually rational choice produces a collectively bad outcome is not solved by exhortation. It is solved by changing the structure, and this one was solved by a default setting.

And a working principle worth carrying into any listening comparison. Because louder is reliably preferred, any comparison that is not level-matched to within a fraction of a decibel is measuring loudness. A great deal of audio folklore (about cables, converters, formats, plug-ins) survives because the versions being compared were not matched, and equipment that adds a little gain on the way through has been winning listening tests for fifty years on that basis alone.

Match the levels, blind the test, and a surprising amount of the difference disappears. Some of it does not, which is where the interesting engineering lives.