Peak normalization matches peaks but not perceived loudness, which drove the loudness war

Peak normalization finds the highest peak in an audio file and scales the whole file so that peak reaches a target ceiling. But the ear responds to average, not peak, level, so equal peaks do not mean equal loudness: a sparse, dynamic track normalized to the same ceiling as a dense, heavily compressed one sounds far quieter. This mismatch created a perverse incentive — producers found that compressing more raised average level while staying within the peak ceiling, so they could sound louder than competitors at the same peak, sacrificing dynamics. That is the ‘loudness war.’ Peak meters such as the QPPM read peaks but do not reflect loudness or long-term average, so they cannot warn that one programme will sound louder than another at equal peak. The fix is loudness normalization (e.g. EBU R 128, targeting LUFS), which turns quiet-but-dynamic material up and brick-walled material down, removing the incentive. Note that DAW peak-normalizing also costs a DSP generation of quality; professional mastering engineers set levels by ear for equal perceived loudness instead.

Examples

A sparse acoustic song normalized to 0 dBFS sounds much quieter than a dense track also at 0 dBFS. Two programmes at identical peak — a 20 LU-range orchestral piece and a brick-walled pop song — differ greatly in loudness; loudness-normalizing both to -23 LUFS turns the orchestra up and the pop song down.

Assessment

Explain why normalizing all songs to the same peak level fails to make them sound equally loud, and why a quasi-peak meter cannot detect which of two equal-peak programmes will sound louder. What does EBU R 128 measure instead, and why does it defuse the loudness war?