Loudness vs. Speech Normalization in Broadcast

摘要

This paper presents an empirical study of the differences between level normalization of programs using the two dominant methods: loudness normalization and speech (“dialog”) normalization. Instead of adding to the continuing debate of the subjective merits of one method over the other, important technical aspects are examined here. The difference in normalization level between the two measures was up to 14 dB. For all films and programs, the loudness method provided the greatest headroom. All programs and a majority of the films could be broadcast at a fixed target level of −24 LKFS (loudness, K-weighted, relative to full scale) without dynamics processing. When it was speech normalized, not a single film could be broadcast at that target level without applying dynamics processing. The study furthermore found a systematic difference between manual speech measurement and automatic speech measurement. Uncertainties in classification, definition, and measurement are described and compared to the requirements for precision in the Advanced Television Systems Committee (ATSC) and the European Broadcasting Union (EBU) loudness-based standards.