Normalizing dialog

Not that loudness

If you look up loudness in your trusty, dusty Audiocyclopedia, you’ll find something that has absolutely nothing to do with 21st century loudness. So forget what you thought you knew about loudness. The only way to calculate digital loudness with scientific accuracy and repeatability is to monitor and record the digital levels of each packet containing audio, and apply an algorithm to determine loudness. While some might argue with the algorithm formulas, measurements made with the same algorithm will at the least be scientifically consistent.

The ITU-R recognized the international need and began work on a comprehensive, scientific, statistically repeatable definition of loudness. In October 2007, the ITU-R introduced BS.1770-1, which was the original digital algorithm designed to measure loudness and true peak. In May 2011, ITU-R adopted BS.1770-2, which added level gating. Level gating is an important component of loudness in long-form content, but more about the technical details of gating later. A couple of months ago, the ITU-R adopted BS.1770-3, which mostly clarified true peak. Defining loudness is truly a work in progress. Fortunately, as definitions are upgraded and updated, most digital test gear can be upgraded with the latest definitions.

The basic unit of measurement of loudness is “Loudness, K-weighted, relative to Full Scale” (LKFS), and it is similar to db Full Scale (dBFS). Zero dB plus headroom equals zero dBFS, which roughly equals zero LKFS, in the most general of terms. In fact, dBFS is derived from the most significant bit of the audio signal. LKFS is dBFS after it has been altered to model the response (not necessarily the frequency response) of the human ear at typical TV listing levels. This is the point at which the measurement slides from objective to slightly subjective. To measure LKFS levels, frequencies above 2kHz on all 5.1 channels are pre-emphasized by 4dB, while the low end is deemphasized starting at 200Hz by -2dB to -13dB at 20Hz. The RMS power average for the complete audio file is then calculated, the power in dB of each channel is added, and an LKFS number is established. LKFS is interchangeable with LUFS.

LKFS is the basic unit of measurement of nearly all loudness measurements, and it is critical to ITU-R BS.1770 and subsequent itineration and crucial to the establishment of DIALNORM. DIALNORM stands for dialog normalization, and it’s the reference for all loudness measurements. It is a metadata number that controls gain within the Dolby AC-3 system used in ATSC transmission. It is a number from 1-31, with 31 being unity gain. DIALNORM, if properly set, will result in assets being played out at -31 LKFS level, regardless of their loudness. In general, people are migrating to a DIALNORM of 24 to ease confusion downstream.

The DIALNORM value represents the measured dialog level of the signal. It is the standard by which all audio is supposed to be mixed. Typically, dialog might not be the loudest sound, so its average level is measured in LKFS, and its value is set by the operator. Typical TV DIALNORM levels run at -24 LKFS, and that seems to be the unofficial common value most producers are migrating to.

The next "Transition to Digital" DTV tutorial will explain ATSC A/85 Recommended Practices, which was revised to become the implementation document envisioned by the CALM Act.

The author wishes to thank Andrew Sachs with Volicon and Steve Smith with Broadcast Technology Consultants for their assistance and information they provided.