Listening examples
Three excerpts of real programme, each in three versions: the original, and the output of the realtime adjuster in its two 2 ms modes. Switch between them while it plays. The switch keeps the position, so each time you hear the same moment in another version.
In the first two excerpts the original is raised or lowered by one fixed amount, so that over the excerpt it is as loud as mode 1. Otherwise the louder version tends to sound better, whatever was done to it; matched, what is left to hear is how the level moves within the excerpt. The third plays its original as published, because there the change in level is the point: an adjuster should do no more to the sound than the material needs. This programme needs one steady gain to reach the target, and that is what the adjuster gives it. Holding a level steady when nothing calls for movement matters as much as riding one that moves.
Both delay the audio by 2 ms. The standard mode, with a 50 ms delay, rides the level the same way and differs in its peak stage. Each excerpt ends with advice on the mode that suits it.
Press play to load an excerpt; nothing is downloaded before that. Keys 1, 2 and 3 switch the version of the excerpt that is playing, and a click on the graph jumps to that moment. Headphones or good speakers help.
Quiet dialogue, then a passage that is quieter still. The speech detector hears speech for most of the first half (the grey band in the lower graph). Mode 2 lifts the voices, and by 0:30 it adds the full 12 dB. Mode 1 holds the whole programme rather than the voices and leaves them close to where they were: its gain stays near 0 dB until 0:30, then climbs to about +10 dB.
This excerpt plays quieter than the other two: about -22 LUFS in mode 1 and -19 LUFS in mode 2, not -16. In places the original is more than 20 LU under the target, and the adjuster adds at most 12 dB. The original is matched to mode 1, so mode 2 plays about 3 dB louder than both: that difference is the dialogue being lifted.
Advice: mode 2, where the target is meant for the dialogue, as in ATSC A/85. For an EBU R128 target, mode 1: R128 measures the whole programme.
Elephants Dream, © copyright 2006 Blender Foundation / Netherlands Media Art Institute, CC BY
Music and effects, with no dialogue. The quiet opening is lifted by the full 12 dB; the loud stretch around 1:05 is turned down by up to 5.5 dB, and the level comes back up as the music settles. Modes 1 and 2 give identical output here: the speech detector found no dialogue to hold.
Advice: mode 1. On music a speech detector can take a vocal for dialogue, and mode 2 would then hold the vocal at the target.
Spring, © Blender Foundation | cloud.blender.org/spring, CC BY
Evenly mixed speech, published at about -27 LUFS, 11 dB under the target. The adjuster's work here is the level: both modes raise it to the target by a steady 11 to 11.5 dB. Beyond that gain, the peak stage acts on a handful of moments of about 10 ms each, by at most 2 dB, to hold the -1 dBTP ceiling. Apart from the step in level, the three versions should sound alike.
The original plays as published, about 11 dB quieter than the adjusted versions. Set your volume on mode 1 or mode 2 first: switching to the original is a large drop, and switching back an equally large jump.
Advice: either. Mode 2 holds the dialogue and mode 1 the programme, and here the programme is the dialogue.
FOMC press conference, 16 September 2026. Source: Federal Reserve Board (public domain).
These are examples to listen to, not measurements. How closely the adjuster holds its target, case by case, is in the measured results, graded by a published test method.