How to measure loudness to a broadcast target

Broadcast loudness is one number for the whole programme: its integrated loudness, measured to ITU-R BS.1770. Europe delivers at −23 LUFS (EBU R128), the US at −24 LKFS (ATSC A/85), and both cap the true peak. Measuring it takes one command or one dropped file. Hitting it takes one gain change, then a second measurement to prove it.

The targets, from the documents themselves:

These are the defaults for content delivered without other arrangements. A broadcaster's own delivery spec can be stricter: if you were sent one, it wins.

ffmpeg -hide_banner -nostats -i melody.wav -af ebur128=peak=true:framelog=quiet -f null -

  Integrated loudness:
    I:         -16.3 LUFS
    Threshold: -26.4 LUFS

  Loudness range:
    LRA:         0.5 LU
    Threshold: -36.1 LUFS
    LRA low:   -16.5 LUFS
    LRA high:  -16.0 LUFS

  True peak:
    Peak:       -1.0 dBFS

Bring it to the target

Integrated loudness moves decibel for decibel with gain. The track measured −16.3 LUFS and the target is −23, so it needs −6.7 dB:

ffmpeg -hide_banner -loglevel error -i melody.wav -af volume=-6.7dB -ar 48000 melody-r128.wav

Then measure the new file with the same ebur128 command:

  Integrated loudness:
    I:         -23.0 LUFS
  ...
  True peak:
    Peak:       -7.7 dBFS

On target, and the true peak went down by the same 6.7 dB, well under −1 dBTP. For ATSC A/85 the gain is the target minus the reading in the same way: −24 − (−16.3) = −7.7 dB. -ar 48000 writes the 48 kHz that broadcast delivery expects.

The two-pass loudnorm route

FFmpeg's loudnorm filter does the measuring and the gain in one tool. The first pass measures and prints JSON:

ffmpeg -hide_banner -nostats -i melody.wav -af loudnorm=I=-23:TP=-1:LRA=7:print_format=json -f null -

{
	"input_i" : "-16.57",
	"input_tp" : "-1.00",
	"input_lra" : "1.60",
	"input_thresh" : "-26.69",
	"output_i" : "-23.08",
	"output_tp" : "-7.87",
	"output_lra" : "0.60",
	"output_thresh" : "-33.19",
	"normalization_type" : "dynamic",
	"target_offset" : "0.08"
}

The second pass feeds the input_* values back as measured_*, with linear=true so it applies one gain change instead of compressing the audio:

ffmpeg -hide_banner -nostats -i melody.wav -af loudnorm=I=-23:TP=-1:LRA=7:measured_I=-16.57:measured_TP=-1.00:measured_LRA=1.60:measured_thresh=-26.69:offset=0.08:linear=true:print_format=summary -ar 48000 melody-loudnorm.wav

Input Integrated:    -16.3 LUFS
Input True Peak:      -1.0 dBTP
Input LRA:             0.6 LU
Input Threshold:     -26.4 LUFS

Output Integrated:   -22.7 LUFS
Output True Peak:     -7.4 dBTP
Output LRA:            0.7 LU
Output Threshold:    -32.8 LUFS

Normalization Type:   Linear
Target Offset:        -0.3 LU

It landed at −22.7, and ebur128 on the output agrees. The first pass had read the 12-second track as −16.57 where ebur128 (and the second pass itself) read −16.3, so the gain came out 0.3 dB short. On the same track looped to two minutes, the first pass read −16.4: the gap shrinks on longer material, but measure the output either way. 0.3 LU is outside R128's ±0.2 LU QC tolerance, so on this file the plain volume route was the one that delivered.

Two more things about loudnorm. Without linear=true, or when the target can't be reached linearly, it compresses dynamically. And in dynamic mode it writes 192 kHz audio: our test output came out at 192000 Hz. Always give it -ar.

Measured with FFmpeg 7.1.5 (Debian). The browser meter's code, run on the same track, read −16.3 LUFS integrated and −1.0 dBTP true peak.

In the browser

The meter is built to BS.1770-4 and decodes everything at 48 kHz, where the recommendation's filter coefficients apply exactly. Its tests reproduce EBU Tech 3341 cases 1 to 6 to within 0.1 LU and the Tech 3342 loudness range cases. There is no ATSC entry in the target list: for A/85, choose No target and read the integrated figure against −24 yourself.

Once the integrated reading is on target and the true peak is under the ceiling, the file is ready to deliver. Keep the ebur128 summary of the final file: it is the figure a broadcaster's QC will check against.