What the loudness meter is telling you
A loudness meter reads four numbers, and each answers a different question. All of them come from ITU-R BS.1770, which filters the audio with a K-weighting curve before averaging it, so bass-heavy and bright material are compared the way the ear compares them. EBU Tech 3341 sets the time windows. LUFS and LKFS are the same unit under two names; a difference between two readings is expressed in LU.
| Reading | Window | Use it for |
|---|---|---|
| Momentary (M) | Sliding 0.4 s, not gated | Instant feedback while you talk or ride a fader |
| Short-term (S) | Sliding 3 s, not gated | Levelling voice; the EBU recommends it for self-op studios |
| Integrated (I) | Whole programme, gated | The single number you normalise each track to |
| Loudness Range (LRA) | Spread of short-term values | How much a track swings between soft and loud passages |
Integrated loudness ignores silence: blocks of 400 ms quieter than -70 LUFS are dropped, then anything more than 10 LU below the remaining average is dropped too. That is why a quiet intro does not drag a track's figure down. LRA is the distance between the 10th and 95th percentile of the short-term readings, so a single fade-out or one loud hit does not dominate it. The EBU advises against quoting LRA for anything shorter than a minute, such as a station ID.
True peak (dBTP) is measured on an oversampled copy of the signal, because the real waveform can rise between two samples. A file whose sample peaks sit at 0 dBFS can still clip once it is decoded, so you set your ceiling in dBTP, not in sample peak.
Which target to pick
Choose one integrated target for the whole station and stay on it. The EBU publishes the figures below; your choice depends on where people listen.
| Situation | Integrated target | True-peak ceiling | Where it comes from |
|---|---|---|---|
| Producing programmes and exchanging them | -23.0 LUFS (live shows may sit within ±1.0 LU) | -1 dBTP | EBU R 128 |
| Radio or stream distribution, default | -23.0 LUFS | Per EBU Tech 3344 | EBU R 128 s3 and s2 |
| Stream heard on phones and in noisy places | Interim range -20.0 to -16.0 LUFS | Leave more room at low bitrates | EBU R 128 s2, Tech 3401 |
| FFmpeg loudnorm with no options set | -24.0 LUFS | -2.0 dBTP | FFmpeg documentation |
Most internet stations compete with music apps on phones, which is exactly the case the interim range was written for. A station playing classical music or drama has good reason to sit lower, near -23 LUFS, and keep its dynamics. Whatever you pick, write it on the studio wall and give it to every DJ.
Lossy encoding adds overshoot. EBU Tech 3344 notes that relatively low bitrates generally produce more of it, so a stream at a low bitrate needs a lower limiter setting than the same audio at a high one. Measure the decoded stream, not only your files.
Levelling the music library
Normalise the music before it reaches air, rather than squashing an uneven playlist with a heavy limiter at the output. EBU Tech 3401 recommends exactly that order: level each file, or store its loudness as playback metadata, and keep strong processing for distribution.
- Copy the originals to a separate folder first and confirm the copy opened and played before you change anything. Never write normalised files over the only copy.
- Measure each track with FFmpeg's
ebur128filter (command below). Addpeak=trueto the filter, asebur128=peak=true, and the summary also reports the true peak. - For each track, run FFmpeg's
loudnormfilter twice. First pass: set your targetI,TPandLRAandprint_format=json, and read the measured values it prints. Second pass: give those back asmeasured_I,measured_LRA,measured_TPandmeasured_threshwithlinear=true, so the track is turned up or down as a whole instead of compressed. - Set an output sample rate with
-ar. The documentation warns that in dynamic mode loudnorm upsamples to 192 kHz to find true peaks. - Measure the new file again. Integrated should land on your target and true peak below your ceiling.
ffmpeg -nostats -i input.mp3 -filter_complex ebur128 -f null -
Linear mode has two conditions: the target LRA must not be lower than the track's own LRA, and the gain must not push true peak over the target TP. If either fails, loudnorm falls back to dynamic mode, which compresses. A very quiet, very wide classical recording is the usual case; set a higher LRA target for it, or accept a lower level for that piece.
Then upload the levelled files to your AutoDJ library in the control panel, and keep the untouched originals in your backup folder.
Keeping live voices on target
Speech is where most level jumps come from: a presenter who leans in, a guest on a quiet laptop, a caller on a phone. Set the meter to the target and read it as relative LU, with your target shown as 0 LU.
- Watch short-term loudness and keep the voice within ±2 LU of 0. On momentary loudness the EBU suggests a ±3 LU band instead.
- Set mic gain so the voice sits on target at a normal speaking distance, then leave it alone; ride the fader, not the preamp.
- Put a true-peak limiter last in the chain at your ceiling, so a laugh or a dropped mic cannot clip the encoder.
- One EBU member broadcaster plays music at 0 LU and sets presenters 1 to 3 LU below it. Try it if your voice sounds pushed over the music.
If you broadcast from your own encoder, set levels in your mixer or processing before the encoder; our support desk can check the encoder settings with you.
Checks before you call it done
- Play a mixed hour: music, a station ID, an ad, a voice link.
- Point FFmpeg's
ebur128at your listen link instead ofinput.mp3, and stop it after ten minutes or more with FFmpeg's-toption. - Read the summary. Integrated should be within about 1 LU of your target, and true peak (with
peak=true) below your ceiling. - Listen to the joins on a phone speaker and on headphones. Nobody should need the volume control between items.
If integrated reads low but true peak is at the ceiling, the material is too dynamic for the target; lower the target or add gentle compression. If one item jumps out, measure that file on its own.
What your station plan does for you
Get Stream Hosting serves every station over HTTPS from Icecast or SHOUTcast, and the SSL radio streaming plans differ in what they handle for you. On the free plan you relay a single live feed at 32 kbps from the Icecast or SHOUTcast encoder you already use, so all levelling happens on your side; at that bitrate leave generous true-peak room.
- Premium: AutoDJ at 128 kbps with crossfade and silence trimming, a browser studio encoder with voice and music processing and a warning when the air goes silent, automatic music ducking whenever you speak, and dead-air alerts. Level your library before uploading.
- Platinum: even loudness on every track in AutoDJ, cue, intro and outro points, up to 320 kbps and 3 streams so phones can take a lighter bitrate. Since lighter streams are lower bitrates, check true peak on each one.
- On any plan you can see your current audience live; on Platinum you also get tune-outs for each track, so you can see whether listeners leave on a loud ad or a quiet link.
The plan page does not state the loudness target the even-loudness feature uses, so measure your listen link once after switching it on and set your live voice level to match what you read. For the same discipline on video channels, see live video streaming.
When a target will not fix it
- A track that was clipped in mastering stays distorted at any level; replace the file.
- A remote DJ sending an already limited, very loud feed will still sound flat after it is turned down. Agree the target with them before the show.
- Loudness normalisation cannot fix a poor speech-to-music balance inside one recording; that is a mixing job.
- Players that apply their own volume levelling may move your stream again. Your job is a consistent stream; theirs is out of your hands.