Radio audio processing before the encoder: levels, compression and limits

Radio audio processing for an internet station means shaping tone with EQ, evening out level with a compressor or AGC, and holding peaks under a ceiling, all before the encoder turns your mix into a stream. Aim the result at a measured loudness and a true-peak limit, not at the meter's red zone. This run sheet gives the numbers the broadcast standards publish, the starting settings one common encoder documents, and the checks that tell you it worked.

Steps
8
Reviewed
Runs on
Radio
Written by
Get Stream Hosting

By the Get Stream Hosting team

Every step was checked against the software's official documentation by independent reviewers before publishing. Last reviewed .

Pre-flight check

The signal chain, in order

Every live show passes through the same five stages, and processing belongs in the middle three. Your station's server relays the encoded stream it receives, so any level or tone problem leaving your studio reaches every listener unchanged.

  1. Source: microphone, music player, call-in audio. Set each one at a sensible gain before anything else.
  2. EQ: small tone corrections, for example taming a boomy voice.
  3. Compressor or AGC: narrows the gap between loud and quiet passages.
  4. Peak limiter: a hard ceiling so nothing overshoots after encoding.
  5. Encoder: MP3, AAC or Opus at your plan's bitrate, sent to the server.

The European Broadcasting Union's radio guideline (EBU Tech 3401) separates production from distribution: keep programmes open and dynamic while you make them, and do the dynamic adaptation for a particular platform at the distribution stage. For a small station, the processing in front of your encoder is that distribution stage.

Targets to set before touching a knob

Pick a loudness target and a peak ceiling first; every other setting serves those two numbers. Loudness is measured in LUFS with a meter that follows ITU-R BS.1770, and peaks in dBTP (true peak, which catches the overs a plain sample-peak meter misses).

SettingValueWhere it comes from
Programme loudness, default-23 LUFSEBU R 128 and its streaming supplement R 128 s2
Louder stream, if you choose onebetween -20 and -16 LUFSEBU R 128 s2 and s3
Tolerance for live programmeswithin 1 LU of the targetEBU R 128
True-peak ceiling, uncompressed audio-1 dBTPEBU R 128
True-peak ceiling ahead of common lossy codecs-2 dBTP, lower at low bitratesEBU Tech 3344

If you want the stream nearer the level of music apps, the EBU puts it in the -20 to -16 LUFS window instead. One German public broadcaster described in Tech 3401 sends its internet streams at -16 LUFS with a -1 dBTP ceiling. Whatever you pick, keep it the same for every show, so listeners never reach for the volume between DJs.

Settings to copy in BUTT

BUTT, one of the encoders that connects to our radio service, has a DSP tab with a 10-band equalizer and a dynamic range compressor, each switched on separately for the stream and for the local recording. Its manual gives this setup order, condensed here:

  1. Play the loudest source you will broadcast, usually music, and line it up with the master gain slider (it runs from -24 to +24 dB; double-click resets it to 0).
  2. Turn the compressor on and set threshold and ratio. Leave attack and release at their defaults unless you have a reason.
  3. Raise the makeup gain until the level is back where it was before compression.
  4. Talk into the microphone. Your voice should now sit closer to the music. If it is still too quiet, set makeup gain to 0 and go back to step 2.

Tip: tick Rec as well as Stream for both processors if you want the local recording processed the same way as the stream.

ContentThresholdRatioNotes
Music, or mixed music and speechfairly high2 to 3Subtle; add a separate microphone compressor for the voice
Speech onlylow5 or moreEvens out different speakers and distance from the mic
Fast transientsas aboveas aboveAggressive Mode stops the compressor averaging power over time

The equalizer moves each band between -15 and +15 dB with a Q of 2. Use cuts of a few dB rather than large boosts; every boost eats headroom the limiter then has to claw back. If you stream Opus, BUTT always resamples to 48 kHz, so set your sound device to 48 kHz and BUTT does not need to resample.

The server address, port and password go in BUTT's server settings; paste them straight from your station's control panel instead of retyping them.

AGC and two-stage levelling

An automatic gain control rides the overall level like a hand on the fader, while a compressor narrows the gap between loud and quiet passages. Tech 3401 reports that level-based AGCs and compressors react quickly but land only within about 3 LU of the target, so one broadcaster it documents routes its voice sources (microphones, phone and web callers) to channels with a loudness-based AGC directly before the fader, which finishes the job, and leaves its file-based music channels unprocessed.

  • Give voice sources such as microphones and phone callers their own levelling, separate from the music.
  • Limit how far any AGC may boost. In that example, AGC gain is capped at 3 LU to stop crosstalk between microphones building up.
  • Normalise pre-recorded items to one loudness before they air, so the live processing has less to correct. A loudness-normalised library let one broadcaster back off its AGC and compression.

What over-processing sounds and measures like

Over-processing is the signal squeezed so hard it stays loud but loses its shape. The EBU's radio supplement describes it as a "less dynamic, more squashed sound" next to a normalised chain at the same loudness. You can catch it on a meter before listeners complain:

  • Loudness Range (LRA) close to zero on music that should breathe. LRA is the gap between the loud and soft parts over time.
  • A small Peak-to-Loudness Ratio (PLR), the gap between loudness and true peak. Tech 3401 calls PLR the measure of how much compression has been applied.
  • True peaks above your ceiling after the encoder. Lossy coding creates inter-sample peaks higher than the sample values, and those turn into distortion.

Louder is not better on a stream. The EBU notes that streaming platforms increasingly normalise loudness themselves, so a crushed signal may simply be turned down to match its neighbours, keeping the damage and losing the advantage.

Checks before you go live

Record ten minutes of a typical show through your full chain, then measure the file with FFmpeg's EBU R128 scanner. This command comes from the FFmpeg filter documentation; replace input.mp3 with your recording:

ffmpeg -nostats -i input.mp3 -filter_complex ebur128 -f null -

The summary at the end prints integrated loudness (I, in LUFS) and Loudness Range (LRA, in LU). Add the filter option peak=true to the ebur128 filter to see true peak (TPK) too. You want I within a unit or so of your target, true peak under your ceiling, and an LRA that is not near zero for music. Listen on a phone speaker and on headphones as well; the meter confirms what the ears suspect.

Processing on our radio service

Our HTTPS radio streaming plans give you processing in two places. The browser studio encoder, included with Premium and Platinum, mixes your microphone and music, applies voice and music processing, encodes at up to 256 kbps (automatic or fixed quality) and warns you if the air goes silent. AutoDJ on those plans adds crossfades and silence trimming, and Platinum adds even loudness on every track, so the music between live shows already sits at one level.

  • Bitrate shapes how hard to process. The Free plan carries 32 kbps, which suits talk; Premium, 128 kbps; Platinum tops out at 320 kbps and runs 3 streams, letting phones pick a lighter one. The lower the bitrate, the more headroom the codec needs.
  • Bring the chain you already run. BUTT, Mixxx, SAM Broadcaster, RadioBOSS and other Icecast or SHOUTcast sources connect, so your existing processing stays in place.
  • Talk over music ducks the music under your voice on paid plans, so the compressor is not left doing that job.

The free plan suits a talk station or a test of your chain. Listeners on your own station app hear the same processed stream, and if you also run a live video channel, holding it to the same loudness target keeps the two level with each other.

When it goes wrong

  • Crackle on loud passages only: true peaks over the ceiling. Lower the limiter ceiling or the makeup gain, then measure again.
  • Music fine, voice buried: the compressor is set for music. Put a separate compressor on the microphone, as the BUTT manual advises.
  • Level drifts between DJs: each one uses their own encoder settings. Agree one loudness target and share the same settings.
  • Dead-air warning while someone is talking: the source reaches the encoder far too quietly. Check the source gain and the master gain before touching the compressor.
  • Still stuck: open a ticket or start a chat with your encoder name, codec and bitrate.

Questions people ask

Do I need a hardware processor for an internet station?

No. Software in the encoder, such as the EQ and compressor in BUTT, covers the basics for most small stations. Hardware helps on shows that mix speech and music: the BUTT manual suggests setting its compressor for music and adding another, typically a hardware module, between the microphones and the final mix. Start with software and measure before buying anything.

What loudness should my stream measure?

The EBU's default is -23 LUFS. If you want to sit nearer music apps, its streaming and radio supplements give a range of -20 to -16 LUFS. Pick one value and hold every show to it.

Why does my stream distort when the meter never hits 0 dB?

A sample-peak meter cannot see peaks that fall between samples, and lossy encoding creates more of them. Measure true peak with a BS.1770 meter and keep it at -2 dBTP or lower ahead of a lossy codec, and lower still at low bitrates, as EBU Tech 3344 advises.

Should talk radio be compressed harder than music?

Yes. The BUTT manual suggests a low threshold and a ratio of 5 or more for pure speech, which keeps voices intelligible and evens out different speakers. Music takes a gentler ratio of 2 to 3 with a fairly high threshold.

Does AutoDJ need the same processing as live shows?

AutoDJ plays files from your library, so normalise them to one loudness before upload. On Platinum, AutoDJ also applies even loudness on every track. Then set your live chain to the same target so the handover from AutoDJ to a live DJ does not jump.

Radio

Put your processed stream on air

Connect your encoder, set your loudness target and send a clean, even signal over HTTPS to every listener.