To normalise audio loudness to EBU R128, measure the programme's integrated loudness in LUFS and adjust it to the target of -23 LUFS, while keeping true peaks at or below -1 dBTP. In FFmpeg, the loudnorm filter does this, and running it in two passes - one to measure, one to apply - gives the most accurate and natural-sounding result.
This guide explains what LUFS and true peak mean, which targets suit broadcast, streaming and podcasts, and how to normalise video audio correctly with FFmpeg without harming quality.
What EBU R128 is
EBU R128 is a recommendation from the European Broadcasting Union for measuring and normalising loudness. Before it, programmes were levelled by peak meters, which led to adverts and some shows sounding much louder than others. R128 replaced peak-based levelling with a measurement of perceived loudness, based on the ITU-R BS.1770 algorithm, so that everything plays back at a consistent level.
It defines three main measurements:
- Integrated loudness - the average loudness over the whole programme, measured in LUFS (Loudness Units relative to Full Scale). The R128 target is -23 LUFS, with a tolerance of around plus or minus 0.5 LU for most programmes.
- Loudness range (LRA) - how much the loudness varies over the programme, measured in LU. A film has a wide range; a podcast or advert usually has a narrow one.
- Maximum true peak - the highest level the waveform reaches, including peaks that fall between samples. R128 sets a limit of -1 dBTP.
One LU is the same size as one decibel, so raising a file by 3 LU makes it 3 dB louder.
Which target to use
R128's -23 LUFS is the standard for European broadcast. Other destinations use different levels, and many streaming services publish their own normalisation targets:
| Destination | Typical integrated target | Typical true peak |
|---|---|---|
| European broadcast (EBU R128) | -23 LUFS | -1 dBTP |
| US broadcast (ATSC A/85) | -24 LKFS | -2 dBTP |
| Online video and music streaming | Roughly -14 to -16 LUFS | -1 dBTP |
| Podcasts | Around -16 LUFS (stereo) | -1 dBTP |
LKFS and LUFS are the same unit under different names. Streaming services generally turn loud content down to their reference level, and some do not turn quiet content up, so for online delivery a master somewhat louder than -23 LUFS is usual. Platforms do not always publish exact figures and they change, so treat the streaming numbers as typical rather than official. For YouTube-specific export advice, see the best video export settings for YouTube.
Normalising versus compressing
Loudness normalisation changes the overall level of a file - like turning a volume knob. It does not change the balance between quiet and loud parts. If a programme has a very wide loudness range (quiet dialogue and loud explosions, for example), simply normalising it may leave the dialogue too quiet. In that case, the loudness range has to be reduced with compression, which FFmpeg's loudnorm can do automatically in its dynamic mode, or which you can handle more carefully in an audio editor.
Normalising with FFmpeg
Single pass (quick)
ffmpeg -i input.mp4 -c:v copy -af loudnorm=I=-23:TP=-1:LRA=11 -c:a aac -b:a 192k -ar 48000 output.mp4
This copies the video untouched and normalises the audio towards -23 LUFS. Single-pass mode cannot see the whole file in advance, so it adjusts dynamically as it goes. That works, but it can make subtle level changes and lands less precisely on the target.
Note the -ar 48000: loudnorm works internally at a very high sample rate (192 kHz), and without setting the output rate you can end up with an unexpectedly high-sample-rate file.
Two pass (accurate)
First, measure the file and print the statistics:
ffmpeg -i input.mp4 -af loudnorm=I=-23:TP=-1:LRA=11:print_format=json -f null NUL
At the end of the output, FFmpeg prints values including input_i, input_tp, input_lra, input_thresh and target_offset. Feed those into the second pass:
ffmpeg -i input.mp4 -c:v copy -af loudnorm=I=-23:TP=-1:LRA=11:measured_I=-30.2:measured_TP=-4.5:measured_LRA=7.1:measured_thresh=-40.6:offset=0.3:linear=true -c:a aac -b:a 192k -ar 48000 output.mp4
Replace the example numbers with your own measurements. With linear=true, FFmpeg applies a single fixed gain change when it can, which preserves the original dynamics exactly. If the required gain would push peaks over the limit, or the loudness range is wider than your LRA target, it falls back to dynamic mode. On macOS or Linux, replace NUL with /dev/null.
Just measuring
To check a file's loudness without changing it, the ebur128 filter prints integrated loudness, loudness range and, with peak=true, true peak:
ffmpeg -i input.mp4 -af ebur128=peak=true -f null NUL
Tips for good results
- Normalise after editing. Integrated loudness is a property of the finished programme, so normalise the final mix rather than individual clips.
- Use a decent audio bitrate. Normalisation means re-encoding the audio. AAC at 192 kbps or more for stereo, or a lossless codec such as FLAC or PCM for masters, avoids adding audible artefacts.
- Watch the true peak. Lossy encoders can create peaks slightly higher than the original, which is one reason the limit is -1 dBTP rather than 0.
- Be careful with quiet sources. Raising a very quiet recording by a large amount also raises background noise.
- Check surround downmixes. If you downmix 5.1 to stereo, measure and normalise the stereo result, not the original.
If the audio drifts out of sync after re-encoding, which can happen when sample rates or timestamps change, see fixing audio out of sync after converting. And for processing whole folders of files to the same loudness target, batch converting videos on Windows covers the workflow.
Doing it with Movie Encoder
Movie Encoder includes loudness normalisation to EBU R128 in its audio settings, alongside volume, delay correction, fades, downmix, channel layout, sample rate and bitrate controls. Audio can be encoded as AAC, MP3, Opus, Vorbis, FLAC, ALAC, PCM, AC-3 or E-AC-3, while the video stream is copied or encoded in the same job.
Because normalisation can be part of a saved preset in the batch queue or a watch folder, you can apply the same loudness target to every file automatically, and the command preview shows the exact FFmpeg arguments used.
Frequently asked questions
What is the EBU R128 loudness target?
EBU R128 specifies an integrated loudness of -23 LUFS with a maximum true peak of -1 dBTP. It is the standard for European broadcast.
Should I use -23 LUFS for YouTube or podcasts?
Usually not. Online platforms and podcast apps typically play back at around -14 to -16 LUFS, so a master at -23 LUFS may sound quiet. Pick the target that matches where the file will be played.
Does loudness normalisation reduce audio quality?
A simple gain change does not alter the sound, but the audio has to be re-encoded, so use a generous bitrate or a lossless codec. If the filter needs to compress the dynamics to hit the target, that does change the character of the mix.
