Normalize audio loudness

LUFS in plain words

The old way to normalize audio was to find the single loudest moment in a file and turn the whole thing up until that moment touched the ceiling. It is simple, but two files treated that way can still sound very different: one sharp drum hit can stop a whole song from being raised.

Loudness normalization measures what you actually hear instead. LUFS stands for Loudness Units relative to Full Scale. The measurement takes in the whole recording, gives extra weight to the frequencies the ear is most sensitive to, and comes up with one number for how loud it feels overall. The numbers are negative, and closer to zero is louder: −14 LUFS is louder than −16, which is louder than −23.

This way of measuring comes from EBU R128, the loudness standard of European broadcasters. The second figure, −1.5 dBTP, is a ceiling for true peaks, the highest points the wave reaches once it is played back. Staying under it leaves room, so the file does not clip when it is compressed to MP3 or played through a small speaker.

Why −16 LUFS, and when you need another level

Different places aim for different levels. These are the targets you meet most often:

Octoolo's Even out loudness always aims at −16 LUFS, the common level for spoken audio. It suits podcasts, lectures, voice notes and mixes of songs taken from different albums, and it is loud enough to hear clearly on a phone speaker without pushing music into distortion. There is no setting for another target, so if a broadcaster or platform asks you to deliver at exactly −23 LUFS, use a tool that lets you set the number and checks the result.

What evening out does to a recording

Octoolo uses FFmpeg's loudnorm filter for this. It aims for an even level across the whole recording while keeping a natural difference between loud and soft passages, so a talk where one speaker mumbles and another booms comes out closer together without being squashed flat.

• Each sound is measured on its own. That is what makes two interviews recorded separately sound like one conversation once they are joined.
• Noise comes up with quiet voices. Switch on Reduce background noise as well: Octoolo takes out the noise first, then evens out what is left.
• The copy is made once. It is kept on your computer, so trims, splits, fades and speed changes work on the evened-out sound straight away, and your original file is untouched.

There is no batch run over a folder. To even out a series of files, put them in the Audio Editor, switch it on for each and export them together as one file, or open them one by one to save them separately.

Where evening out helps most

• Podcasts with guests who joined from different rooms, microphones and headsets.
• Playlists and mixes built from albums decades apart, where older masters are much quieter than new ones.
• Lecture and sermon series recorded in different halls, so listeners stop reaching for the volume button.
• Phone voice memos, which swing from whisper-quiet to loud depending on how the phone was held.
• Voice-overs that need a steady level before music goes underneath them on another track.

In each case the aim is the same: the listener sets the volume once, on a phone, in the car or at a desk, and never has to touch it again. Combine it with a gentle fade at each join and a series of recordings plays like one continuous show.