FFmpeg - audio - short video - creators - LUFS
4 min lectura

Quiet voice in short videos? Measure LUFS and adjust with FFmpeg

Measure and normalize voice audio with FFmpeg, then check the encoded file. Tested commands and clear limits for creators.

Quiet voice in short videos? Measure LUFS and adjust with FFmpeg

Two short videos were recorded with the same microphone, yet viewers must turn up the phone for one of them. Peak level alone may look fine in both files. Perceived loudness across the whole clip is a different measurement. Before publishing a voice-led video, FFmpeg can measure integrated loudness, normalize a copy and give you a clear number to compare with a listening test.

This guide uses FFmpeg's documented loudnorm filter. I ran the commands locally on a six-second synthetic tone: the measured input was −38.2 LUFS and the encoded output was about −16.1 LUFS. A tone can confirm that the pipeline runs, but it cannot tell us whether real speech, music or a social platform's encoding sounds good. The −16 LUFS target here is an editorial starting point for comparing clips, not an official requirement from Instagram, YouTube or TikTok.

Read the three measurements first

FFmpeg describes loudnorm as EBU R128 loudness normalization. I is integrated loudness in LUFS, a whole-clip measure of perceived level. TP is maximum true peak in dBTP, a ceiling useful for avoiding clipping after encoding. LRA describes variation in loudness over time. None of these values replaces listening to the opening, the loudest section and the ending.

For a series of creator videos, a consistent reference is usually more useful than making every clip as loud as possible. Setting TP=-1.5 specifies a ceiling; it does not force the result to reach exactly −1.5 dBTP. My synthetic test ended around −12 dBTP while reaching the integrated-loudness target.

1. Measure your original clip

Install FFmpeg from the official download page. Replace input.mp4 with your file:

ffmpeg -hide_banner -i input.mp4 \
  -af "loudnorm=I=-16:TP=-1.5:LRA=11:print_format=summary" \
  -f null -

This command does not create a video. It prints Input Integrated, Input True Peak and other values in the terminal. It also calculates a hypothetical output, so use the lines beginning with Input when describing the existing file. If the video has no audio track, fix that before running this workflow. If it is already close to your chosen reference and sounds clear, processing may be unnecessary.

2. Export a review copy

When the container and codecs are compatible, you can preserve the video frames and recode only the audio. This example writes MP4 video with AAC audio:

ffmpeg -hide_banner -i input.mp4 \
  -map 0:v:0 -map 0:a:0 \
  -c:v copy -af "loudnorm=I=-16:TP=-1.5:LRA=11" \
  -ar 48000 -c:a aac -b:a 192k \
  output-review.mp4

The two -map selectors require one video and one audio stream; choose the correct tracks if your file has several. -c:v copy leaves the images alone. -ar 48000 sets an explicit sample rate, useful because FFmpeg notes that true-peak detection may upsample internally. Keep your original and compare the copy before replacing anything.

3. Measure the encoded result, then listen

Run the measurement on output-review.mp4. Its Input Integrated and Input True Peak now describe the actual encoded output. Then listen on a phone at a normal level, including quiet speech, the loudest music and the ending. If music masks your voice, lower it in the mix; normalizing the combined track does not separate the two.

ffmpeg -hide_banner -i output-review.mp4 \
  -af "loudnorm=I=-16:TP=-1.5:LRA=11:print_format=summary" \
  -f null -

In the local synthetic test, the original WAV read -38.2 LUFS. After AAC export, the final file measured about -16.1 LUFS and -12.0 dBTP. That demonstrates the command path, not an ideal sound for speech. Silence, background noise, compression and a voice's natural dynamics can all change the outcome.

When to use two passes

loudnorm supports single-pass operation and a two-pass workflow using measurements from the first pass. If a client requires a precise delivery specification for a whole series, follow FFmpeg's documented two-pass method and verify each final file. For everyday creator clips, start by fixing the recording and the voice/music balance. A distorted recording cannot be repaired simply by changing LUFS.

Try this with two real videos you intend to publish together. Measure both, normalize a copy of the quieter one, and compare them on the same phone. Keep the version that is easier to understand, not simply the one with the most attractive number. If your next problem is file size for messaging, see our separate compression guide after fixing the mix.

Comentarios

Aún no hay comentarios. Sé el primero.

Deja un comentario