Loudness normalization makes a batch of files sound equally loud: podcast episodes recorded at uneven levels, interview clips from different sources pasted together, or a track that has to reach -14 LUFS before it is uploaded. It is the “Loudness normalization” mode of the “Volume” tab in the audio/video workbench, and it runs ffmpeg's loudnorm filter as an EBU R128 two-pass job.

The three presets are Podcast / Voice at -16 LUFS with a -1.5 dBTP true-peak ceiling and 11 LU loudness range; Music streaming at -14 LUFS, -1 dBTP, 11 LU; and Broadcast EBU R128 at -23 LUFS, -2 dBTP, 7 LU. “Two-pass” means the whole file is walked once with measurement only (integrated loudness, true peak, LRA, threshold), and the measurements are fed back to loudnorm in pass 2 for linear gain, which pumps less than the single-pass dynamic mode. For video, only the audio is re-encoded and the picture is copied.

1. 01Add files and select them— switch to the “Volume” tab and click “Loudness normalization”.
2. 02Pick a preset— “Podcast / Voice”, “Music streaming” or “Broadcast EBU R128”; the panel writes out the target LUFS and the true-peak ceiling.
3. 03Start normalizing (two passes)— the queue first shows “Loudness analysis (pass 1 of 2)”, and when it finishes it submits “Loudness normalization -16 LUFS (pass 2 of 2)” automatically.
4. 04Download— when pass 2 completes, download <original-name>_loudness.mp3 and similar; video keeps its source container, for example <original-name>_loudness.mp4.

Two passes take time

— the whole file is walked twice, so long audio is clearly slower than “Volume adjust”; parameters cannot be changed between the passes.

Measurement fallback

— with long silent stretches or a very short file, loudnorm may not print its measurement JSON; the panel then suggests the dynamic single-pass mode, whose loudness will drift a little.

Sample rate

— loudnorm works internally at 192 kHz and the tool resamples back to the source rate (48 kHz when unknown).

Audio is always re-encoded

— the track inside a video is re-encoded for its container (AAC 192 kbps in MP4); lossless sources can use FLAC / WAV input and keep lossless output.

What it does not do

— it is not a plain volume boost (that is “Volume adjust”), it does not compress dynamics to one flat level, and the LUFS target cannot be customized.

Both passes run locally in the browser through ffmpeg.wasm and files are never uploaded. On first use the engine, about 31 MB, is downloaded from a CDN (the site asset domain or jsDelivr); that request carries no audio or video content.