Make spoken audio volume more consistent
Updated: November 17, 2026
A quiet lesson followed by a loud introduction forces viewers to reach for the volume control. Diagnose the recording before simply turning every peak up.
Diagnose before editing
Listen to several speech passages at the same playback volume. If one speaker is much quieter, balance the edit first. Normalization is not noise removal and cannot restore clipped speech.
Work on a separate copy
- 1Keep a master and use a local MP4 with an audio track. The command selects the first audio and video streams.
- 2Test the example: it copies the video and re-encodes the processed audio. -16 LUFS and the other values are starting points for this example, not Dembed requirements.
- 3Compare original and output at a fixed listening level, including silence and the loudest passage.
ffmpeg -i input.mp4 -map 0:v:0 -map 0:a:0 -c:v copy -af "loudnorm=I=-16:LRA=11:TP=-1.5" -ar 48000 -c:a aac -b:a 192k speech-normalized.mp4Check before sharing
Listen for hiss brought forward, pumping and harsh peaks. This is a single-pass normalization example, not a guarantee of exact delivery loudness. For a specified target, measure the result and use a measured two-pass workflow. Upload the approved copy and check speech on phone speakers as well as headphones.
Technical reference
https://ffmpeg.org/ffmpeg-filters.html#loudnorm
Frequently asked questions
Not by itself. Room noise, clipping and poor microphone placement need separate treatment.