Audio loudness (LUFS) check and normalization

What it does

Checks with real numbers whether the loudness of audio or video files fits the destination.

• Measures integrated loudness (LUFS), true peak (dBTP), loudness range (LRA) and duration
• Compares with targets: YouTube and Spotify (-14 LUFS), Apple Music and podcasts (-16 LUFS), mono podcasts (-19), broadcast (EBU R128 -23, ATSC A/85 -24, ARIB TR-B32 -24), or any number you give
• Verdicts: too loud, too quiet, peak over the ceiling, or unmeasurable (silence), plus the gain needed to reach the target
• Checks whole folders, reads the audio track of video files, writes a JSON report
• On request, writes normalized copies into a separate folder with ffmpeg's two-pass loudnorm and re-measures them (video streams are copied, not re-encoded)

How to use

python scripts/loudness.py targets
python scripts/loudness.py check episode.wav music/ --target podcast --report loudness.json
python scripts/loudness.py normalize episode.wav --target podcast --out normalized/

Ask your agent whether a file is loud enough for YouTube. It explains the measurements in plain words and, if you want, creates a normalized copy.

Requirements

• Python 3
• ffmpeg (with ffprobe). If it is not found, the script stops and shows how to install it.

Caveats

• Platform targets can change; check the current official spec for important deliveries.
• If reaching the target level would push peaks over the ceiling, ffmpeg compresses the dynamics automatically (shown as "dynamic" in the result).
• Re-encoding to MP3/AAC loses a little quality. Source files are never modified or overwritten.