Normalize Audio Loudness (EBU R128)

@jiixyj1 VersionEnhance

Measures the integrated loudness of an AudioFrame, scales it to a target LUFS, and limits the resulting true peak, emitting a leveled AudioFrame with unchanged sample rate and channel count for consistent downstream levels.

How it fits

Typical backends

A file source is leveled to a target loudness before speech recognition so transcribe_audio_faster_whisper sees consistent input levels.

The leveled audio is sent to a file writer for broadcast or podcast delivery at a fixed loudness spec.

Caveats

AccuracyLoudness is measured over each whole frame, so short frames give noisy estimates; aggregate audio into longer turns upstream for stable leveling.
LimitationThis applies a single integrated-loudness gain, not dynamic-range compression, so the dynamic range inside the frame is preserved.
Parameter interactionThe peak limit in true_peak_dbfs is enforced after the loudness gain is applied, scaling the buffer back down when the post-gain peak would exceed the ceiling; choose target_lufs to match the delivery platform.
FallbackFrames below the silence gate, and empty or zero-sample-rate frames, are passed through untouched so noise is never amplified; an internal measurement error also passes the frame through unchanged.
State lifetimetarget_lufs and true_peak_dbfs are captured once at first use; changing them requires a redeploy.

Versions

19b96a4elatestdefault
Automated release
6/25/2026