Audio Analysis
Normalize / Gain
Brings every clip to the same level: its peak to a target (e.g. -1 dBFS), its average loudness (RMS) to a target, or a fixed gain — never clipping, and never raising near-silence into loud noise.
When to use it
Quiet recordings score lower in models and are missed by thresholds; normalising first makes clips from different devices comparable.
In a pipeline
Audio File InputNormalize / GainAudio Event Classifier
Settings
| Setting | Type | Required | Default or hint |
|---|---|---|---|
| Normalise | select | — | peak |
| Target level (dBFS) | number | — | -1 |
| Target loudness (dBFS) | number | — | -20 |
| Gain (dB) | number | — | 6 |
| Raise by at most (dB) | number | — | 30 |
| Never clip (lower the gain if peaks would pass 0 dBFS) | checkbox | — | true |
Example
Loudness to -20 dBFS
- Normalise=Loudness (RMS) to a level
- Target loudness (dBFS)=-20
| File Name | Gain (dB) | Peak (dBFS) | RMS (dBFS) |
|---|---|---|---|
| quiet.wav | 14.2 | -3.1 | -20 |
| loud.wav | -4.5 | -0.9 | -20 |
Worth knowing
- Raise by at most stops a silent clip being lifted 60 dB into hiss.
- With Never clip, a loudness target that would push peaks past 0 dBFS is lowered just enough, so RMS can end a little below the target.