Making the loudness consistent
Normalising adjusts the overall level of a recording so it sits at a predictable loudness. The point is not to make things louder but to make several things match — an interview where one speaker was close to the microphone and the other was across the room is the case this exists for.
It changes level, not content. Background hiss is raised along with the voice, and a recording that was distorted when it was captured stays distorted; normalising a clipped file makes the clipping easier to hear rather than repairing it.
- Peak versus perceived loudness
- Matching the loudest instant of two files does not make them sound equally loud, because loudness is about the whole recording rather than its single highest sample.
- Do it before joining
- Normalising each part and then merging gives an even result. Normalising after merging just scales the imbalance.