Normalize loudness
What Normalize loudness actually does
Every service that plays audio back turns it down to its own loudness target, so a file mastered louder than that target is not louder on arrival, only more compressed on the way. Getting this right means knowing two numbers: where the recording sits and where it should sit. Most browser normalizers show neither and return a file. This one runs a real loudness analysis, reports the integrated loudness, true peak and loudness range it measured, and then applies the correction to the target chosen here.
How to use it
- Choose an audio file. The waveform is drawn locally.
- Pick the method: integrated loudness for anything destined for a streaming service, or peak and average level for simpler cases.
- Set the target and run.
- Read the measured loudness of the input beside the target, and compare the result against the source.
Useful for
- Bringing a podcast episode to a consistent level across a series.
- Finding out where a recording actually sits before deciding whether it needs anything at all.
- Matching a set of files to one target so a playlist does not jump between tracks.
Limits worth knowing
- Integrated loudness is a measurement over a whole recording; a short clip or a clip that is mostly silence produces a figure that is technically correct and practically misleading.
- Where the correction would push the true peak past its ceiling, the loudness filter falls back to a dynamic mode, so the reported measurement describes the input rather than guaranteeing the output.
- Loudness normalization does not repair clipping that is already in the recording.
Questions people ask
Is the file uploaded?
No. The analysis and the correction both run in this tab.
What target should I use?
Streaming services publish their own; the default here is the common one for speech and music delivery. The right answer depends on where the file is going.
Why is the measured number different from the target after processing?
The reported measurement is of the input. It is shown so the size of the correction is visible rather than hidden.