FFmpeg’s ebur128 readings answer different questions: use integrated loudness for the analyzed program as a whole, momentary and short-term readings to inspect changes over time, LRA to characterize loudness range, and peak readings to inspect signal peaks. Sample peak reports the highest stored sample; true peak requires enabling its oversampled measurement. None of these values is a universal delivery target, so start with the receiving specification and choose the field that matches the decision.
Choose a measurement and inspect it
- Identify the decision first: whole-program loudness, a local change, loudness range, or a peak check. Note the delivery specification, if one exists; do not infer a target from FFmpeg’s display defaults.
- Run
ebur128on the exact audio segment you intend to assess. By default, FFmpeg logs momentary (M), short-term (S), integrated (I), and loudness range (LRA) at 10 Hz. The filter accepts double-precision floating-point audio internally, converting the input if needed. If the analysis is part of a filter chain and the original stream parameters matter downstream, check whether you needaformatoraresampleafter the filter. - To inspect changes visually, enable
video=1. The graph’s main area shows short-term loudness and its gauge shows momentary loudness by default;gauge=shorttermchanges the gauge. With video or metadata output enabled, logging becomes verbose by default, so select an appropriateframelogsetting if logs matter to your workflow. - For peaks, explicitly choose
peak=sample,peak=true, or both. Sample mode reports the highest sample (SPK). True-peak mode oversamples the input and reports true peak (TPK) and per-frame true peak (FTPK); it requires a build withlibswresample. Confirm that the build supports it before treating missing true-peak output as a measured zero. - Compare each result only with a requirement for that same measurement. A short-term reading cannot stand in for integrated loudness, and a sample peak cannot rule out a higher reconstructed peak.
FFmpeg documents the default log cadence and the graph’s distinct 400 ms momentary and 3 s short-term windows in its ebur128 filter documentation. Its peak options and exported values distinguish sample peak in dBFS from true peak in dBFS; the measurement method, not just the displayed unit, distinguishes them.
Hypothetical acceptance worksheet
The observations below are invented to show how to choose a next step, not software-test results, acceptance limits, or evidence that a particular setting will produce a given outcome.
| Question | Hypothetical observation | Field and scope | Next decision |
|---|---|---|---|
| Does the whole program meet its loudness requirement? | A four-minute piece reports I = -18 LUFS; its brief specifies a different integrated-loudness requirement. | Integrated; analyzed program | Compare with the brief’s integrated requirement; do not substitute a peak reading. |
| Where does a passage change? | Dialogue appears steady overall, but music rises during one scene. | M for 400 ms changes; S for 3 s provides a broader view | Inspect that scene on the graph or logs; decide whether the change is intended. |
| Is the program’s loudness range relevant? | The brief asks for an LRA value, and the summary reports LRA = 7 LU. | LRA; range statistic | Compare LRA with that requirement, not with integrated loudness. |
| Are stored samples near a limit? | Peak mode reports SPK = -1.2 dBFS. | Sample peak; highest sample | Use it only for a sample-peak requirement; it does not establish true peak. |
| Could reconstructed peaks differ? | With true-peak mode enabled, the summary reports a higher TPK than SPK. | True peak; oversampled estimate | Check against a true-peak requirement; keep the two peak results distinct. |
Limits and what to do next
LRA is not a moment-by-moment loudness trace, and a high or low LRA alone does not say whether a mix is suitable; use the relevant requirement and listen to the passage in context. The video graph is a display of loudness evolution, not a replacement for the reported measurements. Its green region visualizes a target-relative range; the default target is -23 LUFS, but FFmpeg allows changing it. Treat that display setting as a reference, not a delivery rule.
For mono input intended for playback on a stereo system, consider dualmono=true: FFmpeg says it compensates for a perceptual measurement effect in that case. It does not affect multichannel input. Do not enable it merely because a file has one channel; establish the intended playback first.
volumedetect and astats can help answer other level or signal-statistics questions, but they are not interchangeable with R128 integrated loudness, LRA, or oversampled true-peak measurement. If you use them as a cross-check, name the measurement and its scope explicitly. If expected TPK data is absent, check whether true-peak mode was selected and whether the FFmpeg build includes libswresample. If a result conflicts with a delivery brief, verify the analyzed segment, chosen mode, channel intent, and the brief’s exact metric before changing the mix.
Common questions
Does `ebur128` measure true peak by default?
No. Peak measurement defaults to none; enable peak=true to request oversampled true peak. Choose peak=sample for sample peak, or combine modes when both are needed.
Why are momentary and short-term values different?
They summarize different windows: 400 ms and 3 seconds. A brief event can affect the shorter view differently from the longer one; neither replaces the full-program integrated result.
Is the default `-23 LUFS` target a requirement?
No. It is the default relative-zero target for visualization, and can be changed. Follow the actual delivery specification rather than adopting a display default.
Sources and next step
This guide interprets public documentation and uses explicitly hypothetical acceptance examples, not application tests. Controls and availability can change; verify the cited manual against your installed version.
For work created in Imagild, carry the input requirements and acceptance checklist into the film workspace. The external software controls discussed above belong to their respective products.