Why does FFmpeg loudnorm switch to dynamic processing even when linear mode is enabled?

Why linear=true can become dynamic

According to FFmpeg Filters: loudnorm, linear mode scales the source audio by a constant amount. That approach is possible only when the filter has enough measured information to predict the result safely. If any required measurement is absent, or if the requested targets cannot be met by simple scaling, loudnorm falls back to dynamic normalization.

A dynamic result can therefore follow the documented fallback rather than an ignored option. Check the conditions individually and review the reported mode. Do not treat either a requested mode or a successful render as proof that the output satisfies an external delivery specification.

The prerequisites to check

For the normalization pass, supply all four source statistics using their actual option names: measured_I is the measured integrated loudness, measured_LRA is the measured loudness range, measured_TP is the measured true peak, and measured_thresh is the measured threshold. These describe the input; I, LRA and TP set the requested output targets. Obtain the measurements from an earlier analysis of the same audio and keep the four values together. Supplying only measured_I and measured_TP leaves measured_LRA and measured_thresh missing.

The target loudness range must not be lower than the source loudness range. A linear gain change cannot compress the source’s variation over time, so asking it to reduce that range conflicts with the method. Also check the predicted true peak: the change in integrated loudness must not produce a peak above the chosen maximum true-peak target. If it would, loudnorm reverts to dynamic processing. The target values and measurements must also be coherent and tied to the same source audio.

Illustrative exercise: two measured inputs

Consider this clearly fictional finishing scenario. Input A is a short product demonstration whose earlier analysis reports a source loudness range below the requested target range. Its measured integrated loudness, measured range, measured true peak, and measured threshold are all supplied, and applying the requested gain would keep the predicted true peak below the target ceiling. Linear mode has a valid basis for this input.

Input B is a voiceover-and-music version of the same ad. The command includes measured integrated loudness and measured true peak, but omits measured range and measured threshold. Even if linear=true is present, the complete prerequisite set is missing, so the documented rule requires a fallback to dynamic mode. In another fictional variation, all four measurements are present, but the requested range is narrower than the source range or the gain would push true peak beyond the target. That input also fails the linear conditions.

A practical finishing workflow

As an editorial workflow, keep the approved original and analyze a copy of the exact audio intended for rendering. Record the four measurements and use values from that same input in the normalization pass. Keep the intended integrated-loudness, range and true-peak targets in the record. Changing the mix after measurement means the earlier statistics no longer describe the new source.

Request statistics from the actual normalization pass by adding print_format=json to its loudnorm option list, using the suffix :print_format=json after your targets and measured values. The default, print_format=none, prints no statistics; print_format=summary is the documented readable alternative. This is an option example, not an executed FFmpeg run. Keep the printed report and inspect the normalization mode alongside the output measurements. A report from the earlier analysis pass does not establish the mode used by the final pass. If linear processing is essential, correct the measurements or targets that fail the conditions before rendering again.

Once the fallback is resolved, review the complete finished soundtrack to check whole-file loudness, speech balance and listening conditions before approving delivery.

Illustrative loudnorm linear-mode eligibility record

These are illustrative exercise rows. Save a copy and record your own observations; the example is not an executed acceptance test.

Download review worksheet (CSV)