Put an analysis filter before signal-changing filters when you need to describe the source; put it after them when you need to describe the processed audio. If you need both answers, make two explicitly labeled analysis paths from the same input. There is no universal filter order: the order determines which signal is observed, and trimming silence adds a separate timestamp question.
Choose the signal, then build the path
First write down what the result must answer: “Where is silence in the original recording?” or “What loudness and silence remain after my approved processing?” Filters such as silenceremove and loudnorm alter or analyze audio in different ways; an analysis result only describes the signal at its position in the chain.
Use these steps:
- Identify the input stream and the exact processing you want to assess. For a source audit, do not put a signal-changing filter before the observer.
- For source measurements, place
silencedetectorebur128before processing. Direct the analysis run to a null output, such as-f null -, so it is a measurement pass rather than a delivery render. FFmpeg documentssilencedetectwith this kind of null-output invocation. - For processed measurements, place the observer after the processing filters. For example, to ask about loudness after normalization, put
ebur128afterloudnorm; to ask where silence remains after trimming, putsilencedetectaftersilenceremove. - When both answers matter, run and label both paths against the same input and settings. Do not compare a source measurement from one file with a processed measurement from a different revision.
- If silence trimming is involved, record its timestamp mode and decide whether later event times should follow rewritten output time or retained input-frame times.
These are filter-graph design choices, not Imagild controls or a claim that analysis generates a deliverable media file.
A two-path acceptance worksheet
The following is a hypothetical review of a 60-second narration. The observations are illustrative, not measured results or FFmpeg guarantees. They show how to make a decision without treating two different signals as if they were the same.
| Path and observation | Decision |
|---|---|
Source path: input → silencedetect and ebur128 → null output; the hypothetical log marks a quiet interval near 12 seconds. | Keep this result as evidence about the incoming recording, even if later processing changes that interval. |
Processed path: same input → approved loudnorm settings → ebur128 → null output; the hypothetical summary differs from the source summary. | Label it as a post-normalization measurement; do not substitute it for the source baseline. |
Trim path: same input → silenceremove → silencedetect → null output; an event is reported near 9 seconds. | Interpret it on the trimmed output timeline if timestamps were rewritten; check the selected mode before matching it to the source. |
For actual acceptance, record the input identity, filter order, settings, measurement output and timestamp mode. Decide against the requirement: source diagnosis uses the source path; delivery review uses the path that matches the final processing chain. A null-output analysis run does not create the delivery media.
Limits and what to do when results disagree
FFmpeg defines silenceremove timestamp mode write as rewriting timestamps while keeping the first output frame’s start time; copy keeps each non-dropped frame’s input timestamp. The default is write. As a result, event times after removed intervals may not align with positions in the original recording. See the official `silenceremove` documentation before joining detection markers to source notes.
silencedetect reports a silence start on a frame whose timestamp reaches or passes the detection duration, with the start timestamp carried in metadata; the end and duration metadata are set on the first frame after silence. Its reported times are in seconds. See FFmpeg’s `silencedetect` reference. A marker is not, by itself, proof that a person perceives silence: thresholds and channel handling matter, so confirm that the configured measurement matches the decision you are making.
If the two paths disagree, first check that they use the same input, stream and relevant settings. Then inspect the filter order and timestamp mode. If a processed-path result is unexpected, measure immediately before and after the specific processing filter to narrow down where the change appears. Do not infer a filter’s undocumented behavior from a changed number alone. loudnorm provides loudness statistics and normalization options, but a measurement after it describes its output path, not the untouched source; see the official `loudnorm` documentation.
Three quick questions
Does `silencedetect` have to come first? No. Put it first to inspect incoming audio, or after processing to inspect the result.
Will `-f null -` save a processed audio file? No. It directs that run to a null output for analysis; use a separate, deliberately configured output run for delivery.
Which `silenceremove` timestamp mode should I choose? Choose based on how later timestamps will be interpreted. write rewrites output timing; copy preserves input timestamps on frames that remain. Neither choice makes source-time and output-time positions interchangeable.
Sources and next step
- 8.107 silencedetect — FFmpeg official documentation
- 8.108 silenceremove — FFmpeg official documentation
- 8.97 loudnorm — FFmpeg official documentation
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.