A silencedetect event does not by itself tell you that the whole recording is silent or that audio should be removed. With the default combined-channel mode, the event describes the filter's evaluation of the audio as a combined input. To distinguish a whole-recording silence from a one-channel condition, run the same analysis with mono=1, then compare the channel-specific start, end, and duration metadata. Treat the resulting interval as a candidate for editorial review, not as an automatic cut instruction.
Concrete Judgment and Steps
- Choose and record the analysis criteria.
noiseis the tolerance: it may be expressed as dB or an amplitude ratio.durationis the minimum time that must satisfy that tolerance before the filter reports an event. For example,silencedetect=noise=-50dB:d=2asks for events at or below the selected level lasting at least two seconds. These settings define what is reported; they do not define whether the material is unwanted.
- Run a baseline analysis in the normal combined mode and preserve the reported timestamps. FFmpeg documents the default noise tolerance as
-60dBand the default minimum duration as two seconds, but use explicit values in an audit so the record is unambiguous. The logged time and duration are expressed in seconds.
- Repeat the analysis with
mono=1, for example with an audio filter string such assilencedetect=noise=-50dB:d=2:mono=1. The documentation says that channels are evaluated separately in this mode and that channel-specific metadata uses a.Xsuffix, whereXidentifies the channel. Compare the left and right channel markers at the same timeline positions.
- Classify the event. If both channels report matching or materially overlapping intervals, the recording may contain a whole-program low-level passage under these criteria. If only one channel reports an interval while the other remains active, it is a channel-specific candidate, not proof that the stereo program is silent. Confirm by listening to the affected channel, the other channel, and the ordinary stereo presentation.
- Check the intended content and timing around the interval. The documented start metadata is attached when the detection-duration condition has been reached, while it contains the first frame's silence timestamp. End and duration metadata are set on the first frame after the silence. Therefore, compare the reported boundaries with the audible material and any dialogue, room tone, music, or channel-specific recording role before making an edit.
The official `silencedetect` documentation defines the threshold, duration, mono evaluation, and metadata behavior. The comparison and listening sequence above is an editorial audit method built around those outputs.
A New Acceptance Worksheet
The following is a hypothetical worksheet, not a guarantee of what every file or FFmpeg log will contain. Assume a stereo interview is analyzed with noise=-50dB, duration=2s, and mono=1.
| Candidate interval | Channel observation | Expected content | Decision and reason |
|---|---|---|---|
| 12.4–14.8 s | Left has a silence marker; right has no marker | Speaker remains audible on the right; left microphone may be idle | Further review; do not call the recording silent |
| 12.4–14.8 s | Stereo playback still contains clear speech | Dialogue is expected throughout the exchange | Retain unless a channel repair decision is separately approved |
| 31.0–34.2 s | Both channels show overlapping markers under the same criteria | The script allows a deliberate pause | Retain after content and timing check |
| 31.0–34.2 s | Both channels are marked, but a required word falls inside the interval | Dialogue is expected | Further review; the detector does not override editorial evidence |
| 45.0–47.5 s | Both channels are marked and the passage is an unwanted gap | No speech, music, or timing cue is expected | Consider cutting only after timeline acceptance and approval |
Record the threshold, minimum duration, channel labels, reported start and end, expected material, and decision status. Use actual metadata labels rather than waveform color, track color, or a guessed left/right order; color is only a visual clue.
Limits and the Next Step After Failure
A marker means that the configured level and minimum-duration conditions were met. It does not identify the cause, judge the artistic intent, or prove a defect. A quiet speaker, a disconnected microphone, phase or routing problems, and an intentional pause can lead to different editorial conclusions even when a marker looks similar. The documentation also does not promise that the default combined result can be mapped back to one physical channel; that is why the separate mono run matters.
If the channel comparison is inconsistent with listening, first verify the actual audio stream, channel layout, time base, and the exact filter options used. Then inspect a short region around each boundary and listen separately to each channel. If an edit is approved, validate the resulting dialogue, channel balance, and synchronization at both sides of the cut. FFmpeg's `silenceremove` documentation describes a separate trimming filter with its own thresholds, durations, and multi-channel modes; a silencedetect report alone is not an instruction to invoke it.
Three Short Follow-up Questions
Does a marker mean the samples are digitally zero? No. It means the measured audio met the configured tolerance for at least the configured duration. A low-level room sound may qualify.
Why did the normal run show silence but mono mode show activity on one channel? The combined evaluation and per-channel evaluation answer different questions. Compare the same interval using the channel-specific markers; do not infer channel ownership from the normal log alone.
Should I remove every interval reported on both channels? No. Check whether the pause is intentional, whether required content falls inside it, and whether the edited timeline remains acceptable. Detection supplies evidence for review, not editorial approval.
Sources and next step
- 8.107 silencedetect — FFmpeg official documentation
- 8.108 silenceremove — 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.