How to silence sensitive audio in an interview recording
- Step 1Mark the sensitive passage in seconds — Review the interview and note the start and end of the disclosure in seconds. Decimals help you redact only the sensitive words, not the surrounding answer.
- Step 2Drop the footage onto the tool — Open the audio region muter and drag the clip in. FFmpeg.wasm runs in your browser — the footage is never uploaded. MP4, MOV, MKV, WebM, AVI, M4V, TS accepted; MP4 out.
- Step 3Set Start (s) — Enter the start of the passage into Start (s) — the lower bound of the
between()window. A hair early is safer than a hair late. - Step 4Set End (s) — Enter the end into End (s); it must exceed Start. Extend slightly past the last identifying syllable so nothing leaks at the edge.
- Step 5Run the redaction — The passage is silenced; the rest of the interview and the entire video stream are untouched. Verify with headphones that nothing identifying survives at the boundaries.
- Step 6Redact further passages, then archive — For additional disclosures, feed the MP4 back in and run another pass. Keep the redacted MP4 as the shareable copy and store the original securely and separately.
The only two controls (verified against the tool UI)
The options panel for this tool has exactly two numeric fields. There is no beep/tone option, no fade, no waveform scrubber, and no multi-region list — every claim here is checked against the actual client component.
| Control | What it is | Range | Default |
|---|---|---|---|
| Start (s) | Start of the mute region in seconds. Passed to FFmpeg as the lower bound of between(t, start, end). Decimals allowed (rounded to 3 places, e.g. 12.480) | 0 to clip duration | 0 |
| End (s) | End of the mute region in seconds. Upper bound of the between() expression. Must be greater than Start or the tool stops with *End time must be after start time* | 0 to clip duration | 0 |
| Region count | One region per pass. The tool applies a single between(t, start, end) window. To silence several separate sections, run the tool once per section (the output of one pass is the input to the next) | 1 | 1 |
| Replacement | Silence only — the region's volume is set to 0. There is no beep, tone, hum, or audio substitution. The registry note is explicit: silence is the default to avoid copyright on common beep sounds | fixed | silence |
Audio vs. visual redaction — pick the right tool
This tool redacts the audio only. For a fully de-identified interview you usually combine it with a visual redaction tool.
| What to hide | This tool? | Use instead / also |
|---|---|---|
| Spoken name / address / disclosure | Yes — silence the window | — |
| The person's face | No | face-blur or face-pixelate |
| A document / screen on camera | No | video-redactor (region blur/black box) |
| File metadata (camera, GPS, dates) | No | metadata-scrubber |
| Burned-in caption naming the source | No | Re-export captions via subtitle-burner without the name |
What actually happens to your file
Grounded in the FFmpeg command the tool runs. The video stream is copied bit-for-bit; only the audio is touched. For redaction, the hard silence and the copied (un-re-encoded) video together keep the deliverable both clean and authentic.
| Stage | FFmpeg behaviour | Effect on your file |
|---|---|---|
| Video stream | -c:v copy — the video is stream-copied, never decoded or re-encoded | Zero quality loss on the picture. No generation loss, no re-compression artefacts, frame-identical to the source |
| Audio filter | -af volume=enable='between(t,START,END)':volume=0 — audio plays at full volume everywhere except the region, where volume drops to 0 | Hard silence inside the region; original audio everywhere else |
| Audio re-encode | -c:a aac -b:a 192k — the entire audio track is re-encoded to AAC at 192 kbps, not stream-copied | One transcode pass on the whole track. AAC 192k is transparent for speech/most music, but it is a re-encode, not a copy |
| Container | Output is always written as out_*.mp4 | You always get an .mp4 back, whatever container you fed in (MOV, MKV, WebM, AVI, M4V, TS) |
| Where it runs | FFmpeg.wasm inside your browser tab — no upload | The file never leaves your device. Nothing is sent to a server |
Cookbook
Tight timing for redacting spoken disclosures. Err slightly wide so no identifying syllable survives.
Redact a spoken name
The subject says their full name between 0:42.0 and 0:44.5. Mute the whole utterance with a small safety margin.
Start (s): 41.8 End (s): 44.8 -af volume=enable='between(t,41.800,44.800)':volume=0 Result: the name is silent; the rest of the answer plays.
Redact a longer disclosure
A 12-second passage names an organisation and a location. One window covers it.
Start (s): 312 End (s): 324 The entire disclosure is silenced; surrounding context intact.
Multiple disclosures across an interview
Three separate passages to redact. One pass each; chain the outputs.
Pass 1 -> Start 42 End 45 (download) Pass 2 -> reload, Start 312 End 324 (download) Pass 3 -> reload, Start 905 End 911 (download) All three passages silenced; video copied each pass.
Audio + face redaction together
For full de-identification, silence the audio here, then blur the face in the face tool.
Step 1 (this tool): mute the identifying passages -> out.mp4 Step 2 (face-blur): blur the subject's face on out.mp4 Result: voice and face both protected.
Keep an authenticated original
Because the video is copied, the redacted file's picture is bit-identical to the source — but always retain the unredacted original securely for your records.
Shareable copy: out.mp4 (passages silenced, picture copied) Master copy: original, stored offline / access-controlled
Edge cases and what actually happens
Another voice overlaps the redacted passage
CaveatMuting is time-based, not per-speaker — if the interviewer or background speech overlaps the disclosure window, that audio is silenced too. There is no source separation. If only the subject's words must go but others must stay, the window approach can't isolate by voice.
Identifying syllable survives at the edge
CaveatIf the first or last syllable of a name leaks, widen the window by 0.2–0.5 s. For redaction, erring slightly wide is the safe default — silencing a little surrounding context is preferable to leaking an identifier.
End is less than or equal to Start
RejectedIf End is the same as or earlier than Start, the tool stops immediately with *End time must be after start time. Set both Start and End in the options panel.* Set End to a value strictly greater than Start. A common cause is leaving both at the default 0.
Region runs past the end of the clip
By designThe between(t, start, end) expression simply has no frames to act on past the real end of the audio, so an End value larger than the clip duration silences from Start to the end of the audio and stops. The UI also caps the End field at the detected clip duration, so you normally cannot overshoot.
Audio gets re-encoded even outside the region
ExpectedThe command runs -c:a aac -b:a 192k, which re-encodes the whole audio track, not just the muted window. AAC at 192 kbps is transparent for speech and most music, but if you need a true stream-copy of the untouched audio, this tool is not the right choice — it always does one audio transcode.
Output is MP4 even though you uploaded MOV/MKV/WebM
By designThe tool always writes out_*.mp4. If you need the original container back (for example MKV with multiple tracks), this tool will not preserve it. Use it for the MP4 deliverable, or mute first then remux elsewhere.
Several separate sections to mute
By designEach pass mutes one window. For multiple non-adjacent sections, run the tool repeatedly: mute section 1, download, feed that MP4 back in, mute section 2, and so on. The video is stream-copied each time so the picture never degrades across passes.
You need the face hidden too
Audio onlyThis tool only redacts audio. To obscure the subject's face, run the output through face-blur or face-pixelate. To black out a document on screen, use the video-redactor.
Camera/GPS metadata still identifies the source
Out of scopeSilencing audio doesn't touch container metadata (camera model, GPS, creation dates) that can identify where footage was shot. Run the file through the metadata-scrubber as a separate step.
Very large file on the Free plan
BlockedFree caps a single video at 1 GB. A 1080p hour-long export can exceed that. Pro lifts the cap to 10 GB and Pro-media to 100 GB. Because the video is stream-copied, processing is fast even on large files — the bottleneck is the size cap, not duration.
Frequently asked questions
Is the interview footage uploaded anywhere?
No. FFmpeg.wasm runs entirely in your browser tab, so sensitive footage never leaves your device — essential for source protection.
Can the redacted audio be recovered?
Within the output file the window is set to volume 0 — there is no residual audio in it to recover. The original, unredacted recording still exists wherever you keep it, so store the master securely and separately.
What if another person speaks during the passage I'm redacting?
Muting is time-based, so any audio in the window — including overlapping voices — is silenced. There's no per-speaker isolation; if you must keep one voice and drop another in the same moment, this tool can't separate them.
Does this also hide the person's face?
No — it only redacts audio. Use face-blur or face-pixelate for faces, and video-redactor for documents or screens on camera.
Will the redacted video look re-encoded?
No. The video is stream-copied with -c:v copy, so the picture is frame-identical to the source — no re-encode that could raise authenticity concerns.
How do I redact several disclosures in one interview?
Run one pass per passage: silence the first, download, reload that MP4, silence the next. The copied video doesn't degrade across passes.
Should I redact a little wider than the exact words?
Yes — extend the window by 0.2–0.5 s if a syllable leaks at the edge. For redaction, slightly wide is safer than slightly short.
Does it use a beep over the redacted words?
No — it's silence (volume 0), which is appropriate for evidentiary and editorial redaction. There's no beep or tone option.
What about identifying metadata in the file?
Audio muting doesn't touch metadata. Run the output through the metadata-scrubber to strip camera, GPS, and date fields separately.
What format do I get back?
Always MP4, whatever container you uploaded.
What if I set End before Start?
The tool stops with End time must be after start time. Set End greater than Start.
How large can the interview file be?
Free up to 1 GB; Pro 10 GB; Pro-media and Developer 100 GB. There's no duration cap — only file size and batch count.
Privacy first
Every JAD Video tool runs entirely in your browser via WebCodecs and FFmpeg (WebAssembly). Your video files never leave your device — verified by zero outbound network requests during processing.