How to auto-cut dead air and long pauses from interview recordings
- Step 1Drop the raw interview — Drag your recording onto the silence-stripper tool. Audio or video both work, so a Zoom/Riverside export or a local WAV is fine. Decoding and processing are local — the interview never uploads.
- Step 2Pick a threshold for the noisier party — On a remote interview, set Threshold (dB) for whichever side has the higher noise floor (usually the remote guest). If the guest's room hums,
-35 dBwill register their gaps; a clean studio host's gaps will also be caught. Range is-80…-10, step 1. - Step 3Set the minimum silence (s) — Min silence (s) is the cushion left at each cut. Keep interviews at
0.5–0.8 sso a guest's considered pause still lands as a beat, not a jump cut. Drop to0.3 sonly for very pacey edits. Range0.1…5, step 0.1. - Step 4Run the strip — FFmpeg applies
silenceremovewithdetection=peak: leading silence trimmed once (start_periods=1), every internal and trailing gap tightened (stop_periods=-1). It removes only time — no leveling, no EQ, no de-noise. - Step 5Check the seams around questions — Listen to the moments right after the host's questions, where the longest pauses live. If a guest's answer onset is clipped, lower the threshold magnitude (
-35→-40) or raise Min silence. If hesitation beats remain, lower Min silence. - Step 6Download and finish the interview — The result downloads as
interview-tightened.extin your original format. Two remote rooms usually need cleanup too — follow with ai-noise-reducer to kill the guest's room hum, then loudness-normalizer.
The two real controls
These are the only options the silence stripper exposes, verified against the tool's client component and the FFmpeg filter built in the processor.
| Control | FFmpeg parameter | Default | Range / step | Interview meaning |
|---|---|---|---|---|
| Threshold (dB) | start_threshold + stop_threshold (detection=peak) | -40 dB | -80 to -10, step 1 | Set it for the party with the higher noise floor — usually the remote guest — so their gaps register without chopping the studio host |
| Min silence (s) | start_silence + stop_silence | 0.5 s | 0.1 to 5, step 0.1 | Cushion left at each cut. Keep interviews airier (0.5–0.8 s) so considered pauses keep their weight |
Interview conditions and settings
Starting points for common interview capture setups. Confirm by ear; remote lines vary a lot.
| Setup | Suggested threshold | Suggested min silence | Why |
|---|---|---|---|
| Both parties in studios, separate tracks | -45 dB | 0.6 s | Low floors on both sides; strict threshold, gentle cushion |
| Studio host + kitchen-table guest (single mixed track) | -35 dB | 0.6–0.8 s | Threshold must clear the guest's room tone; the host's clean gaps still cut |
| Phone / VoIP guest with dropouts | -32 to -30 dB | 0.8 s | Variable line; higher cushion avoids choppy edits across dropouts |
| Field interview with ambient noise | -28 to -25 dB | 0.8–1.0 s | High ambient floor; loose threshold and generous cushion keep it natural |
Pro-tier limits for long interviews
The tool requires the Pro tier or higher. Duration and size are separate caps — whichever hits first.
| Tier | Max file size | Max duration | Files per batch |
|---|---|---|---|
| Pro (minimum) | 200 MB | 120 min | 10 |
| Pro-media | 100 GB | Unlimited | 100 |
| Developer | 100 GB | Unlimited | Unlimited |
Cookbook
Interview-specific recipes for the two-control silence stripper. Remember it removes only time and can't tell which speaker is talking — so mixed-noise-floor sessions need a threshold tuned to the noisier party.
Two-studio interview, separate tracks
Both host and guest recorded clean in their own studios. Strip each track at a strict threshold for an even, professional tighten.
Input: guest-track.wav (47 min)
Settings: Threshold -45 dB · Min silence 0.6 s
FFmpeg applies:
silenceremove=start_periods=1:start_silence=0.6:start_threshold=-45dB:
stop_periods=-1:stop_silence=0.6:stop_threshold=-45dB:detection=peak
Output: guest-track-tightened.wav (42 min)
-> ~11% removed; do the same on host-track.wav, then mixMixed-floor single track (studio host + noisy guest)
One mixed file where the host is clean and the guest's room hums. Set the threshold for the guest so their gaps register.
Try -45 dB first -> only the host's gaps cut, guest gaps stay (guest room tone sits above -45 dB) Fix: Threshold -35 dB · Min silence 0.7 s -> both sides' gaps register; tighter, still natural
Flaky VoIP line with dropouts
A phone guest with intermittent dropouts. Use a generous cushion so the edit doesn't sound choppy across the variable line.
Input: phone-interview.mp3 (33 min) Settings: Threshold -31 dB · Min silence 0.8 s Output: phone-interview-tightened.mp3 (29 min) -> dropout silences cut, but the 0.8 s cushion keeps the conversation flowing rather than stuttering
Strip first, then denoise the guest's room
The right order when a remote guest has hiss/hum. Strip silence first so the noise reducer isn't analysing long dead-air stretches.
Step 1 - silence-stripper: raw.wav -> raw-tightened.wav (-35 dB, 0.7 s) Step 2 - ai-noise-reducer (/audio-tools/ai-noise-reducer): raw-tightened.wav -> RNNoise speech model removes hum/hiss Step 3 - loudness-normalizer: -> -16 LUFS for publish
Long-form 90-minute interview on Pro-media
A deep-dive interview that exceeds the Pro 120-minute cap. Move to Pro-media for unlimited duration, or split first.
Input: deepdive.wav (94 min) <- over Pro's 120 min? no, but a 4-hour session would be. For 94 min on Pro it's fine if under 200 MB; a 48 kHz stereo WAV at 94 min is ~970 MB -> exceeds Pro's 200 MB size cap. Options: - Convert to a smaller format first, or - Use Pro-media (100 GB / unlimited duration), or - Split with /audio-tools/audio-splitter, strip, rejoin
Edge cases and what actually happens
One speaker's gaps cut, the other's don't
By designOn a mixed single track the two parties have different noise floors. A threshold strict enough for the clean studio host won't reach the noisy guest's gaps. Set the threshold for the noisier party (raise toward -35/-30) so both register — or strip each speaker's separate track individually.
Guest's answer onset is clipped
ExpectedWhen a guest starts an answer softly, the onset can fall below the threshold and get trimmed. Lower the threshold magnitude (-35 → -40) so soft onsets survive, or raise Min silence so more cushion is kept around each cut.
Edit sounds choppy across a dropout-prone line
ExpectedVoIP dropouts create many short silences; an aggressive Min silence (0.2–0.3 s) stacks lots of tiny cuts and the result stutters. Raise Min silence to 0.8 s+ for variable lines so the conversation flows.
Cross-talk regions are never tightened
By designWhen host and guest overlap there is no true silence to remove, so cuts only happen in single-speaker gaps. That's correct, but pacing can feel uneven on lively interviews. Strip per-track before mixing if you need even tightening across the whole runtime.
Background ambience above the threshold blocks all cuts
By designA field interview with constant traffic or café noise may never drop below the threshold, so nothing is cut. Raise the threshold toward -28/-25 dB, or denoise with ai-noise-reducer first so real gaps fall below the threshold.
File over the Pro 200 MB / 120 min cap
RejectedLong, high-bitrate interview masters can exceed Pro's 200 MB size or 120 min duration (whichever first). Split with audio-splitter and strip the parts, or upgrade to Pro-media (100 GB, unlimited duration).
MP3 export re-encodes the interview
ExpectedOutput matches input format, so an MP3 interview is re-encoded once via libmp3lame to apply the cut. For archival interviews, strip on the lossless WAV and convert to MP3 only when publishing.
Nothing removed on a heavily-compressed line
By designSome VoIP codecs apply aggressive noise gates that fill gaps with comfort noise above the threshold, so no region reads as silent. Try a higher threshold, or denoise first to expose the real gaps.
Frequently asked questions
How do I tighten a remote interview where the host and guest sound very different?
Set the Threshold (dB) for the party with the higher noise floor — usually the remote guest. If their room hums, -35 dB will catch their gaps while the clean studio host's gaps cut too. The cleanest result is to strip each speaker's separate track at its own threshold before mixing.
Does my interview upload anywhere?
No. The tool runs FFmpeg 8.1 in WebAssembly inside your browser tab — the raw interview is processed on your own CPU and never uploaded. That matters for embargoed or NDA interviews. The tool requires the Pro tier (200 MB / 120 min per file).
What threshold and minimum-silence should I use for interviews?
Start at -40 dB / 0.5 s for clean captures. For a noisy remote guest, raise the threshold toward -35/-30 dB and the cushion to 0.6–0.8 s so the edit stays natural across a variable line. Field interviews with ambient noise may need -28/-25 dB.
Will it cut a guest's thoughtful pauses?
The Min silence value sets how much quiet is left at each cut, so at 0.5–0.8 s a considered pause keeps its weight. Long dramatic silences are tightened, not removed entirely. Raise Min silence if pauses lose their impact; lower it for a snappier edit.
Why did nothing get removed from my interview?
The threshold is likely lower than the room's noise floor, so FFmpeg sees no silent regions. Raise the threshold (e.g. -40 → -35 → -30) until the gaps register. Constant background noise (traffic, café) can also block cuts — denoise first with ai-noise-reducer.
Can it tell who is speaking?
No — it only detects level vs. the threshold, not speakers. It tightens single-speaker gaps and leaves cross-talk untouched (there's no silence there to remove). For per-speaker control, strip each isolated track separately, then mix.
What format does the tightened interview come out as?
The same format you uploaded — there's no format selector. A WAV stays WAV at the same sample rate; an MP3 stays MP3. The file is named interview-tightened.ext. To change format afterward, use a converter like wav-to-mp3.
Should I denoise before or after stripping silence?
Strip silence first, then denoise. Removing the long dead-air stretches first means the noise reducer analyses only real audio, and it stops constant ambience from blocking cuts in the first place. Chain into ai-noise-reducer.
Can I process the whole interview series at once?
The Pro tier allows up to 10 files per batch (Pro-media 100, Developer unlimited). They process locally with the same settings. For interviews recorded under different conditions, run them individually so you can tune the threshold per session.
My interview is 90 minutes and the upload was rejected — why?
Probably the size cap, not duration: a long uncompressed stereo WAV easily passes 200 MB even under 120 minutes on Pro. Move to Pro-media (100 GB / unlimited), or split with audio-splitter, strip each part, and rejoin.
Does it level out a quiet guest vs. a loud host?
No — the silence stripper changes only timing, never level. For uneven volumes between host and guest, run speech-leveler after stripping, then loudness-normalizer for the final target.
Is there a one-step way to do the whole interview cleanup?
Yes — podcast-master chains silence handling, noise reduction, leveling, and loudness in a single pass. Use the standalone silence stripper when you want manual control over exactly how aggressively the gaps are cut.
Privacy first
Every JAD Audio tool runs entirely in your browser via FFmpeg (WebAssembly) and RNNoise. Your audio files never leave your device — verified by zero outbound network requests during processing.